Hi,
On 2023-01-04 16:18:38 -0500, Tom Lane wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
> > If we don't do as Peter suggests, then any difference between the
> > results of one build system and the other could either be a bug or an
> > intentional deviation. There will be no easy way to know which it is.
> > And if or when people switch build systems, stuff will be randomly
> > different, and they won't understand why.
Given the difference is "localized", I think calling this out in the docs
would contain confusion.
> > I hear your point too. It's unpleasant for you to spend a lot of
> > effort overriding meson's behavior if the result is arguably worse
> > than the default, and it has the effect of carrying forward in
> > perpetuity hacks that may not have been a good idea in the first
> > place, or may not be a good idea any more. Those seem like valid
> > concerns, too.
This specific instance luckily is trivial to change from code POV.
> Yeah. I think the way forward probably needs to be to decide that
> we are (or are not) going to make changes to the installation tree
> layout, and then make both build systems conform to that. I don't
> really buy the argument that it's okay to let them install different
> layouts. I *am* prepared to listen to arguments that "this is dumb
> and we shouldn't do it anymore".
What exactly shouldn't we do anymore?
I just want to re-iterate that, in my understanding, what we're talking about
here is just whether libdir defaults to just "lib" or whether it adapts to the
platform default (so we end up with libdir as 'lib64' or
'lib/x86_64-linux-gnu' etc). And *not* whether we should continue to force
"postgresql" into the paths.
Greetings,
Andres Freund
