On Fri, Dec 11, 2020 at 01:55:11AM +0100, Stefan Sperling wrote:
> Off-list, jrick suggested to keep the behaviour intact where histedit's
> commit operation is aborted if the saved log message is empty.

This turned out to be a deeper rabbit hole than expected.

There is a lot of fallout from allowing an empty log message to abort
the histedit operation. I've written tests in an attempt to cover the
expected behaviour, and it has now become clear that a lot more work
is required in order to get this work properly.

For now, I'd like to stick to the originally proposed change, and accept
the fact that saving an empty log message won't abort the operation.

All the rest can be revisted after the 0.45 release has been cut.

ok?

diff a2c1255649ab1b8d0fc66ec6ff07646b6176e764 c10001dba47001c869fa9d1b8f1981b66c10cccc
blob - 7bdd9e142a5f26254c73c308684bb6d6da95ef20
blob + 997426c9255fb2e425f894acf4aa233d6fdb7722
--- got/got.c
+++ got/got.c
@@ -8018,7 +8018,10 @@ histedit_edit_logmsg(struct got_histedit_list_entry *h
 	if (err) {
 		if (err->code != GOT_ERR_COMMIT_MSG_EMPTY)
 			goto done;
-		err = got_object_commit_get_logmsg(&hle->logmsg, commit);
+		err = NULL;
+		hle->logmsg = strdup(new_msg);
+		if (hle->logmsg == NULL)
+			err = got_error_from_errno("strdup");
 	}
 done:
 	if (logmsg_path && unlink(logmsg_path) != 0 && err == NULL)