bug#56292: 29.0.50; Compilation failure since recent loaddefs speed up

bug#56292: 29.0.50; Compilation failure since recent loaddefs speed up

[Sean Whitton]

Hello,

When I try to byte compile bongo.el[1] with recent master I get this
strange error:

    ~/tmp/bongo.el:5341:35: Error: Wrong type argument: sequencep, 0

Line 5341 is this:

        (defun bongo-player-update-elapsed-time (player elapsed-time)
      "Set PLAYER's `elapsed-time' property to ELAPSED-TIME,
    unless PLAYER's last seek happened less than N seconds ago, where N
    is the value of PLAYER's `time-update-delay-after-seek' property."
      (let ((delay (bongo-player-get player 'time-update-delay-after-seek)))
        (when (or (null delay) (zerop delay)
                  (let ((time (bongo-player-get player 'last-seek-time)))
                    (or (null time)
                        (time-less-p (seconds-to-time delay)
  >>>> this one >>>>                 (subtract-time (current-time) time)))))
          (bongo-player-put player 'elapsed-time elapsed-time))))

which doesn't look suspicious.

Git bisection says 1d4e903417 is the first bad commit.  I did a
bootstrap build for each step of the bisection, with this script:

    git clean -xdff
    make -j4 || exit 125
    src/emacs --batch -f batch-byte-compile ~/tmp/bongo.el || exit 1

Presumably the error is coming from a type error somewhere in the
bytecompilation machinery, rather than this line in bongo.el?

[1]  https://github.com/dbrock/bongo/blob/master/bongo.el

-- 
Sean Whitton

bug#56292