Re: [PATCH v6 1/2] scripts: generate_rust_analyzer.py: add versioning infrastructure

[Tamir Duberstein]

On Thu, May 7, 2026 at 8:57 AM Jesung Yang <y.j3ms.n@xxxxxxxxx> wrote:
>
> On Tue, May 5, 2026 at 7:01 AM Tamir Duberstein <tamird@xxxxxxxxxx> wrote:
> > On Mon, 04 May 2026 22:20:59 +0900, Jesung Yang <y.j3ms.n@xxxxxxxxx> wrote:
> > > diff --git a/scripts/generate_rust_analyzer.py b/scripts/generate_rust_analyzer.py
> > > index d5f9a0ca742c..934698c7eb95 100755
> > > --- a/scripts/generate_rust_analyzer.py
> > > +++ b/scripts/generate_rust_analyzer.py
> > > @@ -343,6 +346,153 @@ def generate_crates(
> > > [ ... skip 46 lines ... ]
> > > +    #        git -C ./rust-analyzer describe --tags --abbrev=0 "$hash"
> > > +    #      )
> > > +    #
> > > +    #      link_prefix="https://github.com/rust-lang/rust-analyzer/releases/tag";
> > > +    #      echo "$link_prefix/$tag_predates"
> > > +    #      ```
> >
> > I think we talked about this being a bit more defensive with (at least)
> > set -eu. How's this?
> >
> >   #!/usr/bin/env bash
> >
> >   set -euo pipefail
> >
> >   RUST_VERSION=$1
> >
> >   grep_args=()
> >   while IFS= read -r subject; do
> >     grep_args+=(--grep "$subject")
> >   done < <(git -C rust log \
> >       --fixed-strings \
> >       --format='%s' \
> >       --grep 'Merge pull request #' \
> >       --merges \
> >       --no-follow \
> >       -n 10 \
> >       "$RUST_VERSION" \
> >       -- src/tools/rust-analyzer
> >   )
> >
> >   tag_predates=$(
> >     git -C rust-analyzer log \
> >       --fixed-strings \
> >       --format='%(describe:tags,abbrev=0)' \
> >       --merges \
> >       -n 1 \
> >       "${grep_args[@]}"
> >   )
> >
> >   link_prefix="https://github.com/rust-lang/rust-analyzer/releases/tag";
> >   echo "$link_prefix/$tag_predates"
> >
> > Note I added `--merges` in the first `git log`, inlined `LOOKAHEAD` (and
> > reduced it to 10, now that it's only merge commits), and replaced piping
> > into grep with `--grep`. The second and third `git log` commands are
> > combined.
>
> I don't see the `--no-follow` option in the man page, so I'd go for
> `-c log.follow=false`.

https://lore.kernel.org/git/20260507-document-log-no-follow-v1-1-46ce02490eba@xxxxxxxxx/

> Should we also use that for the second `git log`?
> Otherwise, merge commits might be lost, right?

The man page says:

  --follow
      Continue listing the history of a file beyond renames (works
only for a single file).

So the second `git log` is not affected because it doesn't pass paths.
I suppose it doesn't hurt, but can be misleading.


>
> > > [ ... skip 60 lines ... ]
> > > +        return ra_version_output
> > > +    except FileNotFoundError:
> > > +        return None
> > > +
> > > +def map_ra_version_baseline(ra_version_output: str) -> RaVersionInfo:
> > > +    baselines = reversed(RaVersionInfo)
> >
> > This is an iterator but it's possible for both `version_match` and
> > `date_match` to use it, which means the latter will observe an empty
> > iterator.
>
> Nice catch, I'll wrap it in `list()`.
>
> Best regards,
> Jesung