Re: [PATCH v2 0/3] nd/clear-gitenv-upon-use-of-alias

[Jeff King <peff@peff.net>]

On Tue, Dec 22, 2015 at 10:13:03AM -0800, Junio C Hamano wrote:

> > Another by the way, this "forcing aliases as external commands" now
> > shows something like "error: git-log died of signal 13" when the pager
> > exits early, for an alias like "l1 = log --oneline".
> 
> ... and we do not show that when we directly call "git log" is...?
> 
> We do signal this with non-zero exit status like so:
> 
> 	$ GIT_PAGER=true git log --oneline ; echo $?
>         141
> 
> and it is not surprising that the one that is catching the exit
> status of what was spawned and reporting "signal 13".

This sounded very familiar. Apparently I've been running with the patch
below for almost 3 years and never got around to re-examining it.

The original discussion is in:

  http://thread.gmane.org/gmane.comp.version-control.git/212630

Having skimmed through the arguments there, and given that we applied
the patch from the middle of the thread, I think this is a good change.

-- >8 --
Date: Fri, 4 Jan 2013 07:19:35 -0500
Subject: [PATCH] avoid SIGPIPE warnings for aliases

When git executes an alias that specifies an external
command, it will complain if the alias dies due to a signal.
This is usually a good thing, as signal deaths are
unexpected. However, SIGPIPE is not unexpected for many
commands which produce a lot of output; it is intended that
the user closing the pager would kill them them via SIGPIPE.

As a result, the user might see annoying messages in a
scenario like this:

  $ cat ~/.gitconfig
  [alias]
  lgbase = log --some-options
  lg = !git lgbase --more-options
  lg2 = !git lgbase --other-options

  $ git lg -p
  [user hits 'q' to exit pager]
  error: git lgbase --more-options died of signal 13
  fatal: While expanding alias 'lg': 'git lgbase --more-options': Success

Many users won't see this, because we execute the external
command with the shell, and a POSIX shell will silently
rewrite the signal-death exit code into 128+signal, and we
will treat it like a normal exit code. However, this does
not always happen:

  1. If the sub-command does not have shell metacharacters,
     we will skip the shell and exec it directly, getting
     the signal code.

  2. Some shells do not do this rewriting; for example,
     setting SHELL_PATH set to zsh will reveal this problem.

This patch squelches the message, and lets git exit silently
(with the same exit code that a shell would use in case of a
signal).

The first line of the message comes from run-command's
wait_or_whine. To silence that message, we remain quiet
anytime we see SIGPIPE.

The second line comes from handle_alias itself. It calls
die_errno whenever run_command returns a negative value.
However, only -1 indicates a syscall error where errno has
something useful (note that it says the useless "success"
above). Instead, we treat negative returns from run_command
(except for -1) as a normal code to be passed to exit.

Signed-off-by: Jeff King <peff@peff.net>
---
 git.c         | 2 +-
 run-command.c | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/git.c b/git.c
index 6ed824c..34a18a3 100644
--- a/git.c
+++ b/git.c
@@ -252,7 +252,7 @@ static int handle_alias(int *argcp, const char ***argv)
 			alias_argv[argc] = NULL;
 
 			ret = run_command_v_opt(alias_argv, RUN_USING_SHELL);
-			if (ret >= 0)   /* normal exit */
+			if (ret != -1)  /* normal exit */
 				exit(ret);
 
 			die_errno("While expanding alias '%s': '%s'",
diff --git a/run-command.c b/run-command.c
index 13fa452..694a6ff 100644
--- a/run-command.c
+++ b/run-command.c
@@ -245,7 +245,7 @@ static int wait_or_whine(pid_t pid, const char *argv0, int in_signal)
 		error("waitpid is confused (%s)", argv0);
 	} else if (WIFSIGNALED(status)) {
 		code = WTERMSIG(status);
-		if (code != SIGINT && code != SIGQUIT)
+		if (code != SIGINT && code != SIGQUIT && code != SIGPIPE)
 			error("%s died of signal %d", argv0, code);
 		/*
 		 * This return value is chosen so that code & 0xff
-- 
2.7.0.rc2.368.g1cbb535