rather than using an isalnum-equivalent test. The direct answer is that ts_locale.c/.h provides no such test function, which apparently is because there's not a lot of potential callers in the core code. However, both pg_trgm and ltree could benefit from adding one.
There's no semantic hazard here: the documentation I consulted is all pretty explicit that is[w]alnum is true exactly when either is[w]alpha or is[w]digit are. For example, POSIX saith
The iswalpha() and iswalpha_l() functions shall test whether wc is a wide-character code representing a character of class alpha in the current locale, or in the locale represented by locale, respectively; see XBD Locale.
The iswdigit() and iswdigit_l() functions shall test whether wc is a wide-character code representing a character of class digit in the current locale, or in the locale represented by locale, respectively; see XBD Locale.
The iswalnum() and iswalnum_l() functions shall test whether wc is a wide-character code representing a character of class alpha or digit in the current locale, or in the locale represented by locale, respectively; see XBD Locale.
While I didn't try to actually measure it, these functions don't look remarkably cheap. Doing char2wchar() twice when we only need to do it once seems silly, and the libc functions themselves are probably none too cheap for multibyte characters either.
Hence, I propose the attached. I got rid of some places that were unnecessarily checking pg_mblen before applying t_iseq(), too.
regards, tom lane
I see this is already committed, but I'm curious, why do t_isalpha and t_isdigit have the pair of /* TODO */ comments? This unfinished business isn't explained anywhere in the file.
