Hang in make check

Hang in make check

[Jeffrey Walton]

Hi Everyone,

I'm testing Emacs 27.2 on both Linux and Apple. 'make check' is
hanging somewhere around this test:

HOME=/nonexistent EMACSLOADPATH= LC_ALL=C
EMACS_TEST_DIRECTORY=/home/jwalton/Build-Scripts/emacs-27.2/test
"../src/emacs"  --no-init-file --no-site-file --no-site-lisp -L ":."
\
  -l ert  -l lisp/autorevert-tests \
  --batch --eval '(ert-run-tests-batch-and-exit (quote (not (or (tag
:expensive-test) (tag :unstable)))))' > lisp/autorevert-tests.log 2>&1
|| { STAT=$?; cat lisp/autorevert-tests.log; exit $STAT; }
<hang here>

CTRL+C is needed to break into the test.

Jeff

Re: Hang in make check

[Michael Albinus]

Jeffrey Walton <noloader@gmail.com> writes:

> Hi Everyone,

Hi Jeff,

> I'm testing Emacs 27.2 on both Linux and Apple. 'make check' is
> hanging somewhere around this test:
>
> HOME=/nonexistent EMACSLOADPATH= LC_ALL=C
> EMACS_TEST_DIRECTORY=/home/jwalton/Build-Scripts/emacs-27.2/test
> "../src/emacs"  --no-init-file --no-site-file --no-site-lisp -L ":."
> \
>   -l ert  -l lisp/autorevert-tests \
>   --batch --eval '(ert-run-tests-batch-and-exit (quote (not (or (tag
> :expensive-test) (tag :unstable)))))' > lisp/autorevert-tests.log 2>&1
> || { STAT=$?; cat lisp/autorevert-tests.log; exit $STAT; }
> <hang here>
>
> CTRL+C is needed to break into the test.

Does there exist the file test/lisp/autorevert-tests.log? Could you pls
show it?

Is Emacs compiled with file notifications? Which one? You can determine
it with

--8<---------------cut here---------------start------------->8---
# emacs -Q --batch -l filenotify --eval '(message "%s" file-notify--library)'
--8<---------------cut here---------------end--------------->8---

> Jeff

Best regards, Michael.

Re: Hang in make check

[Jeffrey Walton]

[-- Attachment #1: Type: text/plain, Size: 1540 bytes --]

On Sun, Mar 28, 2021 at 4:27 AM Michael Albinus <michael.albinus@gmx.de> wrote:
>
> Jeffrey Walton <noloader@gmail.com> writes:
>
> > I'm testing Emacs 27.2 on both Linux and Apple. 'make check' is
> > hanging somewhere around this test:
> >
> > HOME=/nonexistent EMACSLOADPATH= LC_ALL=C
> > EMACS_TEST_DIRECTORY=/home/jwalton/Build-Scripts/emacs-27.2/test
> > "../src/emacs"  --no-init-file --no-site-file --no-site-lisp -L ":."
> > \
> >   -l ert  -l lisp/autorevert-tests \
> >   --batch --eval '(ert-run-tests-batch-and-exit (quote (not (or (tag
> > :expensive-test) (tag :unstable)))))' > lisp/autorevert-tests.log 2>&1
> > || { STAT=$?; cat lisp/autorevert-tests.log; exit $STAT; }
> > <hang here>
> >
> > CTRL+C is needed to break into the test.
>
> Does there exist the file test/lisp/autorevert-tests.log? Could you pls
> show it?

I'm on Linux and experiencing the hang.

$ cat test/lisp/autorevert-tests.log
Running 3 tests (2021-03-28 04:55:00-0400, selector `(not (or (tag
:expensive-test) (tag :unstable)))')

> Is Emacs compiled with file notifications? Which one? You can determine
> it with
>
> --8<---------------cut here---------------start------------->8---
> # emacs -Q --batch -l filenotify --eval '(message "%s" file-notify--library)'
> --8<---------------cut here---------------end--------------->8---

It appears there are no notifications:

emacs-27.2$ src/emacs -Q --batch -l filenotify --eval '(message "%s"
file-notify--library)'
nil

I also attached config.log. I was trying to configure an 'emacs-nox'.

