Anders Hermansen <anders@yoyo.no> writes:
> * Tom Lane (tgl@sss.pgh.pa.us) wrote:
>> Looking at the source code, it's clear that it's reporting just the
>> first byte of the sequence; the 00 is redundant and probably shouldn't
>> be in the message.
> Yes. Maybe the error messages can be changed so that what actually went
> wrong is more clear? And possibly printing the whole 3-byte sequence?
Any volunteers for that? The specific message in question is in
src/backend/utils/mb/conversion_procs/utf8_and_iso8859_1/utf8_and_iso8859_1.c
else if ((c & 0xe0) == 0xe0)
elog(ERROR, "could not convert UTF8 character 0x%04x to ISO8859-1",
c);
Aside from being unhelpful as to the exact input data, this is wrong in
another way: it ought to be an ereport() not elog(), because it's
certainly not a can't-happen kind of error.
A little bit of grepping turns up a number of similarly deficient
elog and ereport calls in the src/backend/utils/mb/ tree.
There is more useful code for constructing a character description in
pg_verifymbstr() in src/backend/utils/mb/wchar.c. Probably what ought
to happen is to split out a small subroutine along the lines of
char *describe_mb_char(const unsigned char *mbstr, int len)
(returning a palloc'd string "0x....") and then make all the places
that complain about bad multibyte input use it.
Don't have time to deal with it myself, but it seems like a pretty easy
project for anyone wanting to dip their toes in the backend.
regards, tom lane
