Python now supports checking bytecode cache up-to-dateness with a hash of the
source contents rather than volatile source metadata. See the PEP for details.
While a fairly straightforward idea, quite a lot of code had to be modified due
to the pervasiveness of pyc implementation details in the codebase. Changes in
this commit include:
- The core changes to importlib to understand how to read, validate, and
regenerate hash-based pycs.
- Support for generating hash-based pycs in py_compile and compileall.
- Modifications to our siphash implementation to support passing a custom
key. We then expose it to importlib through _imp.
- Updates to all places in the interpreter, standard library, and tests that
manually generate or parse pyc files to grok the new format.
- Support in the interpreter command line code for long options like
--check-hash-based-pycs.
- Tests and documentation for all of the above.
Use sys.modules.get() in the "with _ModuleLockManager(name):" block
to protect the dictionary key with the module lock and use an atomic
get to prevent race condition.
Remove also _bootstrap._POPULATE since it was unused
(_bootstrap_external now has its own _POPULATE object), add a new
_SENTINEL object instead.
* Rewrite importlib _get_module_lock(): it is now responsible to hold
the imp lock directly.
* _find_and_load() now holds the module lock to check if name is in
sys.modules to prevent a race condition
Previously AttributeError was raised, but that's not very reflective of the fact that the requested module can't be found since the specified parent isn't actually a package.
PEP 432 specifies a number of large changes to interpreter startup code, including exposing a cleaner C-API. The major changes depend on a number of smaller changes. This patch includes all those smaller changes.
Special thanks to INADA Naoki for pushing the patch through
the last mile, Serhiy Storchaka for reviewing the code, and to
Victor Stinner for suggesting the idea (originally implemented
in the PyPy project).
Handling zero-argument super() in __init_subclass__ and
__set_name__ involved moving __class__ initialisation to
type.__new__. This requires cooperation from custom
metaclasses to ensure that the new __classcell__ entry
is passed along appropriately.
The initial implementation of that change resulted in abruptly
broken zero-argument super() support in metaclasses that didn't
adhere to the new requirements (such as Django's metaclass for
Model definitions).
The updated approach adopted here instead emits a deprecation
warning for those cases, and makes them work the same way they
did in Python 3.5.
This patch also improves the related class machinery documentation
to cover these details and to include more reader-friendly
cross-references and index entries.
The __class__ cell used by zero-argument super() is now initialized
from type.__new__ rather than __build_class__, so class methods
relying on that will now work correctly when called from metaclass
methods during class creation.
Patch by Martin Teichmann.
Issue #27213: Rework CALL_FUNCTION* opcodes to produce shorter and more
efficient bytecode:
* CALL_FUNCTION now only accepts position arguments
* CALL_FUNCTION_KW accepts position arguments and keyword arguments, but keys
of keyword arguments are packed into a constant tuple.
* CALL_FUNCTION_EX is the most generic, it expects a tuple and a dict for
positional and keyword arguments.
CALL_FUNCTION_VAR and CALL_FUNCTION_VAR_KW opcodes have been removed.
2 tests of test_traceback are currently broken: skip test, the issue #28050 was
created to track the issue.
Patch by Demur Rumed, design by Serhiy Storchaka, reviewed by Serhiy Storchaka
and Victor Stinner.
Windows.
Originally only b'PYTHONCASEOK' was being checked for in os.environ,
but that won't work under Windows where all environment variables are
strings (on OS X they are bytes).
Thanks to Eryk Sun for the bug report.
modules can't be lazily loaded.
Thanks to Python 3.6 allowing for types.ModuleType to have its
__class__ mutated, the restriction can be lifted by calling
create_module() on the wrapped loader.
Issue #26637: The importlib module now emits an ImportError rather than a
TypeError if __import__() is tried during the Python shutdown process but
sys.path is already cleared (set to None).
Issue #26538: libregrtest: Fix setup_tests() to keep module.__path__ type
(_NamespacePath), don't convert to a list.
Add _NamespacePath.__setitem__() method to importlib._bootstrap_external.
importlib.util.LazyLoader.
The class was checking its argument as to whether its implementation
of create_module() came directly from importlib.abc.Loader. The
problem is that the classes coming from imoprtlib.machinery do not
directly inherit from the ABC as they come from _frozen_importlib.
Because the documentation has always said that create_module() was
ignored, the check has simply been removed.
