Show More
@@ -1,329 +1,298 b'' | |||
|
1 | 1 | # __init__.py - Startup and module loading logic for Mercurial. |
|
2 | 2 | # |
|
3 | 3 | # Copyright 2015 Gregory Szorc <gregory.szorc@gmail.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 sys |
|
11 | 11 | |
|
12 | 12 | # Allow 'from mercurial import demandimport' to keep working. |
|
13 | 13 | import hgdemandimport |
|
14 | 14 | |
|
15 | 15 | demandimport = hgdemandimport |
|
16 | 16 | |
|
17 | 17 | __all__ = [] |
|
18 | 18 | |
|
19 | 19 | # Python 3 uses a custom module loader that transforms source code between |
|
20 | 20 | # source file reading and compilation. This is done by registering a custom |
|
21 | 21 | # finder that changes the spec for Mercurial modules to use a custom loader. |
|
22 | 22 | if sys.version_info[0] >= 3: |
|
23 | 23 | import importlib |
|
24 | 24 | import importlib.abc |
|
25 | 25 | import io |
|
26 | 26 | import token |
|
27 | 27 | import tokenize |
|
28 | 28 | |
|
29 | 29 | class hgpathentryfinder(importlib.abc.MetaPathFinder): |
|
30 | 30 | """A sys.meta_path finder that uses a custom module loader.""" |
|
31 | 31 | |
|
32 | 32 | def find_spec(self, fullname, path, target=None): |
|
33 | 33 | # Only handle Mercurial-related modules. |
|
34 | 34 | if not fullname.startswith(('mercurial.', 'hgext.')): |
|
35 | 35 | return None |
|
36 | 36 | # don't try to parse binary |
|
37 | 37 | if fullname.startswith('mercurial.cext.'): |
|
38 | 38 | return None |
|
39 | 39 | # third-party packages are expected to be dual-version clean |
|
40 | 40 | if fullname.startswith('mercurial.thirdparty'): |
|
41 | 41 | return None |
|
42 | 42 | # zstd is already dual-version clean, don't try and mangle it |
|
43 | 43 | if fullname.startswith('mercurial.zstd'): |
|
44 | 44 | return None |
|
45 | 45 | # rustext is built for the right python version, |
|
46 | 46 | # don't try and mangle it |
|
47 | 47 | if fullname.startswith('mercurial.rustext'): |
|
48 | 48 | return None |
|
49 | 49 | # pywatchman is already dual-version clean, don't try and mangle it |
|
50 | 50 | if fullname.startswith('hgext.fsmonitor.pywatchman'): |
|
51 | 51 | return None |
|
52 | 52 | |
|
53 | 53 | # Try to find the module using other registered finders. |
|
54 | 54 | spec = None |
|
55 | 55 | for finder in sys.meta_path: |
|
56 | 56 | if finder == self: |
|
57 | 57 | continue |
|
58 | 58 | |
|
59 | 59 | # Originally the API was a `find_module` method, but it was |
|
60 | 60 | # renamed to `find_spec` in python 3.4, with a new `target` |
|
61 | 61 | # argument. |
|
62 | 62 | find_spec_method = getattr(finder, 'find_spec', None) |
|
63 | 63 | if find_spec_method: |
|
64 | 64 | spec = find_spec_method(fullname, path, target=target) |
|
65 | 65 | else: |
|
66 | 66 | spec = finder.find_module(fullname) |
|
67 | 67 | if spec is not None: |
|
68 | 68 | spec = importlib.util.spec_from_loader(fullname, spec) |
|
69 | 69 | if spec: |
|
70 | 70 | break |
|
71 | 71 | |
|
72 | 72 | # This is a Mercurial-related module but we couldn't find it |
|
73 | 73 | # using the previously-registered finders. This likely means |
|
74 | 74 | # the module doesn't exist. |
|
75 | 75 | if not spec: |
|
76 | 76 | return None |
|
77 | 77 | |
|
78 | 78 | # TODO need to support loaders from alternate specs, like zip |
|
79 | 79 | # loaders. |
|
80 | 80 | loader = hgloader(spec.name, spec.origin) |
|
81 | 81 | # Can't use util.safehasattr here because that would require |
|
82 | 82 | # importing util, and we're in import code. |
|
83 | 83 | if hasattr(spec.loader, 'loader'): # hasattr-py3-only |
|
84 | 84 | # This is a nested loader (maybe a lazy loader?) |
|
85 | 85 | spec.loader.loader = loader |
|
86 | 86 | else: |
|
87 | 87 | spec.loader = loader |
|
88 | 88 | return spec |
|
89 | 89 | |
|
90 | 90 | def replacetokens(tokens, fullname): |
|
91 | 91 | """Transform a stream of tokens from raw to Python 3. |
|
