Re: [ANNOUNCE] Git v2.32.0-rc1 - Defect in t0301.3, t1092.6, t5300

[Taylor Blau <me@ttaylorr.com>]

On Tue, May 25, 2021 at 09:17:40AM -0400, Randall S. Becker wrote:
> On May 24, 2021 6:44 PM, Taylor Blau wrote:
> >To: Randall S. Becker <rsbecker@nexbridge.com>
> >Cc: 'Junio C Hamano' <gitster@pobox.com>; 'Jean-Noël AVILA' <jn.avila@free.fr>; git@vger.kernel.org
> >Subject: Re: [ANNOUNCE] Git v2.32.0-rc1 - Defect in t0301.3, t1092.6, t5300
> >
> >On Mon, May 24, 2021 at 06:37:29PM -0400, Randall S. Becker wrote:
> >> On running the test suite on NonStop, t1092 fails in subtest 6 - but
> >> not when run with --verbose and -x, so I can't supply a log. This may
> >> be timing related. I wanted to report this in case anyone had any
> >> insight on whether there are any possible transient vulnerabilities.
> >
> >The failure in t1092 sounds should be addressed by
> >
> >  https://lore.kernel.org/git/pull.960.git.1621886108515.gitgitgadget@gmail.com/
> >
> >> t0301.3 often has transient failures - which has been experienced for
> >> a while. Most importantly, t5300 continues to fail completely on the
> >> x86 platform, but not ia64, which I previously reported.
> >
> >Not sure about these, though.
>
> Thanks will wait for the roll on that one. We had one loggable transient failure in t1096.15:
>
> expecting success of 1092.15 'merge with outside renames':
>         init_repos &&
>
>         for type in out-to-out out-to-in in-to-out
>         do
>                 test_all_match git reset --hard &&
>                 test_all_match git checkout -f -b merge-$type update-deep &&
>                 test_all_match git merge -m "$type" rename-$type &&
>                 test_all_match git rev-parse HEAD^{tree} || return 1
>         done
>
> + init_repos
> Updating files: 100% (14/14), done.
> HEAD is now at 5d0e8a6 initial commit
> Updating files: 100% (14/14), done.
> HEAD is now at 5d0e8a6 initial commit
> Updating files: 100% (14/14), done.
> HEAD is now at 5d0e8a6 initial commit
> Updating files: 100% (10/10), done.
> Updating files: 100% (7/7), done.
> Updating files: 100% (10/10), done.
> Updating files: 100% (7/7), done.
> + test_all_match git reset --hard
> + test_all_match git checkout -f -b merge-out-to-out update-deep
> + test_all_match git merge -m out-to-out rename-out-to-out
> --- full-checkout-err   2021-05-24 22:49:44 +0000
> +++ sparse-checkout-err 2021-05-24 22:49:50 +0000
> @@ -1 +0,0 @@
> Updating files: 100% (5/5), done.
> + return 1
> error: last command exited with $?=1

I'm having trouble reading this: is 'return 1' part of the output
written to sparse-checkout-err?

In either case, the "updating files" progress meter is coming from
unpack-trees.c:get_progress(). Looking a little further:

  - get_progress() is called by check_updates(), which is called by
    unpack_trees() and update_sparsity()

  - update_sparsity() is called by the sparse-checkout builtin, which
    sets 'o.verbose_update = isatty(2)', so that caller is good.

  - 'git merge' is the last command invoked by t1092.15, and it calls
    unpack_trees() from 'builtin/merge.c:read_tree_trivial()', and sets
    'opts.verbose_update = 1'.

Stolee's patch to set GIT_PROGRESS_DELAY=-1 will obviously take care of
this, but we should probably apply something like this on top (and maybe
look further at other callers of unpack_trees() to make sure they set
opts.verbose_update correctly; I stopped after I found the problem
here).

--- >8 ---

diff --git a/builtin/merge.c b/builtin/merge.c
index eddb8ae70d..5f926d3edc 100644
--- a/builtin/merge.c
+++ b/builtin/merge.c
@@ -677,7 +677,7 @@ static int read_tree_trivial(struct object_id *common, struct object_id *head,
 	opts.src_index = &the_index;
 	opts.dst_index = &the_index;
 	opts.update = 1;
-	opts.verbose_update = 1;
+	opts.verbose_update = isatty(2);
 	opts.trivial_merges_only = 1;
 	opts.merge = 1;
 	trees[nr_trees] = parse_tree_indirect(common);