• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1"""
2Import utilities
3
4Exported classes:
5    ImportManager   Manage the import process
6
7    Importer        Base class for replacing standard import functions
8    BuiltinImporter Emulate the import mechanism for builtin and frozen modules
9
10    DynLoadSuffixImporter
11"""
12from warnings import warnpy3k
13warnpy3k("the imputil module has been removed in Python 3.0", stacklevel=2)
14del warnpy3k
15
16# note: avoid importing non-builtin modules
17import imp                      ### not available in Jython?
18import sys
19import __builtin__
20
21# for the DirectoryImporter
22import struct
23import marshal
24
25__all__ = ["ImportManager","Importer","BuiltinImporter"]
26
27_StringType = type('')
28_ModuleType = type(sys)         ### doesn't work in Jython...
29
30class ImportManager:
31    "Manage the import process."
32
33    def install(self, namespace=vars(__builtin__)):
34        "Install this ImportManager into the specified namespace."
35
36        if isinstance(namespace, _ModuleType):
37            namespace = vars(namespace)
38
39        # Note: we have no notion of "chaining"
40
41        # Record the previous import hook, then install our own.
42        self.previous_importer = namespace['__import__']
43        self.namespace = namespace
44        namespace['__import__'] = self._import_hook
45
46        ### fix this
47        #namespace['reload'] = self._reload_hook
48
49    def uninstall(self):
50        "Restore the previous import mechanism."
51        self.namespace['__import__'] = self.previous_importer
52
53    def add_suffix(self, suffix, importFunc):
54        assert hasattr(importFunc, '__call__')
55        self.fs_imp.add_suffix(suffix, importFunc)
56
57    ######################################################################
58    #
59    # PRIVATE METHODS
60    #
61
62    clsFilesystemImporter = None
63
64    def __init__(self, fs_imp=None):
65        # we're definitely going to be importing something in the future,
66        # so let's just load the OS-related facilities.
67        if not _os_stat:
68            _os_bootstrap()
69
70        # This is the Importer that we use for grabbing stuff from the
71        # filesystem. It defines one more method (import_from_dir) for our use.
72        if fs_imp is None:
73            cls = self.clsFilesystemImporter or _FilesystemImporter
74            fs_imp = cls()
75        self.fs_imp = fs_imp
76
77        # Initialize the set of suffixes that we recognize and import.
78        # The default will import dynamic-load modules first, followed by
79        # .py files (or a .py file's cached bytecode)
80        for desc in imp.get_suffixes():
81            if desc[2] == imp.C_EXTENSION:
82                self.add_suffix(desc[0],
83                                DynLoadSuffixImporter(desc).import_file)
84        self.add_suffix('.py', py_suffix_importer)
85
86    def _import_hook(self, fqname, globals=None, locals=None, fromlist=None):
87        """Python calls this hook to locate and import a module."""
88
89        parts = fqname.split('.')
90
91        # determine the context of this import
92        parent = self._determine_import_context(globals)
93
94        # if there is a parent, then its importer should manage this import
95        if parent:
96            module = parent.__importer__._do_import(parent, parts, fromlist)
97            if module:
98                return module
99
100        # has the top module already been imported?
101        try:
102            top_module = sys.modules[parts[0]]
103        except KeyError:
104
105            # look for the topmost module
106            top_module = self._import_top_module(parts[0])
107            if not top_module:
108                # the topmost module wasn't found at all.
109                raise ImportError, 'No module named ' + fqname
110
111        # fast-path simple imports
112        if len(parts) == 1:
113            if not fromlist:
114                return top_module
115
116            if not top_module.__dict__.get('__ispkg__'):
117                # __ispkg__ isn't defined (the module was not imported by us),
118                # or it is zero.
119                #
120                # In the former case, there is no way that we could import
121                # sub-modules that occur in the fromlist (but we can't raise an
122                # error because it may just be names) because we don't know how
123                # to deal with packages that were imported by other systems.
124                #
125                # In the latter case (__ispkg__ == 0), there can't be any sub-
126                # modules present, so we can just return.
127                #
128                # In both cases, since len(parts) == 1, the top_module is also
129                # the "bottom" which is the defined return when a fromlist
130                # exists.
131                return top_module
132
133        importer = top_module.__dict__.get('__importer__')
134        if importer:
135            return importer._finish_import(top_module, parts[1:], fromlist)
136
137        # Grrr, some people "import os.path" or do "from os.path import ..."
138        if len(parts) == 2 and hasattr(top_module, parts[1]):
139            if fromlist:
140                return getattr(top_module, parts[1])
141            else:
142                return top_module
143
144        # If the importer does not exist, then we have to bail. A missing
145        # importer means that something else imported the module, and we have
146        # no knowledge of how to get sub-modules out of the thing.
