Re: [RFC] building postgres with meson -v8 - Mailing list pgsql-hackers

Hi,

On 2022-05-04 13:53:54 +0200, Peter Eisentraut wrote:
> 0001-meson-Assorted-compiler-test-tweaks.patch
> 
> I was going through a diff of pg_config.h between old and new build and
> found a few omissions and small differences.

Thanks, merged that.


> is of course annoying and can be removed eventually, but it's useful when
> analyzing the diff, and since it's already done in other places it seems
> reasonable to apply it consistently.

Yea, I'd tried to minimize the difference at some point, but haven't done so
in a while...


> 0002-meson-Add-pg_walinspect.patch
> 
> This was added more recently and was not ported yet.  Nothing too
> interesting here.

Merged that.


> 0003-meson-Install-all-server-headers.patch
> 
> With this, all the server headers installed by a makefile-based build are
> installed.  I tried to strike a balance between using install_subdir() with
> exclude list versus listing things explicitly. Different variations might be
> possible, but this looked pretty sensible to me.

I locally had something similar, but I'm worried that this approach will be
too fragile. Leads to e.g. editor temp files getting installed. I've merged it
for now, but I think we need a different approach.


> With these patches, the list of files installed with make versus meson match
> up, except for known open items (postmaster symlink, some library naming
> differences, pkgconfig, pgxs, test modules installed, documentation).

I added pkgconfig since then. They're not exactly the same, but pretty close,
except for one thing: Looks like some of the ecpg libraries really should link
to some other ecpg libs? I think we're missing something there... That then
leads to missing requirements in the .pc files.

Re symlink: Do you have an opion about dropping the symlink vs implementing it
(likely via a small helper script?)?

Re library naming: It'd obviously be easy to adjust the library names, but I
wonder if it'd not be worth keeping the _static.a suffix, right now unsuffixed
library name imo is quite confusing.

Re test modules: Not sure what the best fix for that is yet. Except that we
don't have a search path for server libs, I'd just install them to a dedicated
path or add the build dir to the search path. But we don't, so ...

Re docs: I think the best approach here would be to have a new
meson_options.txt option defining whether the docs should be built. But not
quite sure.

Greetings,

Andres Freund