Re: Mark all GUC variable as PGDLLIMPORT

On Thu, Aug 26, 2021 at 05:10:39PM -0400, Andrew Dunstan wrote:
> On 8/26/21 3:57 PM, Robert Haas wrote:
>> On Thu, Aug 26, 2021 at 3:42 PM Magnus Hagander <magnus(at)hagander(dot)net> wrote:
>>> Ugly as it is, I wonder if there's a chance we could just process all
>>> the headers at install times and inject the PGDLLIMPORT. We know which
>>> symvols it is on account of what we're getting in the DEF file.
>>> Not saying that's not a very ugly solution, but it might work?

I missed the word "install" first here :)

>> If it's ugly, that might mean it's a bad idea and we shouldn't do it
>> ... but if it can be made not-too-ugly, it would certainly be nice to
>> be able to stop worrying about this.
>
> How is this going to affect msys builds? No gendef there IIRC. I guess
> some similar procedure might be possible ...

Wouldn't that be needed for cygwin as well? If we go down to enable
that for a maximum number of parameters, I would really agree for
doing things so as this never gets forgotten for new parameters and we
don't have to discuss the matter anymore. With all that in mind, that
would mean a new perl script that does the job, callable by both MSVC
and normal make builds. But we have nothing that does a manipulation
of the installation contents. And couldn't it be a problem if an
installation is overwritten, updated or upgradedd, where there may be
contents not coming from the in-core build process but from some
extension?
--
Michael