On Mon, Feb 26, 2018 at 06:27:06PM +0100, tbo...@web.de wrote:
> @@ -3611,8 +3615,25 @@ int diff_populate_filespec(struct diff_filespec *s,
> unsigned int flags)
> s->size = size;
> s->should_free = 1;
> }
> - }
> - else {
> + if (!s->binary && buffer_is_binary(s->data, s->size) &&
> + buffer_has_utf16_bom(s->data, s->size)) {
> + int outsz = 0;
> + char *outbuf;
> + outbuf = reencode_string_len(s->data, (int)s->size,
> + "UTF-8", "UTF-16", &outsz);
> + if (outbuf) {
> + if (s->should_free)
> + free(s->data);
> + if (s->should_munmap)
> + munmap(s->data, s->size);
> + s->should_munmap = 0;
> + s->data = outbuf;
> + s->size = outsz;
> + s->reencoded_from_utf16 = 1;
> + s->should_free = 1;
> + }
> + }
> + } else {
I don't think it makes sense to do the conversion deep inside
diff_populate_filespec(), because it will kick in much more than you'd
want (e.g., "format-patch | am" would stop working with this patch, I
think).
I think you'd want to hook this in at the same level as fill_textconv().
In fact, one way to do it would be to have the get_textconv() stack just
fill in a special driver that does the auto-detection. This is similar
to my earlier patch, but it avoids overriding the user-facing config for
non-textconv things (and naturally any actual configured textconv filter
would override the auto-detection).
-Peff
