Re: unaccent fails when datlocprovider=i and datctype=C

On 08.03.23 06:49, Jeff Davis wrote:
> $ initdb -D data -N --locale-provider=icu --icu-locale=en --locale=C

Is it even worth supporting that? What is the point of this kind of setup?

> =# create extension unaccent;
> ERROR: invalid multibyte character for locale
> HINT: The server's LC_CTYPE locale is probably incompatible with the
> database encoding.
> CONTEXT: line 1 of configuration file
> ".../share/tsearch_data/unaccent.rules": "¡ !
> "
>
> Cause: t_isspace() implementation is incomplete (notice "TODO"
> comments):
>
> Oid collation = DEFAULT_COLLATION_OID; /* TODO */
> pg_locale_t mylocale = 0; /* TODO */
>
> if (clen == 1 || lc_ctype_is_c(collation))
> return isspace(TOUCHAR(ptr));
>
> char2wchar(character, WC_BUF_LEN, ptr, clen, mylocale);
>
> return iswspace((wint_t) character[0]);
>
> If using datlocprovider=c, then the earlier branch goes straight to
> isspace(). But if datlocprovider=i, then
> lc_ctype_is_c(DEFAULT_COLLATION_OID) returns false, and it goes into
> char2wchar(). char2wchar() is essentially a wrapper around mbstowcs(),
> which does not work on multibyte input when LC_CTYPE=C.
>
> Quick fix (attached): check whether datctype is C rather than the
> default collation.

This seems right. It's unfortunate that we would now have the
possibilty that

lc_ctype_is_c(DEFAULT_COLLATION_OID) != database_ctype_is_c

but that seems to be the nature of things. Maybe a comment somewhere?