Jeff

[-- Attachment #2: emacs-config.log.zip --]
[-- Type: application/zip, Size: 45191 bytes --]

Re: Hang in make check

[Michael Albinus]

[-- Attachment #1: Type: text/plain, Size: 593 bytes --]

Jeffrey Walton <noloader@gmail.com> writes:

Hi Jeff,

> It was created by GNU Emacs configure 27.2, which was
> generated by GNU Autoconf 2.69.  Invocation command line was
>
>   $ ./configure --build=x86_64-pc-linux-gnu --prefix=/usr/local --libdir=/usr/local/lib --without-all --with-xml2 --with-json --with-libgmp --with-zlib --without-gnutls --with-selinux

With this configuration, I could reproduce the problem. The appended
patch fixes this for me, could you pls check?

(There's no need to panic about Emacs 27.2, the patch is in
autorevert-tests.el)

> Jeff

Best regards, Michael.


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: Type: text/x-patch, Size: 932 bytes --]

*** /tmp/ediffv7q1Ft	2021-03-28 14:10:16.784777579 +0200
--- /home/albinus/src/emacs-27/test/lisp/autorevert-tests.el	2021-03-28 14:07:52.481983870 +0200
***************
*** 131,137 ****
                         (format-message
                          "Reverting buffer `%s'\\." (buffer-name buffer))
                         auto-revert--messages)))
!       (if (with-current-buffer buffer auto-revert-use-notify)
            (read-event nil nil 0.1)
          (sleep-for 0.1)))))

--- 131,139 ----
                         (format-message
                          "Reverting buffer `%s'\\." (buffer-name buffer))
                         auto-revert--messages)))
!       (if (and (or file-notify--library
!                    (file-remote-p temporary-file-directory))
!                (with-current-buffer buffer auto-revert-use-notify))
            (read-event nil nil 0.1)
          (sleep-for 0.1)))))

Re: Hang in make check

[Jeffrey Walton]

On Sun, Mar 28, 2021 at 8:14 AM Michael Albinus <michael.albinus@gmx.de> wrote:
>
> Jeffrey Walton <noloader@gmail.com> writes:
>
> Hi Jeff,
>
> > It was created by GNU Emacs configure 27.2, which was
> > generated by GNU Autoconf 2.69.  Invocation command line was
> >
> >   $ ./configure --build=x86_64-pc-linux-gnu --prefix=/usr/local --libdir=/usr/local/lib --without-all --with-xml2 --with-json --with-libgmp --with-zlib --without-gnutls --with-selinux
>
> With this configuration, I could reproduce the problem. The appended
> patch fixes this for me, could you pls check?
>
> (There's no need to panic about Emacs 27.2, the patch is in
> autorevert-tests.el)

Thanks Michael. The patch clears the hang.

Here's one more hang. It happens on both Linux and Apple over SSH. (I
SSH into most of my machines).

  --batch --eval '(ert-run-tests-batch-and-exit (quote (not (or (tag
:expensive-test) (tag :unstable)))))' > src/keyboard-tests.log 2>&1 ||
{ STAT=$?; cat src/keyboard-tests.log; exit $STAT; }

Jeff

Re: Hang in make check

[Michael Albinus]

Jeffrey Walton <noloader@gmail.com> writes:

Hi Jeff,

>> With this configuration, I could reproduce the problem. The appended
>> patch fixes this for me, could you pls check?
>
> Thanks Michael. The patch clears the hang.

Thanks for the confirmation. I've pushed the patch to the emacs-27
branch; will appear with Emacs 27.3 (if this happens to be released) or 28.1.

> Here's one more hang. It happens on both Linux and Apple over SSH. (I
> SSH into most of my machines).
>
>   --batch --eval '(ert-run-tests-batch-and-exit (quote (not (or (tag
> :expensive-test) (tag :unstable)))))' > src/keyboard-tests.log 2>&1 ||
> { STAT=$?; cat src/keyboard-tests.log; exit $STAT; }

I'll take a look on it tomorrow. I cannot promise too much 'cos this is
not my area; hopefully somebody beats me in fixing this.

