Re: make check produces 25 failures and hangs (bug #65176).

[Alan Mackenzie]

Hello, Jens.

On Sun, Aug 13, 2023 at 12:59:26 +0200, Jens Schmidt wrote:
> On 2023-08-12  21:09, Andrea Corallo wrote:

> > To a very quick look seems to me `comp-subr-trampoline-install' is just
> > not called as it should.  AFAIR it should be called from Ffset, I'm
> > really wondering what could have happened.

> Probably because native-comp-enable-subr-trampolines equals nil in an
> "emacs-master -Q"?

> When I compare the following snippet from emacs-master/lisp/loadup.el:

>          (when (and (featurep 'native-compile)
>                     (equal dump-mode "pdump"))
>            ;; Don't enable this before bootstrap is completed, as the
>            ;; compiler infrastructure may not be usable yet.
>            (setq comp-enable-subr-trampolines t))
>                  ^^^^

> to emacs-29:

>        (when (and (featurep 'native-compile)
>                   (equal dump-mode "pdump"))
>          ;; Don't enable this before bootstrap is completed, as the
>          ;; compiler infrastructure may not be usable yet.
>          (setq native-comp-enable-subr-trampolines t))
>                ^^^^^^^^^^^

Excellent debugging work!  Thanks!

> it seems that the android merge seems to be the culprit.  But leaving
> that to others to judge.

There was a variable called comp-enable-subr-trampolines which got
renamed to native-comp-..... on 2023-02-13 by Andrea in commit
5d0912f144.  In the feature/android branch, which was possibly branched
from main _before_ 2023-02-13, this flag was renamed back again to
comp-enable-.... on 2023-02-18 by Po in commit 4ab9fa7507a.  This likely
followed a merge from main onto the android branch.

It seems that native-comp-.... appeared to be a typo in the android
branch, so it got "corrected" to comp-enable-.... to make the software
work.  This "correction" was then forgotten about, an easy mistake to
make.

> In any case, fixing that setq in emacs-master to refer to
> `native-comp-enable-subr-trampolines' again fixes the issue, at least
> as far as my "little repro case" is concerned ... running "make check" 
> ... yep, no more hangs at least on abbrevs-test.el.

I would favour correcting that last use of comp-enable-subr-trampolines.

-- 
Alan Mackenzie (Nuremberg, Germany).