On 11 Sep 2017 15:54, "Eli Zaretskii" <eliz@gnu.org> wrote:

> From: Richard Copley <rcopley@gmail.com>
> Date: Sun, 10 Sep 2017 23:41:45 +0100
> Cc: Tom Tromey <tom@tromey.com>, 28349@debbugs.gnu.org
>
> This is the backtrace when string-match sets the match data to (0 3):

So you are saying the problem is in the call to string-match?  Or does
some code clobber match-data between that call and when the data is
used?


The call to string-match clobbers the correct match data, which was
obtained just before by re-search-forward, in compilation-parse-errors.

FYI this is not new and not related to Tom's change as far as I can see.

In compilation-parse-errors (compile.el:1426) (upcase comments added):

        [...]
        (goto-char start)
        (while (re-search-forward pat end t)  ;; SET MATCH-DATA
          (when (setq props (compilation-error-properties ;; CLOBBER IT
                             file line end-line col end-col (or type 2)
fmt))
        [...]
        ;; USE IT