Hi! On Wed, 2025-01-22 at 21:39:55 +0100, Aurelien Jarno wrote: > It's actually quite noticeable on fast machines for the rtpengine > package, and even with the latest dpkg mitigations. The build time on > fast machine increased on average by a factor ~30, for instance: > amd64: 5m => 2h39 > arm64: 5m => 3h4 > ppc64el: 4m => 1h45a
We have reworked the upstream build system and substantially reduced the build times with GNU make 4.4 (which should even improve the Debian package build times with GNU make <= 4.3), with at least the following commits: https://github.com/sipwise/rtpengine/commit/887fb40f3fa17bd476d4d42ec0de83752a2f006c https://github.com/sipwise/rtpengine/commit/2f53560f250eda9ba533401d45a88f9ee74da77a https://github.com/sipwise/rtpengine/commit/5cb1002949bbcefebf58b9853f339a9fb9739eaa https://github.com/sipwise/rtpengine/commit/bb2bf3a7026e7f7912dffa3ae8ea0e1ef3f29f31 https://github.com/sipwise/rtpengine/commit/7dbe24e4e5fae61b23e677047b28e82ac49c3caf We are waiting a bit for these to settle, and probably next Monday, we'll backport to the rtpengine mr12.5 LTS branch, and then get that uploaded to Debian. The other thing we noticed is that this only triggers when you are pre-setting the variables in the environment before calling make. I'm though wondering if there are any other similar cases of massive build time increases in the archive, although I guess that might not be currently visible until they next get built? Perhaps it would be nice to do a mass rebuild with GNU make 4.3 and 4.4 and compare times? (But I think this was already done before the upload happened, so maybe just comparing the logged times would be enough.) Although this all still seems like a regression somewhere in GNU make 4.4, which I'm surprised no one else has reported or complained about in bug-make (AFAICS). I think I'll be sending a mail there at least to make them aware. Thanks, Guillem
