contrib/tablefunc bugfix

Joe Conway <mail@joeconway.com> writes:

I did find that contrib/cube fails `make installcheck` due to the bison
error message wording change (...syntax error... vs ... parse error
...). It seems contrib/seg was adjusted for this, but cube was not. Is
there a reason for that, or should I send in a patch?

Since we are going to be going through substantial error message wording
changes anyway for 7.4, I'm planning to revert the build-time hack that
preserves the older bison spelling of the error message; we may as
well standardize on the bison 1.875 spelling. So contrib/seg needs its
hack undone, and both need their expected files updated. If you wanna
do that, fine, otherwise I'll get to it when I start doing the backend
error call updates.

regards, tom lane

Tom Lane wrote:

Joe Conway <mail@joeconway.com> writes:

I did find that contrib/cube fails `make installcheck` due to the bison
error message wording change (...syntax error... vs ... parse error
...). It seems contrib/seg was adjusted for this, but cube was not. Is
there a reason for that, or should I send in a patch?

Since we are going to be going through substantial error message wording
changes anyway for 7.4, I'm planning to revert the build-time hack that
preserves the older bison spelling of the error message; we may as
well standardize on the bison 1.875 spelling. So contrib/seg needs its
hack undone, and both need their expected files updated. If you wanna
do that, fine, otherwise I'll get to it when I start doing the backend
error call updates.

OK -- patch attached. Both cube and seg pass `make installcheck` for me now.

Joe

Joe Conway <mail@joeconway.com> writes:

A recent change in the backend exposed a missing include
(lib/stringinfo.h) in contrib/tablefunc. Attached is a fix. Please apply.

Done.

regards, tom lane

Joe Conway <mail@joeconway.com> writes:

I did find that contrib/cube fails `make installcheck` due to the bison
error message wording change (...syntax error... vs ... parse error
...). It seems contrib/seg was adjusted for this, but cube was not. Is
there a reason for that, or should I send in a patch?

Since we are going to be going through substantial error message wording
changes anyway for 7.4, I'm planning to revert the build-time hack that
preserves the older bison spelling of the error message; we may as
well standardize on the bison 1.875 spelling. So contrib/seg needs its
hack undone, and both need their expected files updated. If you wanna
do that, fine, otherwise I'll get to it when I start doing the backend
error call updates.

OK -- patch attached. Both cube and seg pass `make installcheck` for me now.

Applied. contrib passes build, install, installcheck cleanly for me
now. Let me know if you see remaining problems...

regards, tom lane

Tom Lane wrote:

Applied. contrib passes build, install, installcheck cleanly for me
now. Let me know if you see remaining problems...

Thanks. I just updated to cvs tip and installcheck passes for all 90
backend tests and all contrib as well.

Still getting a number of ecpg related warnings on compile, but I assume
we're not worried too much about those as of yet.

Joe

Joe Conway <mail@joeconway.com> writes:

Still getting a number of ecpg related warnings on compile, but I assume
we're not worried too much about those as of yet.

Me too ... I've been starting to think about ragging on Michael about
those. For starters, C files appearing in src/interfaces ought to
include "postgres_fe.h", rather than a random selection of system
headers. Several of the warnings I'm getting would go away if
ecpg/pgtypeslib/ were following that basic portability practice.

regards, tom lane