Re: $ in make's := operator (was: CVS commit: pkgsrc/mk)

[Roland Illig <roland.illig%gmx.de@localhost>]

Am 30.11.2021 um 13:43 schrieb Jonathan Perkin:
> * On 2021-11-30 at 12:33 GMT, Roland Illig wrote:
>
> > > Date: Tue, 30 Nov 2021 09:06:37 +0000
> > > From: Jonathan Perkin <jperkin%netbsd.org@localhost>
> > > To: pkgsrc-changes%NetBSD.org@localhost
> > > Subject: CVS commit: pkgsrc/mk
> > > X-Mailer: log_accum
> > >
> > > Module Name:    pkgsrc
> > > Committed By:    jperkin
> > > Date:        Tue Nov 30 09:06:37 UTC 2021
> > >
> > > Modified Files:
> > >     pkgsrc/mk: bsd.prefs.mk
> > >
> > > Log Message:
> > > mk: Don't expand OPSYS_VERSION early.
> >
> > The goal of this line was to not evaluate the command from line 92 more
> > than once.
>
> It's unclear to me whether it makes any difference, as these variables
> are already wrapped in .if !defined() earlier and set in MAKEFLAGS.
> At the very least it should be documented better, as it's not clear (at
> least to me) why we would need both constructs simultaneously.

Before removing the ':=' assignment to OS_VERSION:

~/proj/pkgsrc/devel/glib2 $ bmake clean UNAME='echo uname 1>&2; uname'
uname
uname

After removing the ':=' assignment to OS_VERSION:

===> Cleaning for glib2-2.70.1
~/proj/pkgsrc/devel/glib2 $ bmake clean UNAME='echo uname 1>&2; uname'
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
uname
===> Cleaning for glib2-2.70.1

So even though OS_VERSION is not used much, it is used in
MACHINE_PLATFORM, and there are many tests for MACHINE_PLATFORM matching
a pattern.

MAKEFLAGS applies only to nested invocations of make, but not to
repeated evaluations in the top-level make process.

There is no built-in evaluate-once behavior for make variables, although
it may be possible to construct one using the '::=' modifier. I'm not
sure yet whether that will be easy enough to understand, and devel/bmake
still has a bug in this area, see usr.bin/make/unit-tests/counter.mk.


> > > Something about NetBSD make(1), at least on 9_STABLE, doesn't like
> > > doing this,
> > > resulting in literal "$$3" being passed to awk instead of being
> > > escaped down
> > > to "$3".  The same construct works fine with pkgsrc bmake(1).
> >
> > For some reason, .MAKE.SAVE_DOLLARS defaults to true in
> > pkgsrc/devel/bmake (for backwards compatibility) and to false in NetBSD
> > as well as Simon's bmake distribution.
>
> Ah, interesting, thanks for explaining.
>
> > Why does a version number of an operating system contain a '$' at all?
>
> I don't understand the question.  They don't?

The only thing the commit does is to remove the assignment to OS_VERSION.

In the commit message you said something about $$3 and awk, so I assumed
that awk was related to OS_VERSION. The $$3 I found in bsd.prefs.mk
applies to LOWER_VARIANT_VERSION, but I don't see how they relate. Did I
misunderstand something here?

Roland