On Wed, Jul 15, 2026 at 04:15:27PM +0200, Santiago Vila wrote: > [ replying to the lists only, please do not cc me in this thread ] >... > Additionally, the revert of geoip-database to an older version is > probably the reason why the tests of lua-geoip now fail, >...
Thanks for mentioning, I broke that and I've just submitted a pu request to fix it. > the new > python triggered a weird bug in sphinx which made between 7 and 10 > packages to FTBFS, and several other packages which used to build ok > in trixie do not build ok anymore. > > I share your concern. We have been unable to ship a stable release > without FTBFS bugs in a very long time, and I believe people in > general is not aware of that. > > > @Santiago: > > How long would a rebuild of all packages in trixie (perhaps with > > exclusions for a few large packages) take? > > [...] > > An archive rebuild takes between 24h and 48h, so your estimate is > correct. > > Unfortunately, there are FTBFS bugs in trixie which I reported shortly > after 13.0 (or even *before*) which are still unfixed as of today, so > I do not perceive true willingness from the project as a whole to ship > point releases without FTBFS bugs (and of course I can't blame anybody > for that because we are all volunteers). > > Sorry, but I can't do archive rebuilds all the time. Understood (and I do really appreciate your work). > I have enough > work doing them after each point release with the hope that maybe the > bugs found could be fixed for the next point release, but we are not > even able to fix those, so I don't think any amount of archive > rebuilds will fix the underlying problem. > > Maybe what we need is a cultural change of some kind. > > (In either case, raising awareness of the problem is indeed a very > good first step). If a list of regressions exists before the release, the SRMs could exclude packages based on that from the point release. The part of my suggestion that would not create additional work for you would be if you did Archive rebuild including all pu updates on Monday R-5 which should be basically the same packages as your rebuild after the point release, but would highlight cases like mariadb and python. Same work for you, but would have prevented shipping a broken python in the point release. It is necessary to get a comment from the SRMs regarding such suggestions, since they are the ones who'd have to integrate it into their release process. > Thanks. cu Adrian
