homepage

pep 0523 renames PyEval_EvalFrameEx to _PyEval_EvalFrameDefault while the gdb python extension only looks for PyEval_EvalFrameEx to understand if it is dealing with a frame.
Final effect is that attaching gdb to a python3.6 process doesnt resolve python objects. Eg. py-list and py-bt dont work properly

New changeset 2e0f4db114424a00354eab889ba8f7334a2ab8f0 by Łukasz Langa (Bruno "Polaco" Penteado) in branch 'master':
bpo-30983: eval frame rename in pep 0523 broke gdb's python extension (#2803)
https://github.com/python/cpython/commit/2e0f4db114424a00354eab889ba8f7334a2ab8f0

New changeset 09b77165e3fffa7b7ff160ad06042cdcfa004bf5 by Łukasz Langa in branch '3.6':
[3.6] bpo-30983: eval frame rename in pep 0523 broke gdb's python extension (GH-2803) (#3090)
https://github.com/python/cpython/commit/09b77165e3fffa7b7ff160ad06042cdcfa004bf5

This change broke test_gdb on most buildbots :-(
http://buildbot.python.org/all/builders/x86%20Ubuntu%20Shared%203.x/builds/1159/steps/test/logs/stdio
======================================================================
FAIL: test_pyup_command (test.test_gdb.StackNavigationTests)
Verify that the "py-up" command works
----------------------------------------------------------------------
Traceback (most recent call last):
 File "/srv/buildbot/buildarea/3.x.bolen-ubuntu/build/Lib/test/test_gdb.py", line 688, in test_pyup_command
 $''')
 File "/srv/buildbot/buildarea/3.x.bolen-ubuntu/build/Lib/test/test_gdb.py", line 259, in assertMultilineMatches
 self.fail(msg='%r did not match %r' % (actual, pattern))
AssertionError: 'Breakpoint 1 (builtin_id) pending.\n[Thread debugging using libthread_db enabled]\nUsing host libthread_db library "/lib/i386-linux-gnu/libthread_db.so.1".\n\nBreakpoint 1, builtin_id (self=self@entry=, v=42) at Python/bltinmodule.c:1103\n1103\t{\n#4 Frame 0x4086b4cc, for file /srv/buildbot/buildarea/3.x.bolen-ubuntu/build/Lib/test/gdb_sample.py, line 10, in baz (args=(1, 2, 3))\n id(42)\n#5 Frame 0x4086b4cc, for file /srv/buildbot/buildarea/3.x.bolen-ubuntu/build/Lib/test/gdb_sample.py, line 10, in baz (args=(1, 2, 3))\n id(42)\n' did not match '^.*\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 7, in bar \\(a=1, b=2, c=3\\)\n baz\\(a, b, c\\)\n$'

I chose to revert the change because I don't have the bandwidth right now to investigate why the change broke test_gdb.
I'm surprised that a change affecting python-gdb.py wasn't properly tested manually using test_gdb.py :-( I understand that Travis CI doesn't have gdb and/or that the test pass in some cases?
The revert only gives us more time to design the proper solution.

New changeset 7cc33998b83df6ab8cf2c1afe16a61ee232e39ec by Victor Stinner in branch 'master':
bpo-30983: Revert changes which broke most buildbots (#3100)
https://github.com/python/cpython/commit/7cc33998b83df6ab8cf2c1afe16a61ee232e39ec

New changeset 98e26979aa4c15b391fd3780340772a17d513d4d by Victor Stinner in branch '3.6':
Revert "[3.6] bpo-30983: eval frame rename in pep 0523 broke gdb's python extension (GH-2803) (#3090)" (#3101)
https://github.com/python/cpython/commit/98e26979aa4c15b391fd3780340772a17d513d4d

In this case it would be great if the buildbots commented on the relevant pull request that it broke the build. Otherwise this creates this dissonance that it was all green during merging of the PR when in fact it wasn't.
I think we need to make test_gdb run on Travis, this is an unfortunate omission.

While our CI is getting better everyday, it still has a few flaws.
There is no automated buildbot failure notification to the author and
committer yet because random failures were very common 6 months ago. I
fixed something like 90% of the most common race conditions (in tests but
also in the code), and now the remaining 10% are the least funny ones :-)
My latest report:
https://haypo.github.io/python-buildbots-2017q2.html
In the meanwhile I check manually buildbots output and report bugs and
regressions.
For gdb, don't worry, the buildbots were fixed quickly, it's fine if the
pre-commit CI missed the issue since gdb changes are rare. I am confident
that we can design a fix without introducing regressions ;-)

I am unable to repro this locally. With the changes to libpython.py test_gdb still passes on CentOS 7 and macOS 10.12.

OK, I got the repro. You have to rebuild python for /Tools/gdb/libpython.py to land in /python-gdb.py. The latter is used during test_gdb.py.

> You have to rebuild python for /Tools/gdb/libpython.py to land in
/python-gdb.py. The latter is used during test_gdb.py.
Ah yes, sorry, that's not obvious. "make" does the copy for you.

The issue originally stems from the fact that a non-debug Python build with --enable-shared is inlining PyEval_EvalFrameEx into _PyEval_EvalFrameDefault.
So, the patch should really become to *replace* discovering PyEval_EvalFrameEx with the discovery of _PyEval_EvalFrameDefault. This might be problematic in the future when there is an actual JIT but currently it's strictly better than the current situation where the gdb bindings simply don't work for shared builds.

New changeset 5fe59f8e3a0a56a155c18f9d581205ec533764b6 by Łukasz Langa in branch 'master':
bpo-30983: [gdb] Fix py-bt, etc. for non-debug shared builds (#3153)
https://github.com/python/cpython/commit/5fe59f8e3a0a56a155c18f9d581205ec533764b6

New changeset 8e572491b59c2334723bfd7411ab2a9fbd100f70 by Łukasz Langa in branch '3.6':
[3.6] bpo-30983: [gdb] Fix py-bt, etc. for non-debug shared builds (GH-3153) (#3192)
https://github.com/python/cpython/commit/8e572491b59c2334723bfd7411ab2a9fbd100f70