Thread: Re: compiling PL/pgSQL plugin with C++

[ redirecting to -hackers ]

=?koi8-r?B?9MHSwdPP1yDnxc/Sx8nKIPfJ1MHM2MXXyd4=?= <Tarasov-G@gaz-is.ru> writes:
> I'm working on development of some PL/pgSQL plugin.
> The smaller part of my code is written on C.
> It's a standard extension code for integration with fmgr (_PG_init ...)
> But bigger part of the code is written on C++.
> And here I need declarations of internal PL/pgSQL structs from plpgsql.h

So ... that's supposed to work, because we have a test script that
verifies that all our headers compile as C++.

Or I thought it was "all", anyway.  Closer inspection shows that it's
not checking src/pl.  Nor contrib.

I propose that we change src/tools/pginclude/cpluspluscheck so that
it searches basically everywhere:

-for f in `find src/include src/interfaces/libpq/libpq-fe.h src/interfaces/libpq/libpq-events.h -name '*.h' -print | \
+for f in `find src contrib -name '*.h' -print | \

However, trying to run that, I find that plpython and plperl are both
seriously in violation of the project convention that headers should
compile standalone.  It looks like most of their headers rely on an
assumption that the calling .c file already included the Python or
Perl headers respectively.

Anybody object to me reshuffling the #include's to make this pass?
I propose doing that for HEAD only, although we should back-patch
George's fix (and any other actual C++ problems we find).

            regards, tom lane

I wrote:
> I propose that we change src/tools/pginclude/cpluspluscheck so that
> it searches basically everywhere:

> -for f in `find src/include src/interfaces/libpq/libpq-fe.h src/interfaces/libpq/libpq-events.h -name '*.h' -print |
\
> +for f in `find src contrib -name '*.h' -print | \

After further experimentation with that, it seems like we'll have
to continue to exclude src/bin/pg_dump/*.h from the C++ check.
pg_dump uses "public" and "namespace" as field names in various
structs, both of which are C++ keywords.  Changing these names
would be quite invasive, and at least in the short run I see no
payoff for doing so.

ecpg/preproc/type.h is also using "new" as a field name, but it
looks like there are few enough references that renaming that
field isn't unreasonable.

There are various other minor issues, but they generally look
fixable with little consequence.

            regards, tom lane

I wrote:
> There are various other minor issues, but they generally look
> fixable with little consequence.

I've now pushed your patch and additional minor fixes, and
we've expanded cpluspluscheck's coverage so we don't miss
such issues in future.

            regards, tom lane