STINNER Victor added the comment:
However I think the fd tests on windows is just fine to be skipped. So what's
the next plan? Shall we continue to work on it or just resolve this issue?
issue23566_fd_tests_v2.patch makes test_faulthandler.py a little more complex.
It's maybe better to just
Wei Wu added the comment:
@haypo, would you review issue23566_fd_tests_v2.patch? It's been a time since
the last update of it.
However I think the fd tests on windows is just fine to be skipped. So what's
the next plan? Shall we continue to work on it or just resolve this issue?
--
Wei Wu added the comment:
The updated patch refactored test code a little bit according to the latest
review comments by Victor.
--
Added file: http://bugs.python.org/file38546/issue23566_fd_tests_v2.patch
___
Python tracker rep...@bugs.python.org
STINNER Victor added the comment:
I reviewed issue23566_fd_tests.patch .
--
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue23566
___
___
Wei Wu added the comment:
I attached a patch that implements the solution described above.
--
Added file: http://bugs.python.org/file38480/issue23566_fd_tests.patch
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue23566
Wei Wu added the comment:
The last approach I proposed requires some change in template code of
check_xxx methods. To make it better, we can add a bool parameter to the
check_xxx functions, True value indicating a fd test. If a filename is given at
the same time, then a fd can get from that
STINNER Victor added the comment:
I commited a fix to repair Windows buildbots.
Or we could reuse the file created by filename in subprocess?
I tried to pass a file descriptor from the parent to the child
process, but this option is complex. It's possible to pass a handle
with
Roundup Robot added the comment:
New changeset 211e29335e72 by Victor Stinner in branch 'default':
Issue #23566: Skip fd tests of test_faulthandler on Windows
https://hg.python.org/cpython/rev/211e29335e72
--
___
Python tracker rep...@bugs.python.org
Changes by Christoph Sieghart s...@0x2a.at:
--
nosy: -sigi
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue23566
___
___
Python-bugs-list mailing
Wei Wu added the comment:
Updated the patch and addressed the previous review comments:
* Fixed the hasattr problem
* Added a default value None for filename in check_dump_traceback
* Reverted unnecessary code change in check_dump_traceback_later
* Added a new paragraph to the enable,
STINNER Victor added the comment:
Let me know if there is possible further improvement in this patch.
I was going to push your change, but I noticed that you use stderr.fileno() in
tests. I modified the test to use a different file descriptor. sys.stderr is
also the default value, so the
STINNER Victor added the comment:
Oops, I forgot Windows. On Windows, we should maybe use stderr.fileno() and
avoid pass_fds.
(Or maybe just skip the tests on Windows?)
http://buildbot.python.org/all/builders/AMD64%20Windows7%20SP1%203.x/builds/5828/steps/test/logs/stdio
Roundup Robot added the comment:
New changeset e78de5b1e3ef by Victor Stinner in branch 'default':
Issue #23566: enable(), register(), dump_traceback() and dump_traceback_later()
https://hg.python.org/cpython/rev/e78de5b1e3ef
--
nosy: +python-dev
___
Wei Wu added the comment:
Or we could reuse the file created by filename in subprocess?
if filename:
file = open(filename, wb)
if use_fd:
file = file.fileno()
else:
file = None
In this case, we need to pass two arguments(both filename and a bool use_fd) to
check_xxx
STINNER Victor added the comment:
issue23566_update.patch looks good to me, but I suggested some minor changes.
Usually, I do such changes myself, but I proposed this issue on the Python
menthorship list, so I prefer to do the changes to learn the process of
reviewing patches and taking
Wei Wu added the comment:
Attached a patch to this issue. Made some changes to faulthandler_get_fileno to
accept integer as fd. If a fd is provided, the file member of fatal_error
struct is set NULL.
An new test case is added and some common testing code is also changed in order
to be
STINNER Victor added the comment:
@kilowu: Thanks, I reviewed your patch.
--
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue23566
___
___
Wei Wu added the comment:
@haypo: Thank you for your review. I attached an updated patch addressing the
review comments. In addition, I also added a change note in Misc/NEWS.
--
Added file: http://bugs.python.org/file38436/issue23566_update.patch
Changes by Christoph Sieghart s...@0x2a.at:
--
nosy: +sigi
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue23566
___
___
Python-bugs-list mailing
New submission from STINNER Victor:
Currently, functions of the faulthandler module require a file-like object
(with a fileno() method). It would be nice to accept also file descriptors
(int).
Example of code using faulthandler which creates an useless file object just to
pass a file
STINNER Victor added the comment:
Only faulthandler_get_fileno() should be modified, but a new unit test should
be added.
--
___
Python tracker rep...@bugs.python.org
http://bugs.python.org/issue23566
___
21 matches
Mail list logo
