Andrew Dunstan <and...@dunslane.net> writes:
> On 2025-08-04 Mo 6:35 AM, John Naylor wrote:
>> There is a risk of breaking applications, although only a few dozen
>> mappings changed. If it were added as a separate encoding, users could
>> opt in.
> That makes sense ... naming the new encoding so as to avoid confusion
> might be a challenge.
We have precedent for that in SHIFT_JIS_2004. Presumably if we
make this a new encoding, it'd be GB18030_2022.
However, adding a new encoding ID is not without breakage risks
of its own, stemming from some code knowing the new ID and others
not. I recall that we had some actual problems of that ilk when
we added SHIFT_JIS_2004, and some of them were pretty subtle.
See e.g. this comment from src/bin/initdb/Makefile:
# Note: it's important that we link to encnames.o from libpgcommon, not
# from libpq, else we have risks of version skew if we run with a libpq
# shared library from a different PG version. Define
# USE_PRIVATE_ENCODING_FUNCS to ensure that that happens.
That was long enough ago that I have little faith either that that
fix still does what it intended to (the code has been rejiggered
significantly since the issue was last battle-tested), or that
there are not similar hazards elsewhere.
So on the whole I'd lean a bit towards just redefining GB18030 as
meaning the new standard. The fact that we don't support it as a
server-side encoding perhaps makes that idea more tenable than it
would be if the encoding governed the interpretation of our own
stored data.
regards, tom lane
