Cython/Build/Dependencies.py

   1 from glob import glob
   2 import re, os, sys
   3
   4 try:
   5     set
   6 except NameError:
   7     # Python 2.3
   8     from sets import Set as set
   9
  10 from distutils.extension import Extension
  11
  12 from Cython import Utils
  13 from Cython.Compiler.Main import Context, CompilationOptions, default_options
  14
  15 # Unfortunately, Python 2.3 doesn't support decorators.
  16 def cached_method(f):
  17     cache_name = '__%s_cache' % f.__name__
  18     def wrapper(self, *args):
  19         cache = getattr(self, cache_name, None)
  20         if cache is None:
  21             cache = {}
  22             setattr(self, cache_name, cache)
  23         if args in cache:
  24             return cache[args]
  25         res = cache[args] = f(self, *args)
  26         return res
  27     return wrapper
  28
  29
  30 def parse_list(s):
  31     if s[0] == '[' and s[-1] == ']':
  32         s = s[1:-1]
  33         delimiter = ','
  34     else:
  35         delimiter = ' '
  36     s, literals = strip_string_literals(s)
  37     def unquote(literal):
  38         literal = literal.strip()
  39         if literal[0] == "'":
  40             return literals[literal[1:-1]]
  41         else:
  42             return literal
  43
  44     return [unquote(item) for item in s.split(delimiter)]
  45
  46 transitive_str = object()
  47 transitive_list = object()
  48
  49 distutils_settings = {
  50     'name':                 str,
  51     'sources':              list,
  52     'define_macros':        list,
  53     'undef_macros':         list,
  54     'libraries':            transitive_list,
  55     'library_dirs':         transitive_list,
  56     'runtime_library_dirs': transitive_list,
  57     'include_dirs':         transitive_list,
  58     'extra_objects':        list,
  59     'extra_compile_args':   list,
  60     'extra_link_args':      list,
  61     'export_symbols':       list,
  62     'depends':              transitive_list,
  63     'language':             transitive_str,
  64 }
  65
  66 def line_iter(source):
  67     start = 0
  68     while True:
  69         end = source.find('\n', start)
  70         if end == -1:
  71             yield source[start:]
  72             return
  73         yield source[start:end]
  74         start = end+1
  75
  76 class DistutilsInfo(object):
  77
  78     def __init__(self, source=None, exn=None):
  79         self.values = {}
  80         if source is not None:
  81             for line in line_iter(source):
  82                 line = line.strip()
  83                 if line != '' and line[0] != '#':
  84                     break
  85                 line = line[1:].strip()
  86                 if line[:10] == 'distutils:':
  87                     line = line[10:]
  88                     ix = line.index('=')
  89                     key = str(line[:ix].strip())
  90                     value = line[ix+1:].strip()
  91                     type = distutils_settings[key]
  92                     if type in (list, transitive_list):
  93                         value = parse_list(value)
  94                         if key == 'define_macros':
  95                             value = [tuple(macro.split('=')) for macro in value]
  96                     self.values[key] = value
  97         elif exn is not None:
  98             for key in distutils_settings:
  99                 if key in ('name', 'sources'):
 100                     pass
 101                 value = getattr(exn, key, None)
 102                 if value:
 103                     self.values[key] = value
 104
 105     def merge(self, other):
 106         if other is None:
 107             return self
 108         for key, value in other.values.items():
 109             type = distutils_settings[key]
 110             if type is transitive_str and key not in self.values:
 111                 self.values[key] = value
 112             elif type is transitive_list:
 113                 if key in self.values:
 114                     all = self.values[key]
 115                     for v in value:
 116                         if v not in all:
 117                             all.append(v)
 118                 else:
 119                     self.values[key] = value
 120         return self
 121
 122     def subs(self, aliases):
 123         if aliases is None:
 124             return self
 125         resolved = DistutilsInfo()
 126         for key, value in self.values.items():
 127             type = distutils_settings[key]
 128             if type in [list, transitive_list]:
 129                 new_value_list = []
 130                 for v in value:
 131                     if v in aliases:
 132                         v = aliases[v]
 133                     if isinstance(v, list):
 134                         new_value_list += v
 135                     else:
 136                         new_value_list.append(v)
 137                 value = new_value_list
 138             else:
 139                 if value in aliases:
 140                     value = aliases[value]
 141             resolved.values[key] = value
 142         return resolved
 143
 144
 145 def strip_string_literals(code, prefix='__Pyx_L'):
 146     """
 147     Normalizes every string literal to be of the form '__Pyx_Lxxx',
 148     returning the normalized code and a mapping of labels to
 149     string literals.
