[PATCH][TRY2] fix false multi-byte matches in some regular expressions

[Roland McGrath]

> Could I use sed for the test case? GNU sed uses libc regexp by default,
> so the possible regression should be catched as well. I still have no C
> language testcase that fails without the fix.

No, that's not going to be reliable.  Surely you can debug sed and discern
the exact sequence of library calls it's making.


Thanks,
Roland