> Jeff

Best regards, Michael.

Re: Hang in make check

[Lars Ingebrigtsen]

Jeffrey Walton <noloader@gmail.com> writes:

> Here's one more hang. It happens on both Linux and Apple over SSH. (I
> SSH into most of my machines).
>
>   --batch --eval '(ert-run-tests-batch-and-exit (quote (not (or (tag
> :expensive-test) (tag :unstable)))))' > src/keyboard-tests.log 2>&1 ||
> { STAT=$?; cat src/keyboard-tests.log; exit $STAT; }

If this is the same as "cd test; make keyboard-tests" I'm unable to
reproduce it on either Debian/bulleseye or Macos/catalina (via ssh):

[larsi@stories ~/src/emacs/emacs-27/test]$ make keyboard-tests
make[1]: Entering directory '/home/larsi/src/emacs/emacs-27/test'
  GEN      src/keyboard-tests.log
Running 1 tests (2021-03-28 21:19:28+0200, selector `(not (tag :unstable))')
   passed  1/1  keyboard-unread-command-events (1.001370 sec)

Ran 1 tests, 1 results as expected, 0 unexpected (2021-03-28 21:19:29+0200, 1.001741 sec)

make[1]: Leaving directory '/home/larsi/src/emacs/emacs-27/test'


-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

Re: Hang in make check

[Jeffrey Walton]

On Sun, Mar 28, 2021 at 3:20 PM Lars Ingebrigtsen <larsi@gnus.org> wrote:
>
> Jeffrey Walton <noloader@gmail.com> writes:
>
> > Here's one more hang. It happens on both Linux and Apple over SSH. (I
> > SSH into most of my machines).
> >
> >   --batch --eval '(ert-run-tests-batch-and-exit (quote (not (or (tag
> > :expensive-test) (tag :unstable)))))' > src/keyboard-tests.log 2>&1 ||
> > { STAT=$?; cat src/keyboard-tests.log; exit $STAT; }
>
> If this is the same as "cd test; make keyboard-tests" I'm unable to
> reproduce it on either Debian/bulleseye or Macos/catalina (via ssh):
>
> [larsi@stories ~/src/emacs/emacs-27/test]$ make keyboard-tests
> make[1]: Entering directory '/home/larsi/src/emacs/emacs-27/test'
>   GEN      src/keyboard-tests.log
> Running 1 tests (2021-03-28 21:19:28+0200, selector `(not (tag :unstable))')
>    passed  1/1  keyboard-unread-command-events (1.001370 sec)
>
> Ran 1 tests, 1 results as expected, 0 unexpected (2021-03-28 21:19:29+0200, 1.001741 sec)

Thanks Lars.

In case it matters, here's what I configure with:

  --without-all
  --with-xml2
  --with-json
  --with-libgmp
  --with-zlib

Jeff

Re: Hang in make check

[Lars Ingebrigtsen]

Jeffrey Walton <noloader@gmail.com> writes:

> In case it matters, here's what I configure with:
>
>   --without-all
>   --with-xml2
>   --with-json
>   --with-libgmp
>   --with-zlib

Ah; "make keyboard-tests" with those options does indeed hang (when
testing on Debian/bullseye).

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

Re: Hang in make check

[Lars Ingebrigtsen]

Lars Ingebrigtsen <larsi@gnus.org> writes:

> Ah; "make keyboard-tests" with those options does indeed hang (when
> testing on Debian/bullseye).

Seems to have something to do with timers and timeouts.  This hangs:

(read-event nil nil 1)

when run in batch mode, apparently (instead of timing out after one
second).

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

Re: Hang in make check

[Eli Zaretskii]

> From: Lars Ingebrigtsen <larsi@gnus.org>
> Date: Sun, 28 Mar 2021 21:51:29 +0200
> Cc: Emacs developers <emacs-devel@gnu.org>
> 
> Lars Ingebrigtsen <larsi@gnus.org> writes:
> 
> > Ah; "make keyboard-tests" with those options does indeed hang (when
> > testing on Debian/bullseye).
> 
> Seems to have something to do with timers and timeouts.  This hangs:
> 
> (read-event nil nil 1)
> 
> when run in batch mode, apparently (instead of timing out after one
> second).

But that's not what the test does (if it did, it would have hanged for
everyone, regardless of the build details).  The test does:

  (let ((unread-command-events nil))
    (should (equal (progn (push ?\C-a unread-command-events)
                          (read-event nil nil 1))
                   ?\C-a))

When there's some stuff in unread-command-events, read-event should
return immediately with the first element of what's in
unread-command-events.  Why doesn't it happen in that build?  Can you
step through the code and tell why not?

Re: Hang in make check

[Lars Ingebrigtsen]

Eli Zaretskii <eliz@gnu.org> writes:

>> Seems to have something to do with timers and timeouts.  This hangs:
>> 
>> (read-event nil nil 1)
>> 
>> when run in batch mode, apparently (instead of timing out after one
>> second).
>
> But that's not what the test does (if it did, it would have hanged for
> everyone, regardless of the build details).

It's not what the test does -- I was stating that this form hangs (when
in batch mode in this Emacs configuration).

I have not debugged further.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

Re: Hang in make check

[Eli Zaretskii]

> From: Lars Ingebrigtsen <larsi@gnus.org>
> Cc: noloader@gmail.com,  emacs-devel@gnu.org
> Date: Mon, 29 Mar 2021 13:00:39 +0200
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> >> Seems to have something to do with timers and timeouts.  This hangs:
> >> 
> >> (read-event nil nil 1)
> >> 
> >> when run in batch mode, apparently (instead of timing out after one
> >> second).
> >
> > But that's not what the test does (if it did, it would have hanged for
> > everyone, regardless of the build details).
> 
> It's not what the test does -- I was stating that this form hangs (when
> in batch mode in this Emacs configuration).

Sorry for my misunderstanding.

Re: Hang in make check

[Eli Zaretskii]

> Date: Mon, 29 Mar 2021 14:04:47 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> Cc: noloader@gmail.com, emacs-devel@gnu.org
> 
> > >> Seems to have something to do with timers and timeouts.  This hangs:
> > >> 
> > >> (read-event nil nil 1)
> > >> 
> > >> when run in batch mode, apparently (instead of timing out after one
> > >> second).
> > >
> > > But that's not what the test does (if it did, it would have hanged for
> > > everyone, regardless of the build details).
> > 
> > It's not what the test does -- I was stating that this form hangs (when
> > in batch mode in this Emacs configuration).
> 
> Sorry for my misunderstanding.

I think the place where read-event hangs in that configuration is in
this fragment from keyboard.c:kbd_buffer_get_event:

 #if !defined HAVE_DBUS && !defined USE_FILE_NOTIFY && !defined THREADS_ENABLED
   if (noninteractive
       /* In case we are running as a daemon, only do this before
	  detaching from the terminal.  */
       || (IS_DAEMON && DAEMON_RUNNING))
     {
       int c = getchar ();  <<<<<<<<<<<<<<<<<<<<<<<<<<
       XSETINT (obj, c);
       *kbp = current_kboard;
       return obj;
     }
 #endif	/* !defined HAVE_DBUS && !defined USE_FILE_NOTIFY && !defined THREADS_ENABLED  */

But we still need to understand why this happens (if that's what
happens) in the real test, because AFAIU when unread-command-events is
non-nil, we were not supposed to get to this place, but instead return
the event in unread-command-events.

Re: Hang in make check

[Lars Ingebrigtsen]

Eli Zaretskii <eliz@gnu.org> writes:

> I think the place where read-event hangs in that configuration is in
> this fragment from keyboard.c:kbd_buffer_get_event:

Yup; looks likely.

> But we still need to understand why this happens (if that's what
> happens) in the real test, because AFAIU when unread-command-events is
> non-nil, we were not supposed to get to this place, but instead return
> the event in unread-command-events.

The test code snippet that's actually hanging is this:

(should (equal (progn (run-with-timer
                       1 nil
                       (lambda () (push '(t . ?\C-b) unread-command-events)))
                      (read-event nil nil 2))
               ?\C-b))

So the `read-event' hangs indefinitely, and then apparently never
discovers that the run-with-timer has pushed something unto
unread-command-events (because it's in that getchar?), I think.

