Re: [PATCH v5 for-5.0] configure: warn if not using a separate build directory

[Peter Maydell]

On Wed, 15 Apr 2020 at 07:13, Markus Armbruster <address@hidden> wrote:
> Peter Maydell <address@hidden> writes:
> > Given where we are in the release cycle, I think this isn't
> > going to go in for 5.0; and it's not really that urgent now
> > we've decided we don't want to actually deprecate in-tree builds.
>
> Have we?
>
> We had a Aleksandar assert that out-of-tree builds can't do certain
> things, which led to the decision to soften this patch's warning from
> "deprecated; better use the grace period to adjust, and here's how to"
> to "not recommended; here's the recommended way".

This is not how I recall the discussion. The reason we decided to
soften this patche's warning was because there was a sizeable
contingent of people who said "I want the basic './configure; make'
commands to keep working and am willing to write the wrapper
makefile that wil cause those to automatically create and use
a build directory under the hood". If the commands will keep working
then there's nothing to deprecate.

("Stuff doesn't currently work with out-of-tree builds" is something
that I argued at that point in the thread was definitely not a reason
not to deprecate.)

> Whether we want to keep sinking time & energy into an extra way to build
> will become irrelevant once we move to Meson, unless Meson deviates from
> its "this is an opinionated build tool, not a 'give users all the rope
> they may possibly want, and then some'" approach in a surprising lapse
> of judgement.

And Meson is the other part of this -- if Meson is coming soon-ish
and will mean users having to change their build commands in some
way anyway, it's better for them if we only make them change once,
when Meson arrives, rather than once now and then again later.

> If we can't reach consensus in time for 5.0, that's regrettable, but I
> accept it.  Our decision making process is open and slow.  Hard to get
> one without the other.
>
> Much harder to accept is us once again defaulting to do nothing because
> deciding what to do involves a tradeoff.

My understanding of the consensus position was "we should stop
supporting in-tree build in the main makefile machinery but should
have a trivial wrapper that creates and uses a build directory
under the hood if the user does run configure/make from the
source tree". I think that should be much less painful to maintain
than handling both setups through the whole of our makefile system.
(I wouldn't personally bother to implement it, but several people
volunteered to do so.)

thanks
-- PMM