Re: Question about test failure on Hydra

[Stephen Berman]

On Tue, 01 Aug 2017 17:23:36 +0200 Stephen Berman <address@hidden> wrote:

> On Tue, 01 Aug 2017 16:49:16 +0300 Eli Zaretskii <address@hidden> wrote:
>
>>> From: Stephen Berman <address@hidden>
>>> Cc: address@hidden
>>> Date: Tue, 01 Aug 2017 11:57:14 +0200
>>> 
>>> On Tue, 01 Aug 2017 06:12:49 +0300 Eli Zaretskii <address@hidden> wrote:
>>> 
>>> >> From: Stephen Berman <address@hidden>
>>> >> Cc: address@hidden
>>> >> Date: Mon, 31 Jul 2017 11:20:10 +0200
>>> >> 
>>> >> (dired test-dir) reverts an existing Dired buffer, because of setting
>>> >> dired-auto-revert-buffer to t.  (When I step through the code, it is
>>> >> after reverting that point moves to the subdirectory line
>>> >
>>> > By "the point moves" you mean what dired-restore-positions does?
>>> 
>>> Yes.
>>
>> Then I'd begin with logging the values saved in POSITIONS that
>> dired-restore-positions uses, perhaps the recorded positions are
>> sometimes wrong.  Then we could take it from there, either by looking
>> how the values are recorded or how they are restored.
>
> In commit c5305ff6 Tino Calancha added more assertions to track point,
> so this may already be enough to locate the problem if the test fails in
> Hydra again.  If it isn't, we can add still more.

The test is still failing on Hydra in the same way (cf. the log of this
build: https://hydra.nixos.org/build/57876570).  I've added messages
that give the position values before and after the dired call, during
which dired-revert is called, which contains the only calls to
dired-save-positions and dired-restore-positions.  I also added the
messages in two more places before the assertion that fails on Hydra,
and for good measure once more after that assertion.  The resulting
messages (except the one before the dired call) should (and on my test
runs are) all be identical, but should differ on Hydra if point is
somehow moving.

Steve Berman