Hi Fab, On Sun, 2026-11-15 at 20:12 +0100, Fab Stz wrote: > Since I chose to rename the package to html-build (to conform to upstream's > name), I created a new repo to start on fresh base: > > https://salsa.debian.org/bastif/html-build
html-build is a bit generic though, IMO whatwg-html-build was a better name. > You can still use the old one, which would permit you to see the recent > changes. > > By the way, this requires the version 0.40 of crate markup5ever_rcdom > (package rust-markup5ever-rcdom). But it has not been uploaded to the archive > yet. > That's because with v0.35 the autopkgtests failed (when mixed with html5ever > 0.40). > You can get the deb by using the repo which is on my debusine workspace [1]. > The snipped for apt sources is displayed there. So that means the crate needs to be updated before this package can be uploaded? [...] > Le samedi 14 novembre 2026 00:32:59 heure normale d’Europe centrale, vous > avez écrit : > > Hi Fab, > > > > - Why are the rust build-deps in d/control or'ed? It shouldn't need to > > be like that. > > I thought it might be easier to maintain. > > Jonas ([email protected]) which is not part of the rust team (but maintains numerous > rust based packages) and doesn't follow the approach of the team, doesn't > always use versioned rust dependencies for their packages. > > I don't have a strong opinion, so did as you asked. With this, any update of > html5ever (& related) crate will require to upload a new version of > html-build having updated relaxed dependencies. This is only required for breaking changes according to semver (e.g. 0.40 -> 0.41 or 1.0 -> 2.0), which is intentional so that when a crate is updated in the archive, the rdeps are reviewed. Either way works, but there is not really any point in ORing both. [...] > > > - Consider adding autopkgtests based on cargo-auto-test. You can see any > > rust-* package on sources.debian.org to see what debcargo generates and > > do something similar to test the package continues to build/compile. > > I discussed this shortly on #debian-rust. For binary crates, there are > usually no autopkgtests. Moreover upstream doesn't have any integration test > of the binary itself. It only has rust unit tests. These unit tests are > already invoked by dh-cargo when the package is built. > > I anyway created a librust-*-dev package which is however unlikely to be > needed by any other rust package. So this is actually somehow only useful to > run the autopkgtests. I was struggling at getting it work with > cargo-auto-test but this is now fine. Using it however requires one to keep > updating the package version if d/tests/control at every update. I only suggested adding autopkgtest because you had the librust-whatwg- html-build-dev package for the crate. If nothing uses it, it would probably be better to get rid of it and the autopkgtest altogether. It can always be reintroduced later if some software needs the library. [...] > > > - Dversionmangle in d/watch is not a valid key. It's Dversion-Mangle. > > Also consider using "auto" value which is the same thing. > > According to manpage for "debian-watch": > "Hyphens in key names are ignored. Matching-Pattern is equivalent to > Matchingpattern" > > So both keys are valid. TIL, thanks for pointing that out! [...] > > - The binary should be installed to usr/bin/whatwg-html-build, not > > usr/bin/html-build. > > Instead of renaming the binary to the name of the package (which is likely a > requirement from the Debian Policy, but I can't find it right now), I > preferred to rename the package from "whatwg-html-build" to "html-build" (in > line with the name in Cargo.toml). That way there is no need to patch the > scripts that call this binary, I find this ambiguousfor the same reasons as above for the package name. Patching is not necessary since you can easily create a fake PATH that symlinks html-build during package build, or take the cargo approach and include a fake bin dir in the package. -- Maytham
signature.asc
Description: This is a digitally signed message part