92 | 92 | |
|
93 | 93 | It is called by the custom module loading machinery to rewrite |
|
94 | 94 | source/tokens between source decoding and compilation. |
|
95 | 95 | |
|
96 | 96 | Returns a generator of possibly rewritten tokens. |
|
97 | 97 | |
|
98 | 98 | The input token list may be mutated as part of processing. However, |
|
99 | 99 | its changes do not necessarily match the output token stream. |
|
100 | 100 | |
|
101 | 101 | REMEMBER TO CHANGE ``BYTECODEHEADER`` WHEN CHANGING THIS FUNCTION |
|
102 | 102 | OR CACHED FILES WON'T GET INVALIDATED PROPERLY. |
|
103 | 103 | """ |
|
104 | 104 | futureimpline = False |
|
105 | 105 | |
|
106 | 106 | # The following utility functions access the tokens list and i index of |
|
107 | 107 | # the for i, t enumerate(tokens) loop below |
|
108 | 108 | def _isop(j, *o): |
|
109 | 109 | """Assert that tokens[j] is an OP with one of the given values""" |
|
110 | 110 | try: |
|
111 | 111 | return tokens[j].type == token.OP and tokens[j].string in o |
|
112 | 112 | except IndexError: |
|
113 | 113 | return False |
|
114 | 114 | |
|
115 | 115 | def _findargnofcall(n): |
|
116 | 116 | """Find arg n of a call expression (start at 0) |
|
117 | 117 | |
|
118 | 118 | Returns index of the first token of that argument, or None if |
|
119 | 119 | there is not that many arguments. |
|
120 | 120 | |
|
121 | 121 | Assumes that token[i + 1] is '('. |
|
122 | 122 | |
|
123 | 123 | """ |
|
124 | 124 | nested = 0 |
|
125 | 125 | for j in range(i + 2, len(tokens)): |
|
126 | 126 | if _isop(j, ')', ']', '}'): |
|
127 | 127 | # end of call, tuple, subscription or dict / set |
|
128 | 128 | nested -= 1 |
|
129 | 129 | if nested < 0: |
|
130 | 130 | return None |
|
131 | 131 | elif n == 0: |
|
132 | 132 | # this is the starting position of arg |
|
133 | 133 | return j |
|
134 | 134 | elif _isop(j, '(', '[', '{'): |
|
135 | 135 | nested += 1 |
|
136 | 136 | elif _isop(j, ',') and nested == 0: |
|
137 | 137 | n -= 1 |
|
138 | 138 | |
|
139 | 139 | return None |
|
140 | 140 | |
|
141 | 141 | def _ensureunicode(j): |
|
142 | 142 | """Make sure the token at j is a unicode string |
|
143 | 143 | |
|
144 | 144 | This rewrites a string token to include the unicode literal prefix |
|
145 | 145 | so the string transformer won't add the byte prefix. |
|
146 | 146 | |
|
147 | 147 | Ignores tokens that are not strings. Assumes bounds checking has |
|
148 | 148 | already been done. |
|
149 | 149 | |
|
150 | 150 | """ |
|
151 | 151 | st = tokens[j] |
|
152 | 152 | if st.type == token.STRING and st.string.startswith(("'", '"')): |
|
153 | 153 | tokens[j] = st._replace(string='u%s' % st.string) |
|
154 | 154 | |
|
155 | 155 | for i, t in enumerate(tokens): |
|
156 | # Convert most string literals to byte literals. String literals | |
|
157 | # in Python 2 are bytes. String literals in Python 3 are unicode. | |
|
158 | # Most strings in Mercurial are bytes and unicode strings are rare. | |
|
159 | # Rather than rewrite all string literals to use ``b''`` to indicate | |
|
160 | # byte strings, we apply this token transformer to insert the ``b`` | |
|
161 | # prefix nearly everywhere. | |
|
162 | if t.type == token.STRING: | |
|
163 | s = t.string | |
|
164 | ||
|
165 | # Preserve docstrings as string literals. This is inconsistent | |
|
166 | # with regular unprefixed strings. However, the | |
|
167 | # "from __future__" parsing (which allows a module docstring to | |
|
168 | # exist before it) doesn't properly handle the docstring if it | |
|
169 | # is b''' prefixed, leading to a SyntaxError. We leave all | |
|
170 | # docstrings as unprefixed to avoid this. This means Mercurial | |
|
171 | # components touching docstrings need to handle unicode, | |
|
172 | # unfortunately. | |
|
173 | if s[0:3] in ("'''", '"""'): | |
|
174 | yield t | |
|
175 | continue | |
|
176 | ||
|
177 | # If the first character isn't a quote, it is likely a string | |
