On Wed, 2025-11-12 at 19:59 +0100, Peter Eisentraut wrote:
> Many of these issues are pre-existing, but I just figured it has
> reached
> a point where we need to do something about it.
I tried to simplify things in this patch series, assuming that we have
some tolerance for small behavior changes.
0001: No behavior change here, same patch as before. Uncontroversial
simplification, so I plan to commit this soon.
0002: change fuzzystrmatch to use ASCII semantics. As far as I can
tell, this only affects the results of soundex(). Before the patch, in
en_US.iso885915, soundex('réd') was 'RÉ30', after the patch it's
'Ré30'. I'm not sure whether the current behavior is intentional or
not. Other functions (daitch_mokotoff, levenshtein, and metaphone) are
unaffected as far as I can tell.
0003+0005: change ltree to use case folding instead of tolower(). I
believe this is a bug fix, because the current code is inconsistent
between ltree_strncasecmp() and ltree_crc32_sz().
0006-0007: Remove char_tolower() API. This also removes the
optimization for single-byte encodings with the libc provider and a
non-C locale, but simplifies the code (the optimization is retained for
the C locale). It's possible to make the lazy-folding optimization work
for all locales without the char_tolower() API by doing something
simlar to what 0004 does for ltree. But to make this work efficiently
for Generic_Text_IC_like() would be a bit more complex: we'd need to
adjust MatchText() to be able to fold the arguments lazily, and perhaps
introduce some kind of casemapping iterator. That's already a pretty
complex function, so I'm hesitant to do that work unless the
optimization is important.
These patches don't get us quite to the point of eliminating the
LC_CTYPE dependency (there's still downcase_identifier() and
pg_strcasecmp() to worry about, and some assorted isxyz() calls to
examine), but they simplify things enough that the path forward will be
easier.
Regards,
Jeff Davis