 150     """
 151     new_code = []
 152     literals = {}
 153     counter = 0
 154     start = q = 0
 155     in_quote = False
 156     raw = False
 157     while True:
 158         hash_mark = code.find('#', q)
 159         single_q = code.find("'", q)
 160         double_q = code.find('"', q)
 161         q = min(single_q, double_q)
 162         if q == -1: q = max(single_q, double_q)
 163
 164         # We're done.
 165         if q == -1 and hash_mark == -1:
 166             new_code.append(code[start:])
 167             break
 168
 169         # Try to close the quote.
 170         elif in_quote:
 171             if code[q-1] == '\\':
 172                 k = 2
 173                 while q >= k and code[q-k] == '\\':
 174                     k += 1
 175                 if k % 2 == 0:
 176                     q += 1
 177                     continue
 178             if code[q:q+len(in_quote)] == in_quote:
 179                 counter += 1
 180                 label = "%s%s" % (prefix, counter)
 181                 literals[label] = code[start+len(in_quote):q]
 182                 new_code.append("%s%s%s" % (in_quote, label, in_quote))
 183                 q += len(in_quote)
 184                 in_quote = False
 185                 start = q
 186             else:
 187                 q += 1
 188
 189         # Process comment.
 190         elif -1 != hash_mark and (hash_mark < q or q == -1):
 191             end = code.find('\n', hash_mark)
 192             if end == -1:
 193                 end = None
 194             new_code.append(code[start:hash_mark+1])
 195             counter += 1
 196             label = "%s%s" % (prefix, counter)
 197             literals[label] = code[hash_mark+1:end]
 198             new_code.append(label)
 199             if end is None:
 200                 break
 201             q = end
 202             start = q
 203
 204         # Open the quote.
 205         else:
 206             raw = False
 207             if len(code) >= q+3 and (code[q+1] == code[q] == code[q+2]):
 208                 in_quote = code[q]*3
 209             else:
 210                 in_quote = code[q]
 211             end = q
 212             while end>0 and code[end-1] in 'rRbBuU':
 213                 if code[end-1] in 'rR':
 214                     raw = True
 215                 end -= 1
 216             new_code.append(code[start:end])
 217             start = q
 218             q += len(in_quote)
 219
 220     return "".join(new_code), literals
 221
 222
 223 def parse_dependencies(source_filename):
 224     # Actual parsing is way to slow, so we use regular expressions.
 225     # The only catch is that we must strip comments and string
 226     # literals ahead of time.
 227     source = Utils.open_source_file(source_filename, "rU").read()
 228     distutils_info = DistutilsInfo(source)
 229     source, literals = strip_string_literals(source)
 230     source = source.replace('\\\n', ' ')
 231     if '\t' in source:
 232         source = source.replace('\t', ' ')
 233     # TODO: pure mode
 234     dependancy = re.compile(r"(cimport +([0-9a-zA-Z_.]+)\b)|(from +([0-9a-zA-Z_.]+) +cimport)|(include +'([^']+)')|(cdef +extern +from +'([^']+)')")
 235     cimports = []
 236     includes = []
 237     externs  = []
 238     for m in dependancy.finditer(source):
 239         groups = m.groups()
 240         if groups[0]:
 241             cimports.append(groups[1])
 242         elif groups[2]:
 243             cimports.append(groups[3])
 244         elif groups[4]:
 245             includes.append(literals[groups[5]])
 246         else:
 247             externs.append(literals[groups[7]])
 248     return cimports, includes, externs, distutils_info
 249
 250
 251 class DependencyTree(object):
 252
 253     def __init__(self, context):
 254         self.context = context
 255         self._transitive_cache = {}
 256
 257     #@cached_method
 258     def parse_dependencies(self, source_filename):
 259         return parse_dependencies(source_filename)
 260     parse_dependencies = cached_method(parse_dependencies)
 261
 262     #@cached_method
 263     def cimports_and_externs(self, filename):
 264         cimports, includes, externs = self.parse_dependencies(filename)[:3]
 265         cimports = set(cimports)
 266         externs = set(externs)
 267         for include in includes:
 268             include_path = os.path.join(os.path.dirname(filename), include)
 269             if not os.path.exists(include_path):
 270                 include_path = self.context.find_include_file(include, None)
 271             if include_path:
 272                 a, b = self.cimports_and_externs(include_path)
 273                 cimports.update(a)
 274                 externs.update(b)
 275             else:
 276                 print("Unable to locate '%s' referenced from '%s'" % (filename, include))
 277         return tuple(cimports), tuple(externs)
 278     cimports_and_externs = cached_method(cimports_and_externs)
 279
 280     def cimports(self, filename):
 281         return self.cimports_and_externs(filename)[0]
 282
 283     #@cached_method
 284     def package(self, filename):
 285         dir = os.path.dirname(filename)
 286         if os.path.exists(os.path.join(dir, '__init__.py')):
 287             return self.package(dir) + (os.path.basename(dir),)
 288         else:
 289             return ()
 290     package = cached_method(package)
 291
 292     #@cached_method
 293     def fully_qualifeid_name(self, filename):
 294         module = os.path.splitext(os.path.basename(filename))[0]
 295         return '.'.join(self.package(filename) + (module,))
 296     fully_qualifeid_name = cached_method(fully_qualifeid_name)
 297
 298     def find_pxd(self, module, filename=None):
 299         if module[0] == '.':
 300             raise NotImplementedError("New relative imports.")
