Re: psql crashes on encoding mismatch

2011/1/13 Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>:
> Hitoshi Harada <umi(dot)tanuki(at)gmail(dot)com> writes:
>> I found a crash case (assertion failure) when runing psql -f
>> utf8_encoded_script.sql against client_encoding = shift_jis in
>> postgresql.conf. Though encoding mismatch is obviously user's fault, a
>> crash doesn't explain anything to him.
>
> I'm not too impressed with this patch: it seems like the most it will
> accomplish is to move the failure to some other, equally obscure, place
> --- because you'll still have a string that's invalidly encoded.
> Moreover, if you've got wrongly encoded data, it wouldn't be hard at all
> for it to mess up psql's lexing; consider cases such as a
> character-that's-not-as-long-as-we-think just in front of a quote mark.
>
> Shouldn't we instead try to verify the multibyte encoding somewhere
> upstream of here?

I had thought it before going into the patch, too. However, the fact
that psql(fe-misc.c) doesn't have PQverfiymb() although it has
PQmblen() implied to me that encoding verification should be done in
server side perhaps. I might be too ignorant to imagine the lexing
problem of your quote mark, but my crash sample has multibyte
characters in sql comment, which is ignored in the server parsing. If
we decided that the case raises error, wouldn't some existing
applications be broken? I can imagine they are in the same situation
of encoding mismatch and are run without problem I found by chance.

Just for reference I attach the case sql file. To reproduce it:

1. initdb
2. edit client_encoding = shift_jis in postgresql.conf
3. start postgres
4. psql -f case_utf8.sql

Note: the line break should be LF as the file stands. CR-LF cannot
reproduce the problem.

Regards,

--
Hitoshi Harada