git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: git-diff in a worktree is an order of magnitude slower?

From
Jeff King <peff@peff.net>
Date
Jun 21, 2026, 17:24 UTC
Message-ID
<20260621172432.GA2206349@coredump.intra.peff.net>
In-Reply-To
<xmqqa4sog1e9.fsf@gitster.g>
On Sat, Jun 20, 2026 at 05:53:02PM -0700, Junio C Hamano wrote:
Show 10 quoted lines
> > which dates to aecbf914c4 (git-diff: resurrect the traditional empty
> > "diff --git" behaviour, 2007-08-31). On my system that comparison is
> > false because the double-negation produces 1
> > (diff_auto_refresh_index=1 or the result of git_config_bool). 
> 
> Not quite.  It was false because double-negation initializes the
> member to 1, which causes a call to diffcore_skip_stat_unmatch()
> be made, *and* the diffcore_skip_stat_unmatch() function did not
> find any ghost changes, i.e., paths that were only stat-dirty hence
> needed a call to refresh_index_quietly().

I think this is the core of the issue. These entries are "racy git dirty" in the sense that their mtimes are the same as the index mtime, and so we double-check the contents. This is the first bullet point under the "Racy Git" section of Documentation/technical/racy-git.adoc.

But diffcore_skip_stat_unmatch() doesn't count them as dirty, so we don't increment the counter, and thus top-level git-diff won't write out the new index. And thus every subsequent diff repeats the same expensive double-check.

But I'm not sure where the blame lies. Either:
  1. diffcore_skip_stat_unmatch() should be counting these in its
     "dirty" counter; or
  2. the index should be marking these differently. The second bullet
     point of that Racy Git section says:
       When the index file is updated that contains racily clean
       entries, cached `st_size` information is truncated to zero
       before writing a new version of the index file.
     Should the index be written out with a 0 size field here, so that
     we know they are dirty and should be updated? I guess that would be
     user-visible, though, because commands that _don't_ update the
     index (like plumbing diff-files) would generate a spurious diff
     there rather than doing the content-level comparison.

I dunno. You had solved most of the racy git stuff before I came along, so I never gave it too much thought (and what little thought I did was many years ago).

Show 11 quoted lines
> > So… has that conditional been quietly dead all this time? I can't
> > imagine that's right, but…
> 
> I initially thought it was an embarrassing thinko, but after seeing
> how .skip_stat_unmatch is used as a 1-based counter (i.e., if the
> member says 42, it means it saw 41 paths that were stat-dirty but
> without actual content change), I do not think so.
> 
> Now, it is a different matter if such a "dual" purpose "more than a
> simple boolean" counter is a good idea.  Apparently it confused both
> of us in this case ;-).
Make that three of us. ;)
-Peff
Previous: Junio C HamanoNext: Jeff King
Message 10 of 19 in “git-diff in a worktree is an order of magnitude slower?”
  1. D. Ben KnobleJun 8, 2026
  2. Jeff KingJun 9, 2026
  3. D. Ben KnobleJun 9, 2026
  4. Jeff KingJun 11, 2026
  5. Junio C HamanoJun 11, 2026
  6. brian m. carlsonJun 11, 2026
  7. D. Ben KnobleJun 20, 2026
  8. Junio C HamanoJun 21, 2026
  9. Junio C HamanoJun 21, 2026
  10. Jeff KingJun 21, 2026
  11. Jeff KingJun 21, 2026
  12. Junio C HamanoJun 21, 2026
  13. Jeff KingJun 21, 2026
  14. Junio C HamanoJun 21, 2026
  15. Junio C HamanoJun 22, 2026
  16. Jeff KingJun 28, 2026
  17. Junio C HamanoJun 21, 2026
  18. Jeff KingJun 21, 2026
  19. D. Ben KnobleJul 3, 2026

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.