|
178 | # prefixing character (such as 'b', 'u', or 'r'. Ignore. | |
|
179 | if s[0] not in ("'", '"'): | |
|
180 | yield t | |
|
181 | continue | |
|
182 | ||
|
183 | # String literal. Prefix to make a b'' string. | |
|
184 | yield t._replace(string='b%s' % t.string) | |
|
185 | continue | |
|
186 | ||
|
187 | 156 | # Insert compatibility imports at "from __future__ import" line. |
|
188 | 157 | # No '\n' should be added to preserve line numbers. |
|
189 | 158 | if ( |
|
190 | 159 | t.type == token.NAME |
|
191 | 160 | and t.string == 'import' |
|
192 | 161 | and all(u.type == token.NAME for u in tokens[i - 2 : i]) |
|
193 | 162 | and [u.string for u in tokens[i - 2 : i]] |
|
194 | 163 | == ['from', '__future__'] |
|
195 | 164 | ): |
|
196 | 165 | futureimpline = True |
|
197 | 166 | if t.type == token.NEWLINE and futureimpline: |
|
198 | 167 | futureimpline = False |
|
199 | 168 | if fullname == 'mercurial.pycompat': |
|
200 | 169 | yield t |
|
201 | 170 | continue |
|
202 | 171 | r, c = t.start |
|
203 | 172 | l = ( |
|
204 | 173 | b'; from mercurial.pycompat import ' |
|
205 | 174 | b'delattr, getattr, hasattr, setattr, ' |
|
206 | 175 | b'open, unicode\n' |
|
207 | 176 | ) |
|
208 | 177 | for u in tokenize.tokenize(io.BytesIO(l).readline): |
|
209 | 178 | if u.type in (tokenize.ENCODING, token.ENDMARKER): |
|
210 | 179 | continue |
|
211 | 180 | yield u._replace( |
|
212 | 181 | start=(r, c + u.start[1]), end=(r, c + u.end[1]) |
|
213 | 182 | ) |
|
214 | 183 | continue |
|
215 | 184 | |
|
216 | 185 | # This looks like a function call. |
|
217 | 186 | if t.type == token.NAME and _isop(i + 1, '('): |
|
218 | 187 | fn = t.string |
|
219 | 188 | |
|
220 | 189 | # *attr() builtins don't accept byte strings to 2nd argument. |
|
221 | 190 | if fn in ( |
|
222 | 191 | 'getattr', |
|
223 | 192 | 'setattr', |
|
224 | 193 | 'hasattr', |
|
225 | 194 | 'safehasattr', |
|
226 | 195 | ) and not _isop(i - 1, '.'): |
|
227 | 196 | arg1idx = _findargnofcall(1) |
|
228 | 197 | if arg1idx is not None: |
|
229 | 198 | _ensureunicode(arg1idx) |
|
230 | 199 | |
|
231 | 200 | # .encode() and .decode() on str/bytes/unicode don't accept |
|
232 | 201 | # byte strings on Python 3. |
|
233 | 202 | elif fn in ('encode', 'decode') and _isop(i - 1, '.'): |
|
234 | 203 | for argn in range(2): |
|
235 | 204 | argidx = _findargnofcall(argn) |
|
236 | 205 | if argidx is not None: |
|
237 | 206 | _ensureunicode(argidx) |
|
238 | 207 | |
|
239 | 208 | # It changes iteritems/values to items/values as they are not |
|
240 | 209 | # present in Python 3 world. |
|
241 | 210 | elif fn in ('iteritems', 'itervalues') and not ( |
|
242 | 211 | tokens[i - 1].type == token.NAME |
|
243 | 212 | and tokens[i - 1].string == 'def' |
|
244 | 213 | ): |
|
245 | 214 | yield t._replace(string=fn[4:]) |
|
246 | 215 | continue |
|
247 | 216 | |
|
248 | 217 | # Emit unmodified token. |
|
249 | 218 | yield t |
|
250 | 219 | |
|
251 | 220 | # Header to add to bytecode files. This MUST be changed when |
|
252 | 221 | # ``replacetoken`` or any mechanism that changes semantics of module |
|
253 | 222 | # loading is changed. Otherwise cached bytecode may get loaded without |
|
254 | 223 | # the new transformation mechanisms applied. |
|
255 | 224 | BYTECODEHEADER = b'HG\x00\x0c' |
|
256 | 225 | |
|
257 | 226 | class hgloader(importlib.machinery.SourceFileLoader): |
|
258 | 227 | """Custom module loader that transforms source code. |
|
259 | 228 | |
|
260 | 229 | When the source code is converted to a code object, we transform |
|
261 | 230 | certain patterns to be Python 3 compatible. This allows us to write code |
|
262 | 231 | that is natively Python 2 and compatible with Python 3 without |
|
263 | 232 | making the code excessively ugly. |
|
264 | 233 | |
|
265 | 234 | We do this by transforming the token stream between parse and compile. |
|
266 | 235 | |