In a previous change, __spec__.parent was prioritized over
__package__. That is a backwards-compatibility break, but we do
eventually want __spec__ to be the ground truth for module details. So
this change reverts the change in semantics and instead raises an
ImportWarning when __package__ != __spec__.parent to give people time
to adjust to using spec objects.
Issue #26107: The format of the co_lnotab attribute of code objects changes to
support negative line number delta.
Changes:
* assemble_lnotab(): if line number delta is less than -128 or greater than
127, emit multiple (offset_delta, lineno_delta) in co_lnotab
* update functions decoding co_lnotab to use signed 8-bit integers
- dis.findlinestarts()
- PyCode_Addr2Line()
- _PyCode_CheckLineNumber()
- frame_setlineno()
* update lnotab_notes.txt
* increase importlib MAGIC_NUMBER to 3361
* document the change in What's New in Python 3.6
* cleanup also PyCode_Optimize() to use better variable names
not defined for a relative import.
This is the start of work to try and clean up import semantics to rely
more on a module's spec than on the myriad attributes that get set on
a module. Thanks to Rose Ames for the patch.
Summary of changes:
1. Coroutines now have a distinct, separate from generators
type at the C level: PyGen_Type, and a new typedef PyCoroObject.
PyCoroObject shares the initial segment of struct layout with
PyGenObject, making it possible to reuse existing generators
machinery. The new type is exposed as 'types.CoroutineType'.
As a consequence of having a new type, CO_GENERATOR flag is
no longer applied to coroutines.
2. Having a separate type for coroutines made it possible to add
an __await__ method to the type. Although it is not used by the
interpreter (see details on that below), it makes coroutines
naturally (without using __instancecheck__) conform to
collections.abc.Coroutine and collections.abc.Awaitable ABCs.
[The __instancecheck__ is still used for generator-based
coroutines, as we don't want to add __await__ for generators.]
3. Add new opcode: GET_YIELD_FROM_ITER. The opcode is needed to
allow passing native coroutines to the YIELD_FROM opcode.
Before this change, 'yield from o' expression was compiled to:
(o)
GET_ITER
LOAD_CONST
YIELD_FROM
Now, we use GET_YIELD_FROM_ITER instead of GET_ITER.
The reason for adding a new opcode is that GET_ITER is used
in some contexts (such as 'for .. in' loops) where passing
a coroutine object is invalid.
4. Add two new introspection functions to the inspec module:
getcoroutinestate(c) and getcoroutinelocals(c).
5. inspect.iscoroutine(o) is updated to test if 'o' is a native
coroutine object. Before this commit it used abc.Coroutine,
and it was requested to update inspect.isgenerator(o) to use
abc.Generator; it was decided, however, that inspect functions
should really be tailored for checking for native types.
6. sys.set_coroutine_wrapper(w) API is updated to work with only
native coroutines. Since types.coroutine decorator supports
any type of callables now, it would be confusing that it does
not work for all types of coroutines.
7. Exceptions logic in generators C implementation was updated
to raise clearer messages for coroutines:
Before: TypeError("generator raised StopIteration")
After: TypeError("coroutine raised StopIteration")
Known limitations of the current implementation:
- documentation changes are incomplete
- there's a reference leak I haven't tracked down yet
The leak is most visible by running:
./python -m test -R3:3 test_importlib
However, you can also see it by running:
./python -X showrefcount
Importing the array or _testmultiphase modules, and
then deleting them from both sys.modules and the local
namespace shows significant increases in the total
number of active references each cycle. By contrast,
with _testcapi (which continues to use single-phase
initialisation) the global refcounts stabilise after
a couple of cycles.
The concept of .pyo files no longer exists. Now .pyc files have an
optional `opt-` tag which specifies if any extra optimizations beyond
the peepholer were applied.
importlib.abc.Loader.exec_module() is also defined.
Before this change, create_module() was optional **and** could return
None to trigger default semantics. This change now reduces the
options for choosing default semantics to one and in the most
backporting-friendly way (define create_module() to return None).
Along the way, dismantle importlib._bootstrap._SpecMethods as it was
no longer relevant and constructing the new function required
partially dismantling the class anyway.
Serhiy Storchaka
Barry Warsaw
Mark Shannon
Brett Cannon
Benjamin Peterson
Yury Selivanov
Antoine Pitrou
Victor Stinner
Milan Oberkirch
Garvit Khatri
Eric Snow
Mandeep Bhutani
Matthias Bussonnier
Steve Dower
Nick Coghlan
Brett Cannon
Martin Panter
Eric V. Smith
Larry Hastings
Yury Selivanov