Grabbing the recently released Python 3.6.0b1, I tried running one of my test suites, but found that some assertions were failing to assert when the package was loaded as a zip file (such as with pytest-runner installed dependencies). I distilled the issue to this:$ cat > mod.pydef test(val):        assert(val)        print(val)$ zip mod.zip mod.pyupdating: mod.py (deflated 20%)$ rm mod.py$ pythonPython 3.6.0b1 (v3.6.0b1:5b0ca4ed5e2f, Sep 12 2016, 09:24:46) [GCC 4.2.1 (Apple Inc. build 5666) (dot 3)] on darwinType "help", "copyright", "credits" or "license" for more information.>>> import sys>>> sys.path.append('mod.zip')>>> import mod>>> mod.test(False)False>>> mod.__loader__<zipimporter object "mod.zip">>>> sys.flags.optimize0I would have expected the call to mod.test to have raised an AssertionError, and on Python 3.5 it does.I searched the what's new and didn't see anything advertising this change, so I suspect it's an unintentional regression.I'm including Brett for his familiarity with importlib, but welcome re-assignment.If I can do more to help, let me know.

I've added Greg and Thomas in case they have any ideas as they have looked at zipimport more recently than I have.

This shouldn't be happening and makes no sense.  It looks like the assert statement was removed at import code compilation time given the pdb trace with it from a zip file vs with it outside of a zip file:>>> pdb.run('mod.test(False)')> <string>(1)<module>()(Pdb) nFalse--Return--> <string>(1)<module>()->None(Pdb) n>>> pdb.run('mod.test(False)')> <string>(1)<module>()->None(Pdb) s--Call--> oss/cpython/build/mod.zip/mod.py(1)test()-> def test(val):(Pdb) s> oss/cpython/build/mod.zip/mod.py(3)test()-> print(val)(Pdb) sFalse--Return--> oss/cpython/build/mod.zip/mod.py(3)test()->None-> print(val)(Pdb) s--Return--> <string>(1)<module>()->None(Pdb) s> oss/cpython/default/Lib/bdb.py(435)run()-> self.quitting = True(Pdb) s>>> vs no zip file:>>> pdb.run('mod.test(False)')> <string>(1)<module>()(Pdb) s--Call--> oss/cpython/build/mod.py(1)test()-> def test(val):(Pdb) s> oss/cpython/build/mod.py(2)test()-> assert(val)(Pdb) sAssertionError> oss/cpython/build/mod.py(2)test()-> assert(val)(Pdb) s--Return--> oss/cpython/build/mod.py(2)test()->None-> assert(val)(Pdb) sAssertionError> <string>(1)<module>()(Pdb) s--Return--> <string>(1)<module>()->None(Pdb) sAssertionError> oss/cpython/default/Lib/bdb.py(431)run()-> exec(cmd, globals, locals)(Pdb) s> oss/cpython/default/Lib/bdb.py(432)run()-> except BdbQuit:(Pdb) s> oss/cpython/default/Lib/bdb.py(435)run()-> self.quitting = True(Pdb) sTraceback (most recent call last):  File "<stdin>", line 1, in <module>  File "oss/cpython/default/Lib/pdb.py", line 1568, in run    Pdb().run(statement, globals, locals)  File "oss/cpython/default/Lib/bdb.py", line 431, in run    exec(cmd, globals, locals)  File "<string>", line 1, in <module>  File "oss/cpython/build/mod.py", line 2, in test    assert(val)AssertionError

from the zip:>>> dis.dis(mod.test)  3           0 LOAD_GLOBAL              0 (print)              2 LOAD_FAST                0 (val)              4 CALL_FUNCTION            1              6 POP_TOP              8 LOAD_CONST               0 (None)             10 RETURN_VALUEfrom the filesystem:>>> dis.dis(mod.test)  2           0 LOAD_FAST                0 (val)              2 POP_JUMP_IF_TRUE         8              4 LOAD_GLOBAL              0 (AssertionError)              6 RAISE_VARARGS            1  3     >>    8 LOAD_GLOBAL              1 (print)             10 LOAD_FAST                0 (val)             12 CALL_FUNCTION            1             14 POP_TOP             16 LOAD_CONST               0 (None)             18 RETURN_VALUE

This is introduced in663a62bcf9c9. The compile optimize flag is opened in the change.

Bah, that's meant to be a -1 in that change, not 1. I must have typo'd when moving from the default branch to put it in 3.5.Anyone's welcome to fix it, or I'll get it when I have time for CPython work.

Here is a patch with a test case.

lgtm

New changeset7bec326972f5 by Berker Peksag in branch '3.5':Issue#28131: Fix a regression in zipimport's compile_source()https://hg.python.org/cpython/rev/7bec326972f5New changeset7a6c0c4e6072 by Berker Peksag in branch '3.6':Issue#28131: Merge from 3.5https://hg.python.org/cpython/rev/7a6c0c4e6072New changeset873760b02024 by Berker Peksag in branch 'default':Issue#28131: Merge from 3.6https://hg.python.org/cpython/rev/873760b02024

Thanks for the report and for the reviews!