On Tue, Oct 29, 2024 at 01:50:39AM -0400, Jeff King wrote:
> On Mon, Oct 28, 2024 at 02:43:42PM +0100, Karthik Nayak wrote:
>
> > diff --git a/builtin/fast-import.c b/builtin/fast-import.c
> > index ffee7d3abd..f4892d7f37 100644
> > --- a/builtin/fast-import.c
> > +++ b/builtin/fast-import.c
> > @@ -806,7 +806,7 @@ static char *keep_pack(const char *curr_index_name)
> > struct strbuf name = STRBUF_INIT;
> > int keep_fd;
> >
> > - odb_pack_name(&name, pack_data->hash, "keep");
> > + odb_pack_name(the_repository, &name, pack_data->hash, "keep");
>
> Why not pack_data->repo here? It's always going to be set to
> the_repository in this program, but I think minimizing the number of
> references to it still has value.
Yeah, I had pointed out a similar thing when I looked at this patch in
the message above yours in this thread.
I think we reached the same conclusion that this isn't strictly
incorrect, because in all of the instances that I looked at, p->repo is
equal to the_repository, so from an external behavior perspective, the
two are equivalent choices.
But I agree that the point is to *use* p->repo and not rely directly on
'the_repository', so that your suggestion here is a good one.
> Earlier I mentioned that another helper could simplify many of these
> sites a little. What I meant was this (on top of what's in your series):
>
> diff --git a/builtin/fast-import.c b/builtin/fast-import.c
> index 9056447bd0..976cb1d77b 100644
> --- a/builtin/fast-import.c
> +++ b/builtin/fast-import.c
> @@ -806,19 +806,19 @@ static char *keep_pack(const char *curr_index_name)
> struct strbuf name = STRBUF_INIT;
> int keep_fd;
>
> - odb_pack_name(the_repository, &name, pack_data->hash, "keep");
> + pack_hashfile(pack_data, &name, "keep");
> keep_fd = odb_pack_keep(name.buf);
> if (keep_fd < 0)
> die_errno("cannot create keep file");
> write_or_die(keep_fd, keep_msg, strlen(keep_msg));
> if (close(keep_fd))
> die_errno("failed to write keep file");
>
> - odb_pack_name(the_repository, &name, pack_data->hash, "pack");
> + pack_hashfile(pack_data, &name, "pack");
> if (finalize_object_file(pack_data->pack_name, name.buf))
> die("cannot store pack file");
>
> - odb_pack_name(the_repository, &name, pack_data->hash, "idx");
> + pack_hashfile(pack_data, &name, "idx");
> if (finalize_object_file(curr_index_name, name.buf))
> die("cannot store index file");
> free((void *)curr_index_name);
> @@ -832,7 +832,7 @@ static void unkeep_all_packs(void)
>
> for (k = 0; k < pack_id; k++) {
> struct packed_git *p = all_packs[k];
> - odb_pack_name(p->repo, &name, p->hash, "keep");
> + pack_hashfile(p, &name, "keep");
> unlink_or_warn(name.buf);
> }
> strbuf_release(&name);
> diff --git a/builtin/pack-redundant.c b/builtin/pack-redundant.c
> index 7d6c47ffd9..d3b5e7e112 100644
> --- a/builtin/pack-redundant.c
> +++ b/builtin/pack-redundant.c
> @@ -690,7 +690,7 @@ int cmd_pack_redundant(int argc, const char **argv, const char *prefix UNUSED, s
> pl = red = pack_list_difference(local_packs, min);
> while (pl) {
> printf("%s\n%s\n",
> - odb_pack_name(repo, &idx_name, pl->pack->hash, "idx"),
> + pack_hashfile(pl->pack, &idx_name, "idx"),
> pl->pack->pack_name);
> pl = pl->next;
> G }
> diff --git a/packfile.c b/packfile.c
> index cfbfcdc2b8..d81a62eb84 100644
> --- a/packfile.c
> +++ b/packfile.c
> @@ -46,6 +46,11 @@ char *odb_pack_name(struct repository *repo, struct strbuf *buf,
> return buf->buf;
> }
>
> +char *pack_hashfile(struct packed_git *p, struct strbuf *out, const char *ext)
> +{
> + return odb_pack_name(p->repo, out, p->hash, ext);
> +}
> +
> static unsigned int pack_used_ctr;
> static unsigned int pack_mmap_calls;
> static unsigned int peak_pack_open_windows;
> diff --git a/packfile.h b/packfile.h
> index 3409aef35d..43c19d7bba 100644
> --- a/packfile.h
> +++ b/packfile.h
> @@ -32,6 +32,9 @@ struct pack_entry {
> char *odb_pack_name(struct repository *repo, struct strbuf *buf,
> const unsigned char *hash, const char *ext);
>
> +/* Like odb_pack_name(), but pull repo and hash from existing packed_git. */
> +char *pack_hashfile(struct packed_git *p, struct strbuf *out, const char *ext);
> +
> /*
> * Return the basename of the packfile, omitting any containing directory
> * (e.g., "pack-1234abcd[...].pack").
>
>
> While coming up with the name, though, I had some second thoughts. The
> interface implies that its the way you should derive a pack-related
> filename from a packed_git. But it really is mis-designed for that
> purpose! The packed_git struct has "foo.pack" or similar in its
> pack_name field, and the correct way to derive the .idx, .bitmap, .keep,
> etc, is by string substitution. While we do tend to name packs
> pack-$hash.pack, most of the code will happily work on
> "some-arbitrary-name.pack". And that's why we have so few
> odb_pack_name() calls in the first place.
>
> IMHO the ones in fast-import should probably be doing that suffix
> replacement instead (and probably we should have a decent helper to
> facilitate that; you can grep for strip_suffix.*pack to see places that
> could potentially use it).
>
> All that said, I don't think it's worth derailing your series to deal
> with that cleanup. That can come later if we want. And if we do that,
> then the pack_hashfile() I suggested above would have no callers,
> because it's the wrong approach.
Heh. I feel like you and I just discussed this on the list together a
couple of days ago. Indeed, there are quite a few that would benefit
from such a cleanup (there are even more if you search for
'strip_suffix.*idx', which would work similarly).
> I do think it's probably worth changing your series to use the
> packed_git repo pointers we already have available, though (i.e., the
> cases I pointed out inline above).
But yeah, we can take that up as a secondary step on top of this series
if we wend up wanting to do that in the future.
Thanks,
Taylor
