Show More
@@ -1,251 +1,251 b'' | |||
|
1 | 1 | # match.py - file name matching |
|
2 | 2 | # |
|
3 | 3 | # Copyright 2008, 2009 Matt Mackall <mpm@selenic.com> and others |
|
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, incorporated herein by reference. |
|
7 | 7 | |
|
8 | 8 | import util, re |
|
9 | 9 | |
|
10 | 10 | class _match(object): |
|
11 | 11 | def __init__(self, root, cwd, files, mf, ap): |
|
12 | 12 | self._root = root |
|
13 | 13 | self._cwd = cwd |
|
14 | 14 | self._files = files |
|
15 | 15 | self._fmap = set(files) |
|
16 | 16 | self.matchfn = mf |
|
17 | 17 | self._anypats = ap |
|
18 | 18 | def __call__(self, fn): |
|
19 | 19 | return self.matchfn(fn) |
|
20 | 20 | def __iter__(self): |
|
21 | 21 | for f in self._files: |
|
22 | 22 | yield f |
|
23 | 23 | def bad(self, f, msg): |
|
24 | 24 | return True |
|
25 | 25 | def dir(self, f): |
|
26 | 26 | pass |
|
27 | 27 | def missing(self, f): |
|
28 | 28 | pass |
|
29 | 29 | def exact(self, f): |
|
30 | 30 | return f in self._fmap |
|
31 | 31 | def rel(self, f): |
|
32 | 32 | return util.pathto(self._root, self._cwd, f) |
|
33 | 33 | def files(self): |
|
34 | 34 | return self._files |
|
35 | 35 | def anypats(self): |
|
36 | 36 | return self._anypats |
|
37 | 37 | |
|
38 | 38 | class always(_match): |
|
39 | 39 | def __init__(self, root, cwd): |
|
40 | 40 | _match.__init__(self, root, cwd, [], lambda f: True, False) |
|
41 | 41 | |
|
42 | 42 | class never(_match): |
|
43 | 43 | def __init__(self, root, cwd): |
|
44 | 44 | _match.__init__(self, root, cwd, [], lambda f: False, False) |
|
45 | 45 | |
|
46 | 46 | class exact(_match): |
|
47 | 47 | def __init__(self, root, cwd, files): |
|
48 | 48 | _match.__init__(self, root, cwd, files, self.exact, False) |
|
49 | 49 | |
|
50 | 50 | class match(_match): |
|
51 | 51 | def __init__(self, root, cwd, patterns, include=[], exclude=[], |
|
52 | 52 | default='glob'): |
|
53 | 53 | """build an object to match a set of file patterns |
|
54 | 54 | |
|
55 | 55 | arguments: |
|
56 | 56 | root - the canonical root of the tree you're matching against |
|
57 | 57 | cwd - the current working directory, if relevant |
|
58 | 58 | patterns - patterns to find |
|
59 | 59 | include - patterns to include |
|
60 | 60 | exclude - patterns to exclude |
|
61 | 61 | default - if a pattern in names has no explicit type, assume this one |
|
62 | 62 | |
|
63 | 63 | a pattern is one of: |
|
64 | 64 | 'glob:<glob>' - a glob relative to cwd |
|
65 | 65 | 're:<regexp>' - a regular expression |
|
66 | 66 | 'path:<path>' - a path relative to canonroot |
|
67 | 67 | 'relglob:<glob>' - an unrooted glob (*.c matches C files in all dirs) |
|
68 | 68 | 'relpath:<path>' - a path relative to cwd |
|
69 | 69 | 'relre:<regexp>' - a regexp that doesn't have to match the start of a name |
|
70 | 70 | '<something>' - one of the cases above, selected by the dflt_pat argument |
|
71 | 71 | """ |
|
72 | 72 | |
|
73 | 73 | roots = [] |
|
74 | 74 | anypats = bool(include or exclude) |
|
75 | 75 | |
|
76 | 76 | if patterns: |
|
77 | 77 | pats = _normalize(patterns, default, root, cwd) |
|
78 | 78 | roots = _roots(pats) |
|
79 | 79 | anypats = anypats or _anypats(pats) |
|
80 | 80 | pm = _buildmatch(pats, '$') |
