It seems better to cache the finder for the cwd under its full path
insetad of '' in case the cwd changes. Otherwise FileFinder needs to
dynamically change itself based on whether it is given '' instead of
caching a finder for every change to the cwd.
Thanks to os.environ under Windows only updating the dict and not the
environment itself (as exposed by nt.environ), tests using
PYTHONCASEOK always fail. Now the tests are skipped when os.environ
does not do what is expected.
This required updating the code to use posix instead of os. This is
all being done to make bootstrapping easier to removing dependencies
that are kept in importlib.__init__ and thus outside of the single
file to bootstrap from.
paths.
__import__ does a little trick when importing from bytecode by
back-patching the co_filename paths to point to the file location
where the code object was loaded from, *not* where the code object was
originally created. This allows co_filename to point to a valid path.
Problem is that co_filename is immutable from Python, so a private
function -- imp._fix_co_filename() -- had to be introduced in order to
get things working properly. Originally the plan was to add a file
argument to marshal.loads(), but that failed as the algorithm used by
__import__ is not fully recursive as one might expect, so to be fully
backwards-compatible the code used by __import__ needed to be exposed.
This closes issue #6811 by taking a different approach than outlined
in the issue.
test_importlib is that it discovers special little race conditions. For
instance, it turns out that importlib would throw an exception if two different
Python processes both tried to create the __pycache__ directory as one process
would succeed, causing the other process to fail as it didn't expect to get any
"help". So now importlib simply stays calm and just accepts someone else did
the work of creating the __pycache__ directory for it, moving on with life.
Closes issue #9572.
This required moving the class from importlib/abc.py into
importlib/_bootstrap.py and jiggering some code to work better with the class.
This included changing how the file finder worked to better meet import
semantics. This also led to fixing importlib to handle the empty string from
sys.path as import currently does (and making me wish we didn't support that
instead just required people to insert '.' instead to represent cwd).
It also required making the new set_data abstractmethod create
any needed subdirectories implicitly thanks to __pycache__ (it was either this
or grow the SourceLoader ABC to gain an 'exists' method and either a mkdir
method or have set_data with no data arg mean to create a directory).
Lastly, as an optimization the file loaders cache the file path where the
finder found something to use for loading (this is thanks to having a
sourceless loader separate from the source loader to simplify the code and
cut out stat calls).
Unfortunately test_runpy assumed a loader would always work for a module, even
if you changed from underneath it what it was expected to work with. By simply
dropping the previous loader in test_runpy so the proper loader can be returned
by the finder fixed the failure.
At this point importlib deviates from import on two points:
1. The exception raised when trying to import a file is different (import does
an explicit file check to print a special message, importlib just says the path
cannot be imported as if it was just some module name).
2. the co_filename on a code object is not being set to where bytecode was
actually loaded from instead of where the marshalled code object originally
came from (a solution for this has already been agreed upon on python-dev but has
not been implemented yet; issue8611).
This required moving the class from importlib/abc.py into
importlib/_bootstrap.py and jiggering some code to work better with the class.
This included changing how the file finder worked to better meet import
semantics. This also led to fixing importlib to handle the empty string from
sys.path as import currently does (and making me wish we didn't support that
instead just required people to insert '.' instead to represent cwd).
It also required making the new set_data abstractmethod create
any needed subdirectories implicitly thanks to __pycache__ (it was either this
or grow the SourceLoader ABC to gain an 'exists' method and either a mkdir
method or have set_data with no data arg mean to create a directory).
Lastly, as an optimization the file loaders cache the file path where the
finder found something to use for loading (this is thanks to having a
sourceless loader separate from the source loader to simplify the code and
cut out stat calls).
Unfortunately test_runpy assumed a loader would always work for a module, even
if you changed from underneath it what it was expected to work with. By simply
dropping the previous loader in test_runpy so the proper loader can be returned
by the finder fixed the failure.
At this point importlib deviates from import on two points:
1. The exception raised when trying to import a file is different (import does
an explicit file check to print a special message, importlib just says the path
cannot be imported as if it was just some module name).
2. the co_filename on a code object is not being set to where bytecode was
actually loaded from instead of where the marshalled code object originally
came from (a solution for this has already been agreed upon on python-dev but has
not been implemented yet; issue8611).
This required moving the class from importlib/abc.py into
importlib/_bootstrap.py and jiggering some code to work better with the class.
This included changing how the file finder worked to better meet import
semantics. This also led to fixing importlib to handle the empty string from
sys.path as import currently does (and making me wish we didn't support that
instead just required people to insert '.' instead to represent cwd).
It also required making the new set_data abstractmethod create
any needed subdirectories implicitly thanks to __pycache__ (it was either this
or grow the SourceLoader ABC to gain an 'exists' method and either a mkdir
method or have set_data with no data arg mean to create a directory).
Lastly, as an optimization the file loaders cache the file path where the
finder found something to use for loading (this is thanks to having a
sourceless loader separate from the source loader to simplify the code and
cut out stat calls).
Unfortunately test_runpy assumed a loader would always work for a module, even
if you changed from underneath it what it was expected to work with. By simply
dropping the previous loader in test_runpy so the proper loader can be returned
by the finder fixed the failure.
At this point importlib deviates from import on two points:
1. The exception raised when trying to import a file is different (import does
an explicit file check to print a special message, importlib just says the path
cannot be imported as if it was just some module name).
2. the co_filename on a code object is not being set to where bytecode was
actually loaded from instead of where the marshalled code object originally
came from (a solution for this has already been agreed upon on python-dev but has
not been implemented yet; issue8611).
Required updating code relying on other modules to switch to _bootstrap's
unique module requirements. This led to the realization that
get_code was being too liberal in its exception catching when calling set_data
by blindly grabbing IOError. Shifted the responsibility of safely ignoring
writes to a read-only path to set_data.
Importlib is still not relying on SourceLoader yet; requires creating a
SourcelessLoader and updating the source finder.
SourceLoader is a simplification of both PyLoader and PyPycLoader. If one only
wants to use source, then they need to only implement get_data and
get_filename. To also use bytecode -- sourceless loading is not supported --
then two abstract methods -- path_mtime and set_data -- need to be implemented.
Compared to PyLoader and PyPycLoader, there are less abstract methods
introduced and bytecode files become an optimization controlled by the ABC and
hidden from the user (this need came about as PEP 3147 showed that not treating
bytecode as an optimization can cause problems for compatibility).
PyLoader is deprecated in favor of SourceLoader. To be compatible from Python
3.1 onwards, a subclass need only use simple methods for source_path and
is_package. Otherwise conditional subclassing based on whether Python 3.1 or
Python 3.2 is being is the only change. The documentation and docstring for
PyLoader explain what is exactly needed.
PyPycLoader is deprecated also in favor of SourceLoader. Because PEP 3147
shifted bytecode path details so much, there is no foolproof way to provide
backwards-compatibility with SourceLoader. Because of this the class is simply
deprecated and users should move to SourceLoader (and optionally PyLoader for
Python 3.1). This does lead to a loss of support for sourceless loading
unfortunately.
At some point before Python 3.2 is released, SourceLoader will be moved over to
importlib._bootstrap so that the core code of importlib relies on the new code
instead of the old PyPycLoader code. This commit is being done now so that
there is no issue in having the API in Python 3.1a1.
Brett Cannon
Philip Jenvey
Charles-François Natali
Antoine Pitrou
Benjamin Peterson
Meador Inge
Florent Xicluna
Vinay Sajip
Éric Araujo
Victor Stinner
Ezio Melotti
Eric V. Smith
Raymond Hettinger
Brett Cannon
Barry Warsaw