Hi all, In this discussion about mandating things, I've been wondering.... On 19-05-2024 9:11 a.m., Jonas Smedegaard wrote:
* mandate VCS-tracking
* Yes
* mandate the use of one specific VCS
* Yes: git
What do people think this should mean, a *should* or *must* in policy? That the package has a RC bug if the packaging isn't tracked in git? And if the packaging is in git but I forgot to push my latest changes to the documented git server (this happens regularly to me as I do most uploads with dgit)? RC? In all suites where the packaging version isn't pushed for? And how about NMU's? (I just checked a random package without git: aspell-am, last non-NMU upload was in 2013)?
If *must* and thus RC, can the issue be fixed by "NMU"? I.e. if the person that did the changes, (and has nice local commits) somehow isn't available, will the package (if not key) be auto-removed? Or can everybody fix the repo? What if you don't have write access to the existing repo? And then what if the uploader comes back and tries to push the real history? What if my harddisk with local changes crashes before I push (again, I forget that sometimes, but for me luckily dgit will mostly have the commits).
Or is this "just a bug", maybe even at level important, but no other consequences. Than I think this discussion is going to be moot. I don't think there's much forcing possible and I think most already agree that stuff *should* be in VCS, so this isn't going to change for those in favor. And does it really add enough value if those that are forced are just going to do "gbp import-dsc" for each upload to the archive on a ./debian only repository? Because that (or better) we could already automate (see also my PS).
I'm against mandating ("must"). I think there's a large majority (maybe even consensus) that believe you *should* have the packaging in VCS (and I agree with that). Most packages already are (hopefully maintained) in git (93% for testing) [1] on salsa (86%) [2], so I propose we stop the discussion. I think what pere did [3] changed more than this whole discussion: already 45 *orphaned* packages converted to git by 25 April, which means my numbers above are on the low side. The remaining 438 QA maintained (!!??) packages constitute close to 20% of the packages not in git.
PaulPS: I've always wondered if the dgit server shouldn't track history, even if uploads don't happen via it. A dgit clone could (should?) already provide available history, even if no upload happened via it yet.
[1] https://trends.debian.net/vcs_testing-stacked.png [2] https://trends.debian.net/vcs-hosting_testing-stacked.png
Attachment:
OpenPGP_signature.asc
Description: OpenPGP digital signature