|
81 | 81 | if include: |
|
82 | 82 | im = _buildmatch(_normalize(include, 'glob', root, cwd), '(?:/|$)') |
|
83 | 83 | if exclude: |
|
84 | 84 | em = _buildmatch(_normalize(exclude, 'glob', root, cwd), '(?:/|$)') |
|
85 | 85 | |
|
86 | 86 | if patterns: |
|
87 | 87 | if include: |
|
88 | 88 | if exclude: |
|
89 | 89 | m = lambda f: im(f) and not em(f) and pm(f) |
|
90 | 90 | else: |
|
91 | 91 | m = lambda f: im(f) and pm(f) |
|
92 | 92 | else: |
|
93 | 93 | if exclude: |
|
94 | 94 | m = lambda f: not em(f) and pm(f) |
|
95 | 95 | else: |
|
96 | 96 | m = pm |
|
97 | 97 | else: |
|
98 | 98 | if include: |
|
99 | 99 | if exclude: |
|
100 | 100 | m = lambda f: im(f) and not em(f) |
|
101 | 101 | else: |
|
102 | 102 | m = im |
|
103 | 103 | else: |
|
104 | 104 | if exclude: |
|
105 | 105 | m = lambda f: not em(f) |
|
106 | 106 | else: |
|
107 | 107 | m = lambda f: True |
|
108 | 108 | |
|
109 | 109 | _match.__init__(self, root, cwd, roots, m, anypats) |
|
110 | 110 | |
|
111 | 111 | def patkind(pat): |
|
112 | 112 | return _patsplit(pat, None)[0] |
|
113 | 113 | |
|
114 | 114 | def _patsplit(pat, default): |
|
115 | 115 | """Split a string into an optional pattern kind prefix and the |
|
116 | 116 | actual pattern.""" |
|
117 | 117 | if ':' in pat: |
|
118 | 118 | pat, val = pat.split(':', 1) |
|
119 | 119 | if pat in ('re', 'glob', 'path', 'relglob', 'relpath', 'relre'): |
|
120 | 120 | return pat, val |
|
121 | 121 | return default, pat |
|
122 | 122 | |
|
123 |
def _globre(pat |
|
|
123 | def _globre(pat): | |
|
124 | 124 | "convert a glob pattern into a regexp" |
|
125 | 125 | i, n = 0, len(pat) |
|
126 | 126 | res = '' |
|
127 | 127 | group = 0 |
|
128 | 128 | def peek(): return i < n and pat[i] |
|
129 | 129 | while i < n: |
|
130 | 130 | c = pat[i] |
|
131 | 131 | i = i+1 |
|
132 | 132 | if c == '*': |
|
133 | 133 | if peek() == '*': |
|
134 | 134 | i += 1 |
|
135 | 135 | res += '.*' |
|
136 | 136 | else: |
|
137 | 137 | res += '[^/]*' |
|
138 | 138 | elif c == '?': |
|
139 | 139 | res += '.' |
|
140 | 140 | elif c == '[': |
|
141 | 141 | j = i |
|
142 | 142 | if j < n and pat[j] in '!]': |
|
143 | 143 | j += 1 |
|
144 | 144 | while j < n and pat[j] != ']': |
|
145 | 145 | j += 1 |
|
146 | 146 | if j >= n: |
|
147 | 147 | res += '\\[' |
|
148 | 148 | else: |
|
149 | 149 | stuff = pat[i:j].replace('\\','\\\\') |
|
150 | 150 | i = j + 1 |
|
151 | 151 | if stuff[0] == '!': |
|
152 | 152 | stuff = '^' + stuff[1:] |
|
153 | 153 | elif stuff[0] == '^': |
|
154 | 154 | stuff = '\\' + stuff |
|
155 | 155 | res = '%s[%s]' % (res, stuff) |
|
156 | 156 | elif c == '{': |
|
157 | 157 | group += 1 |
|
158 | 158 | res += '(?:' |
|
159 | 159 | elif c == '}' and group: |
|
160 | 160 | res += ')' |
|
161 | 161 | group -= 1 |
|
162 | 162 | elif c == ',' and group: |
|
163 | 163 | res += '|' |
|
164 | 164 | elif c == '\\': |
|
165 | 165 | p = peek() |
|
166 | 166 | if p: |
|
167 | 167 | i += 1 |
|
168 | 168 | res += re.escape(p) |
|
169 | 169 | else: |
|
170 | 170 | res += re.escape(c) |
|
171 | 171 | else: |
|
172 | 172 | res += re.escape(c) |
|
173 |
return |
|
|
173 | return res | |
|
174 | 174 | |
|
175 | 175 | def _regex(kind, name, tail): |
|
