From: Junio C Hamano Date: Sat, 01 Nov 2008 05:39:58 GMT Subject: Re: [PATCH] connect.c: add a way for git-daemon to pass an error back to client Message-ID: <7vy7043sy9.fsf@gitster.siamese.dyndns.org> In-Reply-To: "Tom Preston-Werner" writes: > Example > > S: ERR No matching repository. > C: fatal: remote error: No matching repository. I like what this tries to do. I briefly wondered if this should be restricted to the very first message from the other end, but I think it is not necessary. If the remote throws a few valid looking "SHA-1 SP refname" lines and then said "ERR" (which cannot be the beginning of a valid SHA-1), we can safely and unambiguously declare that this is an error message from the remote end. > diff --git a/connect.c b/connect.c > index 0c50d0a..3af91d6 100644 > --- a/connect.c > +++ b/connect.c > @@ -70,6 +70,9 @@ struct ref **get_remote_heads(int in, struct ref **list, > if (buffer[len-1] == '\n') > buffer[--len] = 0; > > + if (len > 4 && !memcmp("ERR", buffer, 3)) Would matching 4 bytes "ERR " here an improvement? You are expecting buffer+4 is where the message begins in die() anyway, and otherwise you would show the message without "N" if you got "ERRNo matching repo". > + die("remote error: %s", buffer + 4); > + It was very considerate that you did not say "server error" in the error message. This code is shared between both the fetch side and the push side.