Re: What's cooking in git.git (Nov 2012, #02; Fri, 9)

From: Jeff King <peff@peff.net>
Date: Nov 11, 2012, 15:48 UTC
Message-ID: <20121111154846.GA13188@sigill.intra.peff.net>
In-Reply-To: <87wqxs4o6f.fsf@Niukka.kon.iki.fi>

On Sun, Nov 11, 2012 at 09:02:48AM +0200, Kalle Olavi Niemitalo wrote:

> If git did the same thing as cvs here, i.e. ignore the signals in
> the parent process only and check the exit status of the editor,
> I think that would be OK.

Silly me. When I thought through the impact of Paul's patch, I knew that we would notice signal death of the editor. But I totally forgot to consider that the blocked signal is inherited by the child process. I think we just need to move the signal() call to after we've forked. Like this (on top of Paul's patch):

diff --git a/editor.c b/editor.c
index 3ca361b..0ed23ce 100644
--- a/editor.c
+++ b/editor.c
@@ -38,11 +38,20 @@ int launch_editor(const char *path, struct strbuf *buffer, const char *const *en
 
 	if (strcmp(editor, ":")) {
 		const char *args[] = { editor, path, NULL };
+		struct child_process p;
 		int ret;
 
+		memset(&p, 0, sizeof(p));
+		p.argv = args;
+		p.env = env;
+		p.use_shell = 1;
+		if (start_command(&p) < 0)
+			return error("unable to start editor '%s'", editor);
+
 		sigchain_push(SIGINT, SIG_IGN);
-		ret = run_command_v_opt_cd_env(args, RUN_USING_SHELL, NULL, env);
+		ret = finish_command(&p);
 		sigchain_pop(SIGINT);
+
 		if (ret)
 			return error("There was a problem with the editor '%s'.",
 					editor);

Note that this will give you a slightly verbose message from git.
Potentially we could notice editor death due to SIGINT and suppress the
message, under the assumption that the user hit ^C and does not need to
be told.

-Peff