Eli Zaretskii <eliz@gnu.org> writes:
> > > One particular aspect of the bookmarks that I'd like to see tested is
> > > non-ASCII bookmark names and annotations.  The code includes some
> > > slightly involved logic for determining how to encode the info on
> > > disk, so I wonder how well does that work in practice.
> >
> > OK.  I added tests for that in the attached patch using strings
> > manually copied from HELLO.  It works just fine to save and then load
> > bookmarks for the set of languages I chose to include characters from.
> > That is, we have the same before bookmark-save as we get after
> > bookmark-load.
>
> Thanks.  Please add to that text something that doesn't yield valid
> UTF-8 byte sequence.  For example, these two strings:
>
>   (decode-coding-string "\xE3\x32\x9A\x36" 'chinese-gb18030)
>   (char-to-string (decode-char 'eight-bit #x81))
>
> (Please include these snippets directly, not their result as a literal
> string, to make sure the unusual characters survive intact.)

OK, added.  The tests still pass (both interactively and in batch mode).

The first one, when I evaluate it, makes Emacs segfault.  Not sure why
it works when I'm running the tests - I'd guess because Emacs is not
trying to display it?  Let me know if you want a backtrace for this
crash and/or a separate bug report.

In any case, I've attached an updated patch with the two above
snippets included in the save/load test.  Not sure if we should keep
the first one around, but I added a warning not to evaluate it.

What do you think?

Thanks,
Stefan Kangas