[PATCH 0/3] some transport-helper "option object-format" confusion

From: Jeff King <peff@peff.net>
Date: Mar 20, 2024, 09:32 UTC
Message-ID: <20240320093226.GA2445531@coredump.intra.peff.net>
In-Reply-To: <20240307084735.GA2072130@coredump.intra.peff.net>

On Thu, Mar 07, 2024 at 03:47:35AM -0500, Jeff King wrote:

Show 12 quoted lines

> I happened to be looking at the output of t5801 for an unrelated
> problem, and I noticed our git-remote-testgit spewing a bunch of shell
> errors. It turns out that its expectations do not quite match what the
> transport-helper code produces.
> 
> This series brings the test and documentation in line with how the
> transport-helper code behaves. But I'm not sure if we should be going
> the other way (see the comments on patch 2 especially), and bringing the
> transport-helper code in line with the others. Hence the RFC.
> 
>   [1/2]: t5801: fix object-format handling in git-remote-testgit
>   [2/2]: doc/gitremote-helpers: match object-format option docs to code

Here's a non-RFC v2 based on the discussion thus far (thanks brian and Eric).

The big change is that instead of changing the docs to match true-less "option object-format", the code is changed to match the docs. That happens in patch 3 (which subsumes the original patch 1). We continue to drop the documentation for the "option object-format sha256" form. But now the commit message justifies it better, and we clean up the stale code in remote-curl.c.

Patch 1 is a small fix for debugging output that I noticed after getting confused. :-/ It's not strictly related and could be taken separately.

Eric mentioned having Git check that the helpers never say ":object-format" unless it was negotiated. I stopped short of that. One, it's a bit tricky to test (since Git will always ask for object-format, you have to teach remote-testgit to optionally send broken output). And two, I'm not sure that being strict has much value here. It keeps remote helpers honest, but the real losers are old versions that do not understand :object-format, which would fail against such a remote. So I dunno. It isn't any harder to do it on top later if we want to.

  [1/3]: transport-helper: use write helpers more consistently
  [2/3]: transport-helper: drop "object-format <algo>" option
  [3/3]: transport-helper: send "true" value for object-format option

 Documentation/gitremote-helpers.txt |  7 ++-----
 remote-curl.c                       |  9 ++-------
 t/t5801/git-remote-testgit          |  4 +++-
 transport-helper.c                  | 11 ++++-------
 4 files changed, 11 insertions(+), 20 deletions(-)

-Peff