[PATCH v2 00/17] document & test fetch pruning & add fetch.pruneTags

As noted in my 87h8quytmq.fsf@evledraar.gmail.com there was a bug I noticed in v3 where it would segfault on some git-fetch invocations, but there were not tests anywhere that caught that.

So in addition to fixing that issue, this fleshens out the testing being set up as part of this series so we'll test those sorts of invocations. It would segfault on some `git fetch <url>`, not `git fetch <name>`.

Ævar Arnfjörð Bjarmason (17):
  fetch: don't redundantly NULL something calloc() gave us

Rephrased commit message.

  fetch: trivially refactor assignment to ref_nr

New, makes a subsequent change smaller.

  fetch: stop accessing "remote" variable indirectly

Typo fix in commit message noted by Junio.

  remote: add a macro for "refs/tags/*:refs/tags/*"

New, makes a subsequent change smaller.

  fetch tests: refactor in preparation for testing tag pruning
  fetch tests: re-arrange arguments for future readability
  fetch tests: add a tag to be deleted to the pruning tests

No changes.

  fetch tests: test --prune and refspec interaction

Changed +refs/tags/*:refs/tags/ to refs/tags/*:refs/tags/. No functional difference, since git doesn't care. Just to be consistent with the macro added earlier & doing the same in commit messages & tests later in the series.

  fetch tests: double quote a variable for interpolation

Now back from an earlier version, needed for a later change.

  fetch tests: expand case/esac for later change

New, makes the next patch smaller / easier to review.

  fetch tests: fetch <url> <spec> as well as fetch [<remote>]

For all `git fetch <name>` we now run another version of the test where we test an equivalent `git fetch <url>`. This sort of exhaustive testing was missing in our whole test suite, and would have caught the segfault in v3.

  git fetch doc: add a new section to explain the ins & outs of pruning
  git remote doc: correct dangerous lies about what prune does
  git-fetch & config doc: link to the new PRUNING section

No changes except omitting the "+" in front of refs/tags/[...] as noted above.

  fetch tests: add scaffolding for the new fetch.pruneTags

Ditto "+" change + minor changes carried over from previous patches.

  fetch: add a --fetch-prune option and fetch.pruneTags config

The bug in v3 was that the remote->fetch variable needs to chaned in lockstep with remote->fetch_refspec, but only the latter was changed. Codepaths that fetched by URL would under --prune-tags expect as many items in both, and segfault on the access to remote->fetch.

As explained in the amended commit message the API is not amenable to ALLOC_GROW, so there's now a add_prune_tags_to_fetch_refspec() function in remote.c which adds the new element to remote->fetch via xrealloc() + memcpy().

Careful review of that most welcome.

There's lots more tests that catch the case where it segfaulted.

  fetch: make the --fetch-prune work with <url>

The previous patch was changed to document that this wouldn't work:

    git fetch <url of origin> --prune --prune-tag

This makes it work, at the cost of some complexity in fetch_one(). I think it makes sense to keep this, I just wanted to split it off from the previous patch to clearly show the hoops we need to jump through for that one case.

 Documentation/config.txt               |  20 ++-
 Documentation/fetch-options.txt        |  17 ++-
 Documentation/git-fetch.txt            |  87 ++++++++++++
 Documentation/git-remote.txt           |  14 +-
 builtin/fetch.c                        |  54 ++++++--
 contrib/completion/git-completion.bash |   2 +-
 remote.c                               |  15 ++
 remote.h                               |   5 +
 t/t5510-fetch.sh                       | 242 +++++++++++++++++++++++++++------
 9 files changed, 395 insertions(+), 61 deletions(-)