what to do about unsupported encodings

Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:

pg_get_encoding_from_locale() in chklocale.c reports this when it finds
a locale with an encoding it does not recognize:

ereport(WARNING,
(errmsg("could not determine encoding for locale \"%s\": codeset
is \"%s\"",
ctype, sys),
errdetail("Please report this to <pgsql-bugs@postgresql.org>.")));

I guess we don't get many of these reports. But when testing out all
the locales that an OS provides, I can produce tons of warnings like
this, mostly related to legacy encodings of various localities.

Should we maintain a list to the effect of, these are encodings we have
heard about but don't support? Or should we just drop the "please
report this" part? I think the latter was added when we were still
breaking in this code, but it seems to have held up well.

I agree we could do without that now.

Slightly related: there are some callers such as PGLC_localeconv that
aren't checking for a failure return. Seems bad.

regards, tom lane

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

On Thu, Oct 27, 2016 at 2:01 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:

Slightly related: there are some callers such as PGLC_localeconv that
aren't checking for a failure return. Seems bad.

And also related to the patch trying to improve OOM handling in
pg_locale.c, the point being that elog() cannot be called as long as
the saved locales are not put back in place. The cleanest fix here
would be to use a goto to take the code past restoring the saved
locales, and then issue and ERROR. I think that it would be better to
hit those two stones at once on the other thread.

For the calls in fe-connect.c, why are the calls of
pg_get_encoding_from_locale() requesting to print a message on error?
Shouldn't the caller request to *not* print an error instead and then
store an error string in conn->errorMessage if the encoding is < 0? It
seems to me that at the end the right call is to just remove
write_message all together and let the caller decide the error message
it wants. For HEAD that would be a fine cleanup to do. A patch does
not look that complicated, I could get one out according to those
lines.

Thoughts?
--
Michael

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers