On Thu, Feb 29, 2024 at 04:01:47AM +0100, Jelte Fennema-Nio wrote:
> On Thu, 29 Feb 2024 at 01:57, Michael Paquier <michael@paquier.xyz> wrote:
>> I have doubts about the changes in raw_pg_bind_textdomain_codeset(),
>> as the encoding could come from the value in the pg_database tuples
>> themselves. The current coding is slightly safer from the perspective
>> of bogus input values as we would loop over pg_enc2gettext_tbl looking
>> for a match. 0003 changes that so as we could point to incorrect
>> memory areas rather than fail safely for the NULL check.
>
> That's fair. Attached is a patch that adds a PG_VALID_ENCODING check
> to raw_pg_bind_textdomain_codeset to solve this regression.
- for (i = 0; pg_enc2gettext_tbl[i].name != NULL; i++)
+ if (!PG_VALID_ENCODING(encoding) || pg_enc2gettext_tbl[encoding] == NULL) {
Shouldn't PG_MULE_INTERNAL point to NULL in pg_enc2gettext_tbl[]?
That just seems safer to me, and more consistent because its values
satisfies PG_VALID_ENCODING().
--
Michael