 301         if filename is not None:
 302             relative = '.'.join(self.package(filename) + tuple(module.split('.')))
 303             pxd = self.context.find_pxd_file(relative, None)
 304             if pxd:
 305                 return pxd
 306         return self.context.find_pxd_file(module, None)
 307     find_pxd = cached_method(find_pxd)
 308
 309     #@cached_method
 310     def cimported_files(self, filename):
 311         if filename[-4:] == '.pyx' and os.path.exists(filename[:-4] + '.pxd'):
 312             self_pxd = [filename[:-4] + '.pxd']
 313         else:
 314             self_pxd = []
 315         a = self.cimports(filename)
 316         b = filter(None, [self.find_pxd(m, filename) for m in self.cimports(filename)])
 317         if len(a) != len(b):
 318             print(filename)
 319             print("\n\t".join(a))
 320             print("\n\t".join(b))
 321         return tuple(self_pxd + filter(None, [self.find_pxd(m, filename) for m in self.cimports(filename)]))
 322     cimported_files = cached_method(cimported_files)
 323
 324     def immediate_dependencies(self, filename):
 325         all = list(self.cimported_files(filename))
 326         for extern in sum(self.cimports_and_externs(filename), ()):
 327             all.append(os.path.normpath(os.path.join(os.path.dirname(filename), extern)))
 328         return tuple(all)
 329
 330     #@cached_method
 331     def timestamp(self, filename):
 332         return os.path.getmtime(filename)
 333     timestamp = cached_method(timestamp)
 334
 335     def extract_timestamp(self, filename):
 336         # TODO: .h files from extern blocks
 337         return self.timestamp(filename), filename
 338
 339     def newest_dependency(self, filename):
 340         return self.transitive_merge(filename, self.extract_timestamp, max)
 341
 342     def distutils_info0(self, filename):
 343         return self.parse_dependencies(filename)[3]
 344
 345     def distutils_info(self, filename, aliases=None, base=None):
 346         return (self.transitive_merge(filename, self.distutils_info0, DistutilsInfo.merge)
 347             .subs(aliases)
 348             .merge(base))
 349
 350     def transitive_merge(self, node, extract, merge):
 351         try:
 352             seen = self._transitive_cache[extract, merge]
 353         except KeyError:
 354             seen = self._transitive_cache[extract, merge] = {}
 355         return self.transitive_merge_helper(
 356             node, extract, merge, seen, {}, self.cimported_files)[0]
 357
 358     def transitive_merge_helper(self, node, extract, merge, seen, stack, outgoing):
 359         if node in seen:
 360             return seen[node], None
 361         deps = extract(node)
 362         if node in stack:
 363             return deps, node
 364         try:
 365             stack[node] = len(stack)
 366             loop = None
 367             for next in outgoing(node):
 368                 sub_deps, sub_loop = self.transitive_merge_helper(next, extract, merge, seen, stack, outgoing)
 369                 if sub_loop is not None:
 370                     if loop is not None and stack[loop] < stack[sub_loop]:
 371                         pass
 372                     else:
 373                         loop = sub_loop
 374                 deps = merge(deps, sub_deps)
 375             if loop == node:
 376                 loop = None
 377             if loop is None:
 378                 seen[node] = deps
 379             return deps, loop
 380         finally:
 381             del stack[node]
 382
 383 _dep_tree = None
 384 def create_dependency_tree(ctx=None):
 385     global _dep_tree
 386     if _dep_tree is None:
 387         if ctx is None:
 388             ctx = Context(["."], CompilationOptions(default_options))
 389         _dep_tree = DependencyTree(ctx)
 390     return _dep_tree
 391
 392 # This may be useful for advanced users?