|
267 | 236 | Implementing transformations invalidates caching assumptions made |
|
268 | 237 | by the built-in importer. The built-in importer stores a header on |
|
269 | 238 | saved bytecode files indicating the Python/bytecode version. If the |
|
270 | 239 | version changes, the cached bytecode is ignored. The Mercurial |
|
271 | 240 | transformations could change at any time. This means we need to check |
|
272 | 241 | that cached bytecode was generated with the current transformation |
|
273 | 242 | code or there could be a mismatch between cached bytecode and what |
|
274 | 243 | would be generated from this class. |
|
275 | 244 | |
|
276 | 245 | We supplement the bytecode caching layer by wrapping ``get_data`` |
|
277 | 246 | and ``set_data``. These functions are called when the |
|
278 | 247 | ``SourceFileLoader`` retrieves and saves bytecode cache files, |
|
279 | 248 | respectively. We simply add an additional header on the file. As |
|
280 | 249 | long as the version in this file is changed when semantics change, |
|
281 | 250 | cached bytecode should be invalidated when transformations change. |
|
282 | 251 | |
|
283 | 252 | The added header has the form ``HG<VERSION>``. That is a literal |
|
284 | 253 | ``HG`` with 2 binary bytes indicating the transformation version. |
|
285 | 254 | """ |
|
286 | 255 | |
|
287 | 256 | def get_data(self, path): |
|
288 | 257 | data = super(hgloader, self).get_data(path) |
|
289 | 258 | |
|
290 | 259 | if not path.endswith(tuple(importlib.machinery.BYTECODE_SUFFIXES)): |
|
291 | 260 | return data |
|
292 | 261 | |
|
293 | 262 | # There should be a header indicating the Mercurial transformation |
|
294 | 263 | # version. If it doesn't exist or doesn't match the current version, |
|
295 | 264 | # we raise an OSError because that is what |
|
296 | 265 | # ``SourceFileLoader.get_code()`` expects when loading bytecode |
|
297 | 266 | # paths to indicate the cached file is "bad." |
|
298 | 267 | if data[0:2] != b'HG': |
|
299 | 268 | raise OSError('no hg header') |
|
300 | 269 | if data[0:4] != BYTECODEHEADER: |
|
301 | 270 | raise OSError('hg header version mismatch') |
|
302 | 271 | |
|
303 | 272 | return data[4:] |
|
304 | 273 | |
|
305 | 274 | def set_data(self, path, data, *args, **kwargs): |
|
306 | 275 | if path.endswith(tuple(importlib.machinery.BYTECODE_SUFFIXES)): |
|
307 | 276 | data = BYTECODEHEADER + data |
|
308 | 277 | |
|
309 | 278 | return super(hgloader, self).set_data(path, data, *args, **kwargs) |
|
310 | 279 | |
|
311 | 280 | def source_to_code(self, data, path): |
|
312 | 281 | """Perform token transformation before compilation.""" |
|
313 | 282 | buf = io.BytesIO(data) |
|
314 | 283 | tokens = tokenize.tokenize(buf.readline) |
|
315 | 284 | data = tokenize.untokenize(replacetokens(list(tokens), self.name)) |
|
316 | 285 | # Python's built-in importer strips frames from exceptions raised |
|
317 | 286 | # for this code. Unfortunately, that mechanism isn't extensible |
|
318 | 287 | # and our frame will be blamed for the import failure. There |
|
319 | 288 | # are extremely hacky ways to do frame stripping. We haven't |
|
320 | 289 | # implemented them because they are very ugly. |
|
321 | 290 | return super(hgloader, self).source_to_code(data, path) |
|
322 | 291 | |
|
323 | 292 | # We automagically register our custom importer as a side-effect of |
|
324 | 293 | # loading. This is necessary to ensure that any entry points are able |
|
325 | 294 | # to import mercurial.* modules without having to perform this |
|
326 | 295 | # registration themselves. |
|
327 | 296 | if not any(isinstance(x, hgpathentryfinder) for x in sys.meta_path): |
|
328 | 297 | # meta_path is used before any implicit finders and before sys.path. |
|
329 | 298 | sys.meta_path.insert(0, hgpathentryfinder()) |
General Comments 0
You need to be logged in to leave comments.
Login now