Was playing with Otto' malloc diff to find leaks and spotted this.
It's not what the malloc diff complains about (the recallocarray()
call in lib/fileindex.c:/^read_fileindex_path around line 591 that I
just can't see -- maybe it's an error in how malloc marks leaks.)

add_entry doesn't release the fileindex entry on error, the callers do
(grep for how got_fileindex_entry_add() is used for reference) and
this is the noly instance where we fail to do so.  Admittedly, it's
very unlikely to leak here, got should have produced an invalid
fileindex and we fail immediately anyway, but it still may be worth,
if not only for reference.  ok?

diff /home/op/w/got
commit - 6be067cef84c15f7e8623ec8fccaf955d98d006b
path + /home/op/w/got
blob - 19b8418d8d3b6252476d3f2048056a0c997b1377
file + lib/fileindex.c
--- lib/fileindex.c
+++ lib/fileindex.c
@@ -752,8 +752,10 @@ got_fileindex_read(struct got_fileindex *fileindex, FI
 		if (err)
 			return err;
 		err = add_entry(fileindex, ie);
-		if (err)
+		if (err) {
+			got_fileindex_entry_free(ie);
 			return err;
+		}
 	}
 
 	n = fread(sha1_expected, 1, sizeof(sha1_expected), infile);