 393 def create_extension_list(patterns, ctx=None, aliases=None):
 394     seen = set()
 395     deps = create_dependency_tree(ctx)
 396     if not isinstance(patterns, list):
 397         patterns = [patterns]
 398     module_list = []
 399     for pattern in patterns:
 400         if isinstance(pattern, str):
 401             filepattern = pattern
 402             template = None
 403             name = '*'
 404             base = None
 405             exn_type = Extension
 406         elif isinstance(pattern, Extension):
 407             filepattern = pattern.sources[0]
 408             if os.path.splitext(filepattern)[1] not in ('.py', '.pyx'):
 409                 # ignore non-cython modules
 410                 module_list.append(pattern)
 411                 continue
 412             template = pattern
 413             name = template.name
 414             base = DistutilsInfo(exn=template)
 415             exn_type = template.__class__
 416         else:
 417             raise TypeError(pattern)
 418         for file in glob(filepattern):
 419             pkg = deps.package(file)
 420             if '*' in name:
 421                 module_name = deps.fully_qualifeid_name(file)
 422             else:
 423                 module_name = name
 424             if module_name not in seen:
 425                 module_list.append(exn_type(
 426                         name=module_name,
 427                         sources=[file],
 428                         **deps.distutils_info(file, aliases, base).values))
 429                 seen.add(name)
 430     return module_list
 431
 432 # This is the user-exposed entry point.
 433 def cythonize(module_list, nthreads=0, aliases=None, **options):
 434     if 'include_path' not in options:
 435         options['include_path'] = ['.']
 436     c_options = CompilationOptions(**options)
 437     cpp_options = CompilationOptions(**options); cpp_options.cplus = True
 438     ctx = c_options.create_context()
 439     module_list = create_extension_list(module_list, ctx=ctx, aliases=aliases)
 440     deps = create_dependency_tree(ctx)
 441     to_compile = []
 442     for m in module_list:
 443         new_sources = []
 444         for source in m.sources:
 445             base, ext = os.path.splitext(source)
 446             if ext in ('.pyx', '.py'):
 447                 if m.language == 'c++':
 448                     c_file = base + '.cpp'
 449                     options = cpp_options
 450                 else:
 451                     c_file = base + '.c'
 452                     options = c_options
 453                 if os.path.exists(c_file):
 454                     c_timestamp = os.path.getmtime(c_file)
 455                 else:
 456                     c_timestamp = -1
 457                 # Priority goes first to modified files, second to direct
 458                 # dependents, and finally to indirect dependents.
 459                 if c_timestamp < deps.timestamp(source):
 460                     dep_timestamp, dep = deps.timestamp(source), source
 461                     priority = 0
 462                 else:
 463                     dep_timestamp, dep = deps.newest_dependency(source)
 464                     priority = 2 - (dep in deps.immediate_dependencies(source))
 465                 if c_timestamp < dep_timestamp:
 466                     print("Compiling %s because it depends on %s" % (source, dep))
 467                     to_compile.append((priority, source, c_file, options))
 468                 new_sources.append(c_file)
 469             else:
 470                 new_sources.append(source)
 471         m.sources = new_sources
 472     to_compile.sort()
 473     if nthreads:
 474         # Requires multiprocessing (or Python >= 2.6)
 475         try:
 476             import multiprocessing
 477             pool = multiprocessing.Pool(nthreads)
 478             pool.map(cythonize_one_helper, to_compile)
 479         except ImportError:
 480             print("multiprocessing required for parallel cythonization")
 481             nthreads = 0
 482     if not nthreads:
 483         for priority, pyx_file, c_file, options in to_compile:
 484             cythonize_one(pyx_file, c_file, options)
 485     return module_list
 486
 487 # TODO: Share context? Issue: pyx processing leaks into pxd module
 488 def cythonize_one(pyx_file, c_file, options=None):
 489     from Cython.Compiler.Main import compile, default_options
 490     from Cython.Compiler.Errors import CompileError, PyrexError
 491
 492     if options is None:
 493         options = CompilationOptions(default_options)
 494     options.output_file = c_file
 495
 496     any_failures = 0
 497     try:
 498         result = compile([pyx_file], options)
 499         if result.num_errors > 0:
 500             any_failures = 1
 501     except (EnvironmentError, PyrexError), e:
 502         sys.stderr.write(str(e) + '\n')
 503         any_failures = 1
 504     if any_failures:
 505         raise CompileError(None, pyx_file)
 506
 507 def cythonize_one_helper(m):
 508     return cythonize_one(*m[1:])