I haven't had time to actually try to debug this; hopefully I'll find
some more time later this week...

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

Re: Hang in make check

[Eli Zaretskii]

> From: Lars Ingebrigtsen <larsi@gnus.org>
> Cc: noloader@gmail.com,  emacs-devel@gnu.org
> Date: Tue, 30 Mar 2021 15:29:33 +0200
> 
> The test code snippet that's actually hanging is this:
> 
> (should (equal (progn (run-with-timer
>                        1 nil
>                        (lambda () (push '(t . ?\C-b) unread-command-events)))
>                       (read-event nil nil 2))
>                ?\C-b))
> 
> So the `read-event' hangs indefinitely, and then apparently never
> discovers that the run-with-timer has pushed something unto
> unread-command-events (because it's in that getchar?), I think.

Yes.  It's actually worse: _because_ the getchar call hangs, Emacs
doesn't get to running the timer function at all.

So I think the code needs to be modified, or maybe simply disabled
when system-configuration-features includes neither DBUS nor NOTIFY
nor THREADS.

Re: Hang in make check

[Andreas Schwab]

On Mär 30 2021, Eli Zaretskii wrote:

> Yes.  It's actually worse: _because_ the getchar call hangs, Emacs
> doesn't get to running the timer function at all.
>
> So I think the code needs to be modified, or maybe simply disabled
> when system-configuration-features includes neither DBUS nor NOTIFY
> nor THREADS.

Perhaps that getchar call should be removed altogether?

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

Re: Hang in make check

[Eli Zaretskii]

> From: Andreas Schwab <schwab@linux-m68k.org>
> Cc: Lars Ingebrigtsen <larsi@gnus.org>,  noloader@gmail.com,
>   emacs-devel@gnu.org
> Date: Tue, 30 Mar 2021 16:14:18 +0200
> 
> On Mär 30 2021, Eli Zaretskii wrote:
> 
> > Yes.  It's actually worse: _because_ the getchar call hangs, Emacs
> > doesn't get to running the timer function at all.
> >
> > So I think the code needs to be modified, or maybe simply disabled
> > when system-configuration-features includes neither DBUS nor NOTIFY
> > nor THREADS.
> 
> Perhaps that getchar call should be removed altogether?

How will keyboard input be received then in those configurations?

This is a corner use case -- why would someone build Emacs with all
those features disabled, and then want to run the test suite?  I'm not
sure we should bother too much about that case.

Re: Hang in make check

[Andreas Schwab]

On Mär 30 2021, Eli Zaretskii wrote:

>> From: Andreas Schwab <schwab@linux-m68k.org>
>> Cc: Lars Ingebrigtsen <larsi@gnus.org>,  noloader@gmail.com,
>>   emacs-devel@gnu.org
>> Date: Tue, 30 Mar 2021 16:14:18 +0200
>> 
>> On Mär 30 2021, Eli Zaretskii wrote:
>> 
>> > Yes.  It's actually worse: _because_ the getchar call hangs, Emacs
>> > doesn't get to running the timer function at all.
>> >
>> > So I think the code needs to be modified, or maybe simply disabled
>> > when system-configuration-features includes neither DBUS nor NOTIFY
>> > nor THREADS.
>> 
>> Perhaps that getchar call should be removed altogether?
>
> How will keyboard input be received then in those configurations?

The same way as in all other configurations?  After all, the majority of
the emacs installations don't include that block of code anyway.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."

Re: Hang in make check

[Eli Zaretskii]

> From: Andreas Schwab <schwab@linux-m68k.org>
> Cc: larsi@gnus.org,  noloader@gmail.com,  emacs-devel@gnu.org
> Date: Tue, 30 Mar 2021 17:16:35 +0200
> 
> >> Perhaps that getchar call should be removed altogether?
> >
> > How will keyboard input be received then in those configurations?
> 
> The same way as in all other configurations?  After all, the majority of
> the emacs installations don't include that block of code anyway.

If tty_read_avail_input is universally supported without blocking,
maybe we should.