On 20.11.2016 00:54, Cyril Brulebois wrote: > A few things happened lately which made me wonder. > > > Looking at bootstrap-base, there's this commit: > https://anonscm.debian.org/cgit/d-i/base-installer.git/commit/?id=a8b68f8e5c268127f443c055a9efdf9740133115 > > which totally broke d-i, which got uploaded on 2016-11-13, and there's > still no released fix for it one whole week after that. > > I'm not thrilled about this kind of “push and forget” (git) or “upload > and forget” (in the archive). We need to do better. > > (About this specific issue I've just checked with Philip Hands, who > seems to be happy with testing etc., so an upload is forthcoming, but I > wanted not to miss the opportunity to point out this kind of issues.) Yeah, my apologies. I misread the patch and assumed wrongly that the variable always has to be set, when it unfortunately relied on the empty variable checking for the directory to be present. (Unnecessarily so, but not failing in the process.) So it looked completely straightforward to me when the original behavior had a subtle twist. I saw the followup thread between Tianon and Philip and thought that a proper patch had been identified. Otherwise I would've went, mea culpa'ed and uploaded a fix. I didn't realize that it still wasn't uploaded, sorry about that. :( I suppose one of the better workflows would be instead of just build testing the commit on Jenkins, to also localudeb it and do a smoke run of d-i to ensure that it's not totally broken. Philip, is that something that would be possible? (And of course briefly testing that myself rather than just build testing. There isn't really any excuse to make here.) Kind regards Philipp Kern
Attachment:
signature.asc
Description: OpenPGP digital signature