147        raise ImportError, 'No module named ' + fqname
148
149    def _determine_import_context(self, globals):
150        """Returns the context in which a module should be imported.
151
152        The context could be a loaded (package) module and the imported module
153        will be looked for within that package. The context could also be None,
154        meaning there is no context -- the module should be looked for as a
155        "top-level" module.
156        """
157
158        if not globals or not globals.get('__importer__'):
159            # globals does not refer to one of our modules or packages. That
160            # implies there is no relative import context (as far as we are
161            # concerned), and it should just pick it off the standard path.
162            return None
163
164        # The globals refer to a module or package of ours. It will define
165        # the context of the new import. Get the module/package fqname.
166        parent_fqname = globals['__name__']
167
168        # if a package is performing the import, then return itself (imports
169        # refer to pkg contents)
170        if globals['__ispkg__']:
171            parent = sys.modules[parent_fqname]
172            assert globals is parent.__dict__
173            return parent
174
175        i = parent_fqname.rfind('.')
176
177        # a module outside of a package has no particular import context
178        if i == -1:
179            return None
180
181        # if a module in a package is performing the import, then return the
182        # package (imports refer to siblings)
183        parent_fqname = parent_fqname[:i]
184        parent = sys.modules[parent_fqname]
185        assert parent.__name__ == parent_fqname
186        return parent
187
188    def _import_top_module(self, name):
189        # scan sys.path looking for a location in the filesystem that contains
190        # the module, or an Importer object that can import the module.
191        for item in sys.path:
192            if isinstance(item, _StringType):
193                module = self.fs_imp.import_from_dir(item, name)
194            else:
195                module = item.import_top(name)
196            if module:
197                return module
198        return None
199
200    def _reload_hook(self, module):
201        "Python calls this hook to reload a module."
202
203        # reloading of a module may or may not be possible (depending on the
204        # importer), but at least we can validate that it's ours to reload
205        importer = module.__dict__.get('__importer__')
206        if not importer:
207            ### oops. now what...
208            pass
209
210        # okay. it is using the imputil system, and we must delegate it, but
211        # we don't know what to do (yet)
212        ### we should blast the module dict and do another get_code(). need to
213        ### flesh this out and add proper docco...
214        raise SystemError, "reload not yet implemented"
215
216
217class Importer:
218    "Base class for replacing standard import functions."
219
220    def import_top(self, name):
221        "Import a top-level module."
222        return self._import_one(None, name, name)
223
224    ######################################################################
225    #
226    # PRIVATE METHODS
227    #
228    def _finish_import(self, top, parts, fromlist):
229        # if "a.b.c" was provided, then load the ".b.c" portion down from
230        # below the top-level module.
231        bottom = self._load_tail(top, parts)
232
233        # if the form is "import a.b.c", then return "a"
234        if not fromlist:
235            # no fromlist: return the top of the import tree
236            return top
237
238        # the top module was imported by self.
239        #
240        # this means that the bottom module was also imported by self (just
241        # now, or in the past and we fetched it from sys.modules).
242        #
243        # since we imported/handled the bottom module, this means that we can
244        # also handle its fromlist (and reliably use __ispkg__).
245
246        # if the bottom node is a package, then (potentially) import some
247        # modules.
248        #
249        # note: if it is not a package, then "fromlist" refers to names in
250        #       the bottom module rather than modules.
251        # note: for a mix of names and modules in the fromlist, we will
252        #       import all modules and insert those into the namespace of
253        #       the package module. Python will pick up all fromlist names
254        #       from the bottom (package) module; some will be modules that
255        #       we imported and stored in the namespace, others are expected
256        #       to be present already.
257        if bottom.__ispkg__:
258            self._import_fromlist(bottom, fromlist)
259
260        # if the form is "from a.b import c, d" then return "b"
261        return bottom
262
263    def _import_one(self, parent, modname, fqname):
264        "Import a single module."
265
266        # has the module already been imported?
267        try:
268            return sys.modules[fqname]
269        except KeyError:
270            pass
271
272        # load the module's code, or fetch the module itself
273        result = self.get_code(parent, modname, fqname)
274        if result is None:
275            return None
276
277        module = self._process_result(result, fqname)
278
279        # insert the module into its parent
280        if parent:
281            setattr(parent, modname, module)
282        return module
283
284    def _process_result(self, result, fqname):
285        ispkg, code, values = result
286        # did get_code() return an actual module? (rather than a code object)
287        is_module = isinstance(code, _ModuleType)
288
289        # use the returned module, or create a new one to exec code into
290        if is_module:
291            module = code
292        else:
293            module = imp.new_module(fqname)
294
295        ### record packages a bit differently??
