Re: build remaining Flex files standalone

Hi,

On 2022-08-16 17:41:43 +0700, John Naylor wrote:
> For v3, I addressed some comments and added .h files to the
> headerscheck exceptions.

Thanks!

> /*
> * NB: include bootparse.h only AFTER including bootstrap.h, because bootstrap.h
> * includes node definitions needed for YYSTYPE.
> */
>
> Future cleanup: I see this in headerscheck:
>
> # We can't make these Bison output files compilable standalone
> # without using "%code require", which old Bison versions lack.
> # parser/gram.h will be included by parser/gramparse.h anyway.
>
> That directive has been supported in Bison since 2.4.2.

2.4.2 is from 2010. So I think we could just start relying on it?

> > > +/* functions shared between guc.c and guc-file.l */
> > > [...]
> > I think I prefer your suggestion of a guc_internal.h upthread.
>
> Started in 0002, but left open the headerscheck failure.
>
> Also, if such a thing is meant to be #include'd only by two generated
> files, maybe it should just live in the directory where they live, and
> not in the src/include dir?

It's not something we've done for the backend afaics, but I don't see a reason
not to start at some point.

> > > From 7d4ecfcb3e91f3b45e94b9e64c7c40f1bbd22aa8 Mon Sep 17 00:00:00 2001
> > > From: John Naylor <john(dot)naylor(at)postgresql(dot)org>
> > > Date: Fri, 12 Aug 2022 15:45:24 +0700
> > > Subject: [PATCH v201 2/9] Build booscanner.c standalone
> >
> > > -# bootscanner is compiled as part of bootparse
> > > -bootparse.o: bootscanner.c
> > > +# See notes in src/backend/parser/Makefile about the following two rules
> > > +bootparse.h: bootparse.c
> > > + touch $@
> > > +
> > > +bootparse.c: BISONFLAGS += -d
> > > +
> > > +# Force these dependencies to be known even without dependency info built:
> > > +bootparse.o bootscan.o: bootparse.h
> >
> > Wonder if we could / should wrap this is something common. It's somewhat
> > annoying to repeat this stuff everywhere.
>
> I haven't looked at the Meson effort recently, but if the build rule
> is less annoying there, I'm inclined to leave this as a wart until
> autotools are retired.

The only complicating thing in the rules there is the dependencies from one .c
file to another .c file.

> > > diff --git a/contrib/cube/cubedata.h b/contrib/cube/cubedata.h
> > > index dbe7d4f742..0b373048b5 100644
> > > --- a/contrib/cube/cubedata.h
> > > +++ b/contrib/cube/cubedata.h
> > > @@ -67,3 +67,7 @@ extern void cube_scanner_finish(void);
> > >
> > > /* in cubeparse.y */
> > > extern int cube_yyparse(NDBOX **result);
> > > +
> > > +/* All grammar constructs return strings */
> > > +#define YYSTYPE char *
> >
> > Why does this need to be defined in a semi-public header? If we do this in
> > multiple files we'll end up with the danger of macro redefinition warnings.
>
> I tried to put all the Flex/Bison stuff in another *_internal header,
> but that breaks the build. Putting just this one symbol in a header is
> silly, but done that way for now. Maybe two copies of the symbol?

The problem is that if it's in a header you can't include another header with
such a define. That's fine if it's a .h that's just intended to be included by
a limited set of files, but for something like a header for a datatype that
might need to be included to e.g. define a PL transform or a new operator or
... This would be solved by the %code requires thing, right?

Greetings,

Andres Freund