warnings: get filename from frame.f_code.co_filename

[takluyver]

BPO	33375
Nosy	@brettcannon, @vstinner, @takluyver
PRs	bpo-33375: Get filename for warnings from frame.f_code.co_filename #6622 bpo-33375: Fix GCC warning in Python/_warnings.c #7556

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2018-06-09.17:10:30.972>
created_at = <Date 2018-04-27.16:38:55.779>
labels = ['3.8', 'library']
title = 'warnings: get filename from frame.f_code.co_filename'
updated_at = <Date 2019-10-22.23:03:40.665>
user = 'https://github.com/takluyver'

bugs.python.org fields:

activity = <Date 2019-10-22.23:03:40.665>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2018-06-09.17:10:30.972>
closer = 'brett.cannon'
components = ['Library (Lib)']
creation = <Date 2018-04-27.16:38:55.779>
creator = 'takluyver'
dependencies = []
files = []
hgrepos = []
issue_num = 33375
keywords = ['patch']
message_count = 9.0
messages = ['315848', '316133', '316134', '316135', '319099', '319161', '319162', '320045', '355172']
nosy_count = 3.0
nosy_names = ['brett.cannon', 'vstinner', 'takluyver']
pr_nums = ['6622', '7556']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue33375'
versions = ['Python 3.8']

[takluyver]

The warnings module tries to find and show the line of code which is responsible for a warning, for the same reasons that tracebacks show a line of code from each stack frame. However, they work in quite different ways.

Native tracebacks, the traceback module and pdb all do something like this:

frame.f_code.co_filename

But warnings does something like this (paraphrasing C code in _warnings.c):

frame.f_globals.get('__file__', sys.argv[0])

This causes problems for interactive interpreters like IPython, because there are multiple pieces of entered code which have to share the same global namespace. E.g. ipython/ipython#11080

This was raised a long time ago in bpo-1692664. Back then, the answer was that co_filename could be wrong if the path of a pyc file changed. However, that issue was fixed in bpo-1180193. And it seems that the co_filename approach must largely work today, because tracebacks and pdb rely on it.

So I'm proposing to make warnings match how those other tools find filenames. I think this should also be a minor simplification of the code.

[takluyver]

Hi Brett! If you get a moment, any review of the linked PR would be welcome. :-)

#6622

[brettcannon]

I just added myself to review the PR. At worst I will get to it during the
sprints at PyCon US.

On Thu, 3 May 2018 at 12:52 Thomas Kluyver <report@bugs.python.org> wrote:

Thomas Kluyver <thomas@kluyver.me.uk> added the comment:

Hi Brett! If you get a moment, any review of the linked PR would be
welcome. :-)

#6622

----------

---

Python tracker <report@bugs.python.org>
<https://bugs.python.org/issue33375\>

---

[takluyver]

Thanks! No rush, I just thought I'd take the opportunity when you added yourself to the nosy list.

[brettcannon]

New changeset 11a8966 by Brett Cannon (Thomas Kluyver) in branch 'master':
bpo-33375: Get filename for warnings from frame.f_code.co_filename (GH-6622)
11a8966

[brettcannon]

New changeset 3f45f5d by Brett Cannon (Zackery Spytz) in branch 'master':
bpo-33375: Fix GCC warning in Python/_warnings.c (GH-7556)
3f45f5d

[brettcannon]

Thanks everyone for making this happen!

[vstinner]

This change introduced a regression: bpo-33912 "[EASY] test_warnings: test_exec_filename() fails when run with -Werror". I consider that it's an easy issue, please don't fix it, but explain how to fix it and let a newcomer to fix it!
https://mail.python.org/pipermail/python-committers/2017-June/004564.html

[vstinner]

I closed bpo-8787 as a duplicate of this issue.