archive mirror
 help / color / mirror / Atom feed
From: Jeff King <>
To: Johannes Schindelin <>
Cc: Alexandr Miloslavskiy <>,
	Alexandr Miloslavskiy via GitGitGadget  <>,, Junio C Hamano <>
Subject: Re: [PATCH v2 1/1] vreportf: Fix interleaving issues, remove 4096 limitation
Date: Sat, 26 Oct 2019 17:36:48 -0400	[thread overview]
Message-ID: <> (raw)
In-Reply-To: <>

On Sat, Oct 26, 2019 at 10:56:45PM +0200, Johannes Schindelin wrote:

> Back to the issue at hand: I did open a GitGitGadget PR with my proposed
> change, in the hopes that I could somehow fast-track this fix into the
> CI/PR builds over at, but there are
> problems: it seems that now there is an at least occasional broken pipe
> in the same test when run on macOS.

Yes, I think that's another issue in the same test. There's more
discussion further down in the thread I linked earlier, starting here:

and I think Gábor's solution here:

is the right direction (and note that this _isn't_ just a test artifact,
but a bug that occasionally hits real-world cases, too).

> There _also_ seems to be something spooky going on in t3510.12 and .13,
> where the expected output differs from the actual output only by a
> re-ordering of the lines:
> -- snip --
> [...]
> +++ diff -u expect advice
> --- expect	2019-10-25 22:17:44.982884700 +0000
> +++ advice	2019-10-25 22:17:45.278884500 +0000
> @@ -1,3 +1,3 @@
>  error: cherry-pick is already in progress
> -hint: try "git cherry-pick (--continue | --skip | --abort | --quit)"
>  fatal: cherry-pick failed
> +hint: try "git cherry-pick (--continue | --skip | --abort | --quit)"
> -- snap --

Hrm. I'd have thought those are both coming from the same process. Which
implies that we're not fflushing stderr before calling write(2). But
your patch seems to do so...

<scratches head> Aha. I think you force-pushed up as I am typing this.
:) So I think that is indeed the solution.

> So much as I would love to see the flakiness of t5516 be fixed as soon
> as possible, I fear we will have to look at the underlying issue a bit
> closer: there are two processes writing to `stderr` concurrently. I
> don't know whether there would be a good way for the `stderr` of the
> `upload-pack` process to be consumed by the `fetch` process, and to be
> printed by the latter.

The worst part is that this message already _is_ consumed by fetch: we
send it twice, once over the sideband, and once directly to stderr. In
most cases the stderr version is lost, but some server providers might
be collecting it. I wouldn't mind seeing the direct-to-stderr one
dropped. There's some more discussion in (from the same thread linked


  reply	other threads:[~2019-10-26 21:36 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-10-22 14:39 [PATCH 0/1] vreportf: Fix interleaving issues, remove 4096 limitation Alexandr Miloslavskiy via GitGitGadget
2019-10-22 14:39 ` [PATCH 1/1] " Alexandr Miloslavskiy via GitGitGadget
2019-10-22 14:45 ` [PATCH v2 0/1] " Alexandr Miloslavskiy via GitGitGadget
2019-10-22 14:45   ` [PATCH v2 1/1] " Alexandr Miloslavskiy via GitGitGadget
2019-10-25 11:37     ` Johannes Schindelin
2019-10-25 12:38       ` Alexandr Miloslavskiy
2019-10-25 14:02         ` Johannes Schindelin
2019-10-25 14:15           ` Alexandr Miloslavskiy
2019-10-25 21:28             ` Johannes Schindelin
2019-10-25 22:11           ` Jeff King
2019-10-26  8:02             ` Alexandr Miloslavskiy
2019-10-26 20:56             ` Johannes Schindelin
2019-10-26 21:36               ` Jeff King [this message]
2019-10-28 16:05                 ` Johannes Schindelin
2019-10-26 21:56               ` Johannes Schindelin
2019-10-26 22:05                 ` Johannes Schindelin

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \ \ \ \ \ \ \ \

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).