Bug #13155: Segfault testing Pry

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: ruby 2.4.1p111 (2017-03-22 revision 58053) [powerpc64le-linux]
• Backport: 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN

[vo.x (Vit Ondruch)]

Trying to fix Pry on Fedora Rawhide, I get random segfaults. Unfortunately, I can't see any pattern in them. The backtrace is always different. Attaching two logs in case it might make the issue more obvious. Here are some package versions which might be of interest:

$ rpm -q readline
readline-7.0-4.fc26.x86_64

$ rpm -q glibc
glibc-2.24.90-28.fc26.x86_64

[vo.x (Vit Ondruch)]

This appears to crash on Travis (Ubuntu) as well:

https://travis-ci.org/pry/pry/jobs/183781286

[sorah (Sorah Fukumori)]

• Status changed from Open to Feedback

I couldn't reproduce with 2.4.1 and trunk. Maybe it got a fix after 2.4.0 release.

[vo.x (Vit Ondruch)]

Will see as soon as 2.4.1 lands in Fedora. Will keep an eye on this.

[vo.x (Vit Ondruch)]

• File segv3 segv3 added
• File segv4 segv4 added
• ruby -v changed from ruby 2.4.0p0 (2016-12-24 revision 57164) [x86_64-linux] to ruby 2.4.1p111 (2017-03-22 revision 58053) [powerpc64le-linux]

This was triggered again during automatic rebuild of Fedora's Pry package

https://apps.fedoraproject.org/koschei/build/2977931

and it also failed in my subsequent build. See the attached logs. So 2.4.1 is still affected :/

[wanabe (_ wanabe)]

I guess it is the same issue as #13605 because of the log pattern.
(using pry, T_NONE, newobj_slowpath_wb_protected and gc_mark_children)

[wanabe (_ wanabe)]

• Related to Bug #13605: GC bug calling `ObjectSpace.each_object` added

[vo.x (Vit Ondruch)]

• Status changed from Feedback to Closed

This seems to be fixed by Ruby 2.4.2, because there was no build failure since this version is available in Fedora. So it was probably the same issue as #13605 pointed out by wanabe.