176 | 176 | '''convert a pattern into a regular expression''' |
|
177 | 177 | if not name: |
|
178 | 178 | return '' |
|
179 | 179 | if kind == 're': |
|
180 | 180 | return name |
|
181 | 181 | elif kind == 'path': |
|
182 | 182 | return '^' + re.escape(name) + '(?:/|$)' |
|
183 | 183 | elif kind == 'relglob': |
|
184 |
return _globre(name |
|
|
184 | return '(?:|.*/)' + _globre(name) + tail | |
|
185 | 185 | elif kind == 'relpath': |
|
186 | 186 | return re.escape(name) + '(?:/|$)' |
|
187 | 187 | elif kind == 'relre': |
|
188 | 188 | if name.startswith('^'): |
|
189 | 189 | return name |
|
190 | 190 | return '.*' + name |
|
191 |
return _globre(name |
|
|
191 | return _globre(name) + tail | |
|
192 | 192 | |
|
193 | 193 | def _buildmatch(pats, tail): |
|
194 | 194 | """build a matching function from a set of patterns""" |
|
195 | 195 | try: |
|
196 | 196 | pat = '(?:%s)' % '|'.join([_regex(k, p, tail) for (k, p) in pats]) |
|
197 | 197 | if len(pat) > 20000: |
|
198 | 198 | raise OverflowError() |
|
199 | 199 | return re.compile(pat).match |
|
200 | 200 | except OverflowError: |
|
201 | 201 | # We're using a Python with a tiny regex engine and we |
|
202 | 202 | # made it explode, so we'll divide the pattern list in two |
|
203 | 203 | # until it works |
|
204 | 204 | l = len(pats) |
|
205 | 205 | if l < 2: |
|
206 | 206 | raise |
|
207 | 207 | a, b = _buildmatch(pats[:l//2], tail), _buildmatch(pats[l//2:], tail) |
|
208 | 208 | return lambda s: a(s) or b(s) |
|
209 | 209 | except re.error: |
|
210 | 210 | for k, p in pats: |
|
211 | 211 | try: |
|
212 | 212 | re.compile('(?:%s)' % _regex(k, p, tail)) |
|
213 | 213 | except re.error: |
|
214 | 214 | raise util.Abort("invalid pattern (%s): %s" % (k, p)) |
|
215 | 215 | raise util.Abort("invalid pattern") |
|
216 | 216 | |
|
217 | 217 | def _globprefix(pat): |
|
218 | 218 | '''return the non-glob prefix of a path, e.g. foo/* -> foo''' |
|
219 | 219 | root = [] |
|
220 | 220 | for p in pat.split('/'): |
|
221 | 221 | if '[' in p or '{' in p or '*' in p or '?' in p: |
|
222 | 222 | break |
|
223 | 223 | root.append(p) |
|
224 | 224 | return '/'.join(root) or '.' |
|
225 | 225 | |
|
226 | 226 | def _normalize(names, default, root, cwd): |
|
227 | 227 | pats = [] |
|
228 | 228 | for kind, name in [_patsplit(p, default) for p in names]: |
|
229 | 229 | if kind in ('glob', 'relpath'): |
|
230 | 230 | name = util.canonpath(root, cwd, name) |
|
231 | 231 | elif kind in ('relglob', 'path'): |
|
232 | 232 | name = util.normpath(name) |
|
233 | 233 | |
|
234 | 234 | pats.append((kind, name)) |
|
235 | 235 | return pats |
|
236 | 236 | |
|
237 | 237 | def _roots(patterns): |
|
238 | 238 | r = [] |
|
239 | 239 | for kind, name in patterns: |
|
240 | 240 | if kind == 'glob': |
|
241 | 241 | r.append(_globprefix(name)) |
|
242 | 242 | elif kind in ('relpath', 'path'): |
|
243 | 243 | r.append(name or '.') |
|
244 | 244 | elif kind == 'relglob': |
|
245 | 245 | r.append('.') |
|
246 | 246 | return r |
|
247 | 247 | |
|
248 | 248 | def _anypats(patterns): |
|
249 | 249 | for kind, name in patterns: |
|
250 | 250 | if kind in ('glob', 're', 'relglob', 'relre'): |
|
251 | 251 | return True |
General Comments 0
You need to be logged in to leave comments.
Login now