296        module.__importer__ = self
297        module.__ispkg__ = ispkg
298
299        # insert additional values into the module (before executing the code)
300        module.__dict__.update(values)
301
302        # the module is almost ready... make it visible
303        sys.modules[fqname] = module
304
305        # execute the code within the module's namespace
306        if not is_module:
307            try:
308                exec code in module.__dict__
309            except:
310                if fqname in sys.modules:
311                    del sys.modules[fqname]
312                raise
313
314        # fetch from sys.modules instead of returning module directly.
315        # also make module's __name__ agree with fqname, in case
316        # the "exec code in module.__dict__" played games on us.
317        module = sys.modules[fqname]
318        module.__name__ = fqname
319        return module
320
321    def _load_tail(self, m, parts):
322        """Import the rest of the modules, down from the top-level module.
323
324        Returns the last module in the dotted list of modules.
325        """
326        for part in parts:
327            fqname = "%s.%s" % (m.__name__, part)
328            m = self._import_one(m, part, fqname)
329            if not m:
330                raise ImportError, "No module named " + fqname
331        return m
332
333    def _import_fromlist(self, package, fromlist):
334        'Import any sub-modules in the "from" list.'
335
336        # if '*' is present in the fromlist, then look for the '__all__'
337        # variable to find additional items (modules) to import.
338        if '*' in fromlist:
339            fromlist = list(fromlist) + \
340                       list(package.__dict__.get('__all__', []))
341
342        for sub in fromlist:
343            # if the name is already present, then don't try to import it (it
344            # might not be a module!).
345            if sub != '*' and not hasattr(package, sub):
346                subname = "%s.%s" % (package.__name__, sub)
347                submod = self._import_one(package, sub, subname)
348                if not submod:
349                    raise ImportError, "cannot import name " + subname
350
351    def _do_import(self, parent, parts, fromlist):
352        """Attempt to import the module relative to parent.
353
354        This method is used when the import context specifies that <self>
355        imported the parent module.
356        """
357        top_name = parts[0]
358        top_fqname = parent.__name__ + '.' + top_name
359        top_module = self._import_one(parent, top_name, top_fqname)
360        if not top_module:
361            # this importer and parent could not find the module (relatively)
362            return None
363
364        return self._finish_import(top_module, parts[1:], fromlist)
365
366    ######################################################################
367    #
368    # METHODS TO OVERRIDE
369    #
370    def get_code(self, parent, modname, fqname):
371        """Find and retrieve the code for the given module.
372
373        parent specifies a parent module to define a context for importing. It
374        may be None, indicating no particular context for the search.
375
376        modname specifies a single module (not dotted) within the parent.
377
378        fqname specifies the fully-qualified module name. This is a
379        (potentially) dotted name from the "root" of the module namespace
380        down to the modname.
381        If there is no parent, then modname==fqname.
382
383        This method should return None, or a 3-tuple.
384
385        * If the module was not found, then None should be returned.
386
387        * The first item of the 2- or 3-tuple should be the integer 0 or 1,
388            specifying whether the module that was found is a package or not.
389
390        * The second item is the code object for the module (it will be
391            executed within the new module's namespace). This item can also
392            be a fully-loaded module object (e.g. loaded from a shared lib).
393
394        * The third item is a dictionary of name/value pairs that will be
395            inserted into new module before the code object is executed. This
396            is provided in case the module's code expects certain values (such
397            as where the module was found). When the second item is a module
398            object, then these names/values will be inserted *after* the module
399            has been loaded/initialized.
400        """
401        raise RuntimeError, "get_code not implemented"
402
403
404######################################################################
405#
406# Some handy stuff for the Importers
407#
408
409# byte-compiled file suffix character
410_suffix_char = __debug__ and 'c' or 'o'
411
412# byte-compiled file suffix
413_suffix = '.py' + _suffix_char
414
415def _compile(pathname, timestamp):
416    """Compile (and cache) a Python source file.
417
418    The file specified by <pathname> is compiled to a code object and
419    returned.
420
421    Presuming the appropriate privileges exist, the bytecodes will be
422    saved back to the filesystem for future imports. The source file's
423    modification timestamp must be provided as a Long value.
424    """
425    codestring = open(pathname, 'rU').read()
426    if codestring and codestring[-1] != '\n':
427        codestring = codestring + '\n'
428    code = __builtin__.compile(codestring, pathname, 'exec')
429
430    # try to cache the compiled code
431    try:
432        f = open(pathname + _suffix_char, 'wb')
433    except IOError:
434        pass
435    else:
436        f.write('\0\0\0\0')
437        f.write(struct.pack('<I', timestamp))
438        marshal.dump(code, f)
439        f.flush()
440        f.seek(0, 0)
441        f.write(imp.get_magic())
442        f.close()
443
444    return code
445
446_os_stat = _os_path_join = None
447def _os_bootstrap():
448    "Set up 'os' module replacement functions for use during import bootstrap."
449
450    names = sys.builtin_module_names
451
452    join = None
453    if 'posix' in names:
454        sep = '/'
455        from posix import stat
456    elif 'nt' in names:
457        sep = '\\'
458        from nt import stat
459    elif 'dos' in names:
460        sep = '\\'
461        from dos import stat
462    elif 'os2' in names:
463        sep = '\\'
464        from os2 import stat
465    else:
466        raise ImportError, 'no os specific module found'
467
468    if join is None:
469        def join(a, b, sep=sep):
470            if a == '':
471                return b
472            lastchar = a[-1:]
473            if lastchar == '/' or lastchar == sep:
474                return a + b
475            return a + sep + b
476
477    global _os_stat
478    _os_stat = stat
479
480    global _os_path_join
481    _os_path_join = join
482
483def _os_path_isdir(pathname):
484    "Local replacement for os.path.isdir()."
485    try:
486        s = _os_stat(pathname)
487    except OSError:
488        return None
489    return (s.st_mode & 0170000) == 0040000
490
491def _timestamp(pathname):
492    "Return the file modification time as a Long."
493    try:
494        s = _os_stat(pathname)
495    except OSError:
496        return None
497    return long(s.st_mtime)
498
499
500######################################################################
501#
502# Emulate the import mechanism for builtin and frozen modules
503#
504class BuiltinImporter(Importer):
505    def get_code(self, parent, modname, fqname):
506        if parent:
507            # these modules definitely do not occur within a package context
508            return None
509
510        # look for the module
511        if imp.is_builtin(modname):
512            type = imp.C_BUILTIN
513        elif imp.is_frozen(modname):
514            type = imp.PY_FROZEN
515        else:
516            # not found
517            return None
518
519        # got it. now load and return it.
520        module = imp.load_module(modname, None, modname, ('', '', type))
521        return 0, module, { }
522
523
524######################################################################
525#
526# Internal importer used for importing from the filesystem
527#
528class _FilesystemImporter(Importer):
529    def __init__(self):
530        self.suffixes = [ ]
531
532    def add_suffix(self, suffix, importFunc):
533        assert hasattr(importFunc, '__call__')
534        self.suffixes.append((suffix, importFunc))
535
536    def import_from_dir(self, dir, fqname):
537        result = self._import_pathname(_os_path_join(dir, fqname), fqname)
538        if result:
539            return self._process_result(result, fqname)
540        return None
541
542    def get_code(self, parent, modname, fqname):
543        # This importer is never used with an empty parent. Its existence is
544        # private to the ImportManager. The ImportManager uses the
545        # import_from_dir() method to import top-level modules/packages.
546        # This method is only used when we look for a module within a package.
547        assert parent
548
549        for submodule_path in parent.__path__:
550            code = self._import_pathname(_os_path_join(submodule_path, modname), fqname)
551            if code is not None:
552                return code
553        return self._import_pathname(_os_path_join(parent.__pkgdir__, modname),
554                                     fqname)
555
556    def _import_pathname(self, pathname, fqname):
557        if _os_path_isdir(pathname):
558            result = self._import_pathname(_os_path_join(pathname, '__init__'),
559                                           fqname)
560            if result:
561                values = result[2]
562                values['__pkgdir__'] = pathname
563                values['__path__'] = [ pathname ]
564                return 1, result[1], values
565            return None
566
567        for suffix, importFunc in self.suffixes:
568            filename = pathname + suffix
569            try:
570                finfo = _os_stat(filename)
571            except OSError:
572                pass
573            else:
574                return importFunc(filename, finfo, fqname)
575        return None
576
577######################################################################
578#
579# SUFFIX-BASED IMPORTERS
580#
581
582def py_suffix_importer(filename, finfo, fqname):
583    file = filename[:-3] + _suffix
584    t_py = long(finfo[8])
585    t_pyc = _timestamp(file)
586
587    code = None
588    if t_pyc is not None and t_pyc >= t_py:
589        f = open(file, 'rb')
590        if f.read(4) == imp.get_magic():
591            t = struct.unpack('<I', f.read(4))[0]
592            if t == t_py:
593                code = marshal.load(f)
594        f.close()
595    if code is None:
596        file = filename
597        code = _compile(file, t_py)
598
599    return 0, code, { '__file__' : file }
600
601class DynLoadSuffixImporter:
602    def __init__(self, desc):
603        self.desc = desc
604
605    def import_file(self, filename, finfo, fqname):
606        fp = open(filename, self.desc[1])
607        module = imp.load_module(fqname, fp, filename, self.desc)
608        module.__file__ = filename
609        return 0, module, { }
610
611
612######################################################################
613
614def _print_importers():
615    items = sys.modules.items()
616    items.sort()
617    for name, module in items:
618        if module:
619            print name, module.__dict__.get('__importer__', '-- no importer')
620        else:
621            print name, '-- non-existent module'
622
623def _test_revamp():
624    ImportManager().install()
625    sys.path.insert(0, BuiltinImporter())
626
627######################################################################
628
629#
630# TODO
631#
632# from Finn Bock:
633#   type(sys) is not a module in Jython. what to use instead?
634#   imp.C_EXTENSION is not in Jython. same for get_suffixes and new_module
635#
636#   given foo.py of:
637#      import sys
638#      sys.modules['foo'] = sys
639#
640#   ---- standard import mechanism
641#   >>> import foo
642#   >>> foo
643#   <module 'sys' (built-in)>
644#
645#   ---- revamped import mechanism
646#   >>> import imputil
647#   >>> imputil._test_revamp()
648#   >>> import foo
649#   >>> foo
650#   <module 'foo' from 'foo.py'>
651#
652#
653# from MAL:
654#   should BuiltinImporter exist in sys.path or hard-wired in ImportManager?
655#   need __path__ processing
656#   performance
657#   move chaining to a subclass [gjs: it's been nuked]
658#   deinstall should be possible
659#   query mechanism needed: is a specific Importer installed?
660#   py/pyc/pyo piping hooks to filter/process these files
661#   wish list:
662#     distutils importer hooked to list of standard Internet repositories
663#     module->file location mapper to speed FS-based imports
664#     relative imports
665#     keep chaining so that it can play nice with other import hooks
666#
667# from Gordon:
668#   push MAL's mapper into sys.path[0] as a cache (hard-coded for apps)
669#
670# from Guido:
671#   need to change sys.* references for rexec environs
672#   need hook for MAL's walk-me-up import strategy, or Tim's absolute strategy
673#   watch out for sys.modules[...] is None
674#   flag to force absolute imports? (speeds _determine_import_context and
675#       checking for a relative module)
676#   insert names of archives into sys.path  (see quote below)
677#   note: reload does NOT blast module dict
678#   shift import mechanisms and policies around; provide for hooks, overrides
679#       (see quote below)
680#   add get_source stuff
681#   get_topcode and get_subcode
682#   CRLF handling in _compile
683#   race condition in _compile
684#   refactoring of os.py to deal with _os_bootstrap problem
685#   any special handling to do for importing a module with a SyntaxError?
686#       (e.g. clean up the traceback)
687#   implement "domain" for path-type functionality using pkg namespace
688#       (rather than FS-names like __path__)
689#   don't use the word "private"... maybe "internal"
690#
691#
692# Guido's comments on sys.path caching:
693#
694# We could cache this in a dictionary: the ImportManager can have a
695# cache dict mapping pathnames to importer objects, and a separate
696# method for coming up with an importer given a pathname that's not yet
697# in the cache.  The method should do a stat and/or look at the
698# extension to decide which importer class to use; you can register new
699# importer classes by registering a suffix or a Boolean function, plus a
700# class.  If you register a new importer class, the cache is zapped.
701# The cache is independent from sys.path (but maintained per
702# ImportManager instance) so that rearrangements of sys.path do the
703# right thing.  If a path is dropped from sys.path the corresponding
704# cache entry is simply no longer used.
705#
706# My/Guido's comments on factoring ImportManager and Importer:
707#
708# > However, we still have a tension occurring here:
709# >
710# > 1) implementing policy in ImportManager assists in single-point policy
711# >    changes for app/rexec situations
712# > 2) implementing policy in Importer assists in package-private policy
713# >    changes for normal, operating conditions
714# >
715# > I'll see if I can sort out a way to do this. Maybe the Importer class will
716# > implement the methods (which can be overridden to change policy) by
717# > delegating to ImportManager.
718#
719# Maybe also think about what kind of policies an Importer would be
720# likely to want to change.  I have a feeling that a lot of the code
721# there is actually not so much policy but a *necessity* to get things
722# working given the calling conventions for the __import__ hook: whether
723# to return the head or tail of a dotted name, or when to do the "finish
724# fromlist" stuff.
725#
726