Ikiwiki should really survive being asked to work with a git branch that has no existing commits.

    mkdir iki-gittest
    cd iki-gittest
    GIT_DIR=barerepo.git git init
    git clone barerepo.git srcdir
    ikiwiki --rcs=git srcdir destdir

I've fixed this initial construction case, and, based on my testing, I've also fixed the post-update executing on a new master, and ikiwiki.cgi executing on a non-existent master cases.

Please commit so my users stop whining at me about having clean branches to push to, the big babies.

Summary: Change three scary loud failure cases related to empty branches into three mostly quiet success cases.

[[!tag patch]]

> FWIW, [[The_TOVA_Company]] apparently wants this feature (and I hope
> I don't mind that I mention they were willing to pay someone for it,
> but I told them I'd not done any of the work. :) )
> 
> Code review follows, per hunk.. --[[Joey]] 

<pre>
diff --git a/IkiWiki/Plugin/git.pm b/IkiWiki/Plugin/git.pm
index cf7fbe9..e5bafcf 100644
--- a/IkiWiki/Plugin/git.pm
+++ b/IkiWiki/Plugin/git.pm
@@ -439,17 +439,21 @@ sub git_commit_info ($;$) {
 
 	my @opts;
 	push @opts, "--max-count=$num" if defined $num;
-
-	my @raw_lines = run_or_die('git', 'log', @opts,
-		'--pretty=raw', '--raw', '--abbrev=40', '--always', '-c',
-		'-r', $sha1, '--', '.');
-
+	my @raw_lines;
 	my @ci;
-	while (my $parsed = parse_diff_tree(\@raw_lines)) {
-		push @ci, $parsed;
-	}
+        
+	# Test to see if branch actually exists yet.
+	if (run_or_non('git', 'show-ref', '--quiet', '--verify', '--', 'refs/heads/' . $config{gitmaster_branch}) ) {
+		@raw_lines = run_or_die('git', 'log', @opts,
+			'--pretty=raw', '--raw', '--abbrev=40', '--always', '-c',
+			'-r', $sha1, '--', '.');
+
+		while (my $parsed = parse_diff_tree(\@raw_lines)) {
+			push @ci, $parsed;
+		}
 
-	warn "Cannot parse commit info for '$sha1' commit" if !@ci;
+		warn "Cannot parse commit info for '$sha1' commit" if !@ci;
+	};
 
 	return wantarray ? @ci : $ci[0];
 }
</pre>

My concern is that this adds a bit of slowdown (git show-ref is fast, but
It's still extra work) to a very hot code path that is run to eg,
update recentchanges after every change. 

Seems not ideal to do extra work every time to handle a case
that will liternally happen a maximum of once in the entire lifecycle of a
wiki (and zero times more typically, since the setup automator puts in a
.gitignore file that works around this problem).

So as to not just say "no" ... what if it always tried to run git log,
and if it failed (or returned no parsed lines, then it could look 
at git show-ref to desice whether to throw an error or not.
--[[Joey]] 

<pre>
@@ -474,7 +478,10 @@ sub rcs_update () {
 	# Update working directory.
 
 	if (length $config{gitorigin_branch}) {
-		run_or_cry('git', 'pull', '--prune', $config{gitorigin_branch});
+		run_or_cry('git', 'fetch', '--prune', $config{gitorigin_branch});
+		if (run_or_non('git', 'show-ref', '--quiet', '--verify', '--', 'refs/remotes/' . $config{gitorigin_branch} . '/' . $config{gitmaster_branch}) ) {
+			run_or_cry('git', 'merge', $config{gitorigin_branch} . '/' . $config{gitmaster_branch});
+		}
 	}
 }
</pre>

Same concern here about extra work. Code path is nearly as hot, being
called on every refresh. Probably could be dealt with similarly as above.

Also, is there any point in breaking the pull up into a
fetch followed by a merge? --[[Joey]] 

<pre>
@@ -559,7 +566,7 @@ sub rcs_commit_helper (@) {
 	# So we should ignore its exit status (hence run_or_non).
 	if (run_or_non('git', 'commit', '-m', $params{message}, '-q', @opts)) {
 		if (length $config{gitorigin_branch}) {
-			run_or_cry('git', 'push', $config{gitorigin_branch});
+			run_or_cry('git', 'push', $config{gitorigin_branch}, $config{gitmaster_branch});
 		}
 	}
 	
</pre>

This seems fine to apply. --[[Joey]]