What the summary says, python version used was 3.9 and there was nothing interesting to see in the logs, which was: [core:notice] [pid 28230:tid 139632084309376] AH00052: child pid 28432 exit signal Segmentation fault (11) I tried to go back to 4.7.1, but didn't complete the test as I needed mercurial in python 3.8 as well; time constraints pushed me to add the ~amd64 keyword to 4.9.0, where things work just fine. I also asked in #gentoo if anybody was experiencing this, but I suspect that I might be one of a select group who run the mercurial server in Gentoo, so got no answer. I am reporting this for the record and in case somebody else bumps into the same issue. I doubt I'll be able to run any more tests in this case. Perhaps it's time to stabilise 4.9.0?
Created attachment 761190 [details] Output of `emerge --info` Last thoughts: I did forget in this instance to try compiling with -O2, perhaps this is worth trying for anybody else who bumps into this.
Now that the need for a mod_wsgi that supports has been addressed with 4.9.0, it might make more sense to just remove 4.7.1-r1.
Supports 3.10, that is. It is too early to stabilize 4.9.0, since that only entered the tree recently.
Created attachment 761386 [details] backtrace Looks like it goes beyond mercurial. Getting the same crashes with a "hello world" script, and seeing them fixed on 4.9
Created attachment 761387 [details] emerge --info to accompany the backtrace -O2 didn't help in my case
hgweb is working for me, but all my wep apps using dev-python/bottle-0.12.19 (on a local apache server) started segfaulting since ww-apache/mod_wsgi-4.7.1-r1 was rebuilt on 2022-01-03. They had been fine since June last year with this version of mod_wsgi. It seems to be 10 of the mpm_prefork threads segfault at the same moment. Everything is working again with www-apache/mod_wsgi-4.9.0
So I checked, and turns out I'm using hgweb.cgi rather than hgweb.wsgi.
Confirm from my side, this segfault present with python 3.9, apache 2.4.52 and mod_wsgi 4.7.1-r1. It ALWAYS segfault even on hello-world python script. With mod_wsgi 4.9.0 no crash with hello world and with production code on servers.
(In reply to Dirkjan Ochtman from comment #3) > Supports 3.10, that is. > > It is too early to stabilize 4.9.0, since that only entered the tree > recently. If 4.9.0 fixes a lot of regressions, it's fine to stable it, and better than leaving stable broken if that's the status quo (but not aware of the background here).
For reference, we have 2 stable versions: - One works but requires python 3.8 - The -r1 one that's broken as per this bug The unstable version fixes everything. And because I'm not a Gentoo developer/security/staff/etc., I can say that I would stabilise, but it's up to you folks. Rules are rules after all :-)
I've reverted the buggy changes in 4.7.1-r1 and requested stabilization for 4.9.0 in bug 831061.
