[PATCH 0/11] Updated 'quickfetch' series

[PATCH 0/11] Updated 'quickfetch' series

[Shawn O. Pearce]

So what started out as a 1 patch hack has now grown into an 11
patch series.  *sigh*

  1)  Fix memory leak in traverse_commit_list

This one seems obvious to me.  No real harm to not having it but
later on we'd see the downside to the leak in git-fetch.

  2)  Reorganize the object memory pools so we can release them
  3)  Define free_all_objects to deallocate object pools
  4)  Allow pooled nodes to be recycled back onto a free list
  5)  Bulk allocate struct commit_list, as we do for struct commit

This part of the series is only lightly tested thus far.  It rewrites
the object allocators in alloc.c to allow complete destruction of
the object hash and its contents in a single call.  This way we
can rerun the revision machinary a second time in the same process
without the mess of clear_commit_flags() or worring about parents
having been rewritten.

We also decrease memory usage by moving struct commit_list into the
pool of things managed by alloc.c.  Given how many of these suckers
we are allocating (3x as many as commits in a --topo-order!) the
malloc overhead saved per parent pointer is probably worth the pain
of reviewing this series.

  6)  git-fetch: Release objects used by a prior transport

Makes use of the above to ensure transports that are called directly
in-process (e.g. fetch-pack) don't get confused by a prior invocation
of that same transport.  I don't think fetch-pack was meant to be
run twice in the same process...

  7)  git-fetch: Limit automated tag following to only fetched objects

Rewrites the rules behind automated tag following.  The new rule is
simpler to understand, has some measure of safety attached to it, and
actually works when we turn on quickfetch behavior in a transport.

  8)  run-command: Support sending stderr to /dev/null
  9)  rev-list: Introduce --quiet to avoid /dev/null redirects
 10)  git-fetch: avoid local fetching from alternate (again)
 11)  git-fetch: test avoiding unnecessary copying from alternates

Turn on quickfetch for the native git transport.  As I'm sitting
here writing this I'm wondering if the quickfetch thing shouldn't
be higher up, like in git-fetch itself, so that we can also use
it in the HTTP transport.  But it should work now with the earlier
stuff out of the way.

-- 
Shawn.