Andrew Lee <[email protected]> writes: > Hi team, > > I just quickly checked and found that the kms release was 2 days ago and > can easily see the tag on master branch. > > For not been tagged upstream release, we usually just use git HEAD in > debian/watch file. We may simply use this way to package a git version > included the tagged submodule release instead of creating fragmentation.
Are you saying we should make golang-google-cloud track git HEAD? The ML_DSA_44 stuff was pushed back in February 2026, it is not that recent. For some reason I was looking at the https://code.googlesource.com/gocloud main branch that stopped being updated in 2024, which explains why I didn't see it in main. I do see it in the https://github.com/googleapis/google-cloud-go.git main branch now that I look at the right place. So indeed I think tracking git HEAD would solve my problem. The v0.123.0 git tag is from September 2025 so I think we should stop tracking the vX.Y.Z tags. I've asked upstream if they have any recommendation for us: https://github.com/googleapis/google-cloud-go/issues/20885 Bumping this will require an epoch, any objections to doing that? Maybe we could upload to experimental first -- I suspect there will be a substantial work to make HEAD build. /Simon > Or we better discuss a new method for module-aware packaging. I have an > idea in mind, but need to make a proof of concept work. Go compiler is > query modules via goproxy by default. If we can use go compiler to query > and fetch the zip file from goproxy with uscan. It will be much easier for > tracking and fetch the source of a module. And might be possible to do some > automation for packaging with salsa-ci. > > Best regards, > > -Andrew > > Daniel Baumann <[email protected]>於 2026年11月11日週三,22:21寫道: > >> On 11/11/26 13:37, Simon Josefsson wrote: >> > There is a middle-ground: just break off some packages, but not all. >> >> yes, "and"... >> > migrate all consumers of golang-google-cloud kms/ to also add the new >> > golang-google-cloud-kms-dev package >> >> that sounds like a lot of work from a generic point of view, and becomes >> a hassle when you have to undo it and merge it back once it's not >> feasible anymore because the two diverged too much. >> >> this multiplies with each submodule you're breaking out. >> >> personally, I'd rather spend the time do breaking them off completely >> and be done with it. ymmv. >> >> > As a temporary solution until we establish some consensus, I would still >> > consider that vendoring the newer kms/ files may be the simplest fix. >> >> I'd consider vendoring to be even worse than the "middle-ground", >> updating and tracking individual copies, even if it's only temporary >> time intended (even disregarding that most of the time it later becomes >> more permant for $reasons), is horrible and always results in much more >> work, stuff that gets forgotten, etc. that (often) other people have to >> cleanup years later. >> >> Regards, >> Daniel >> >>
signature.asc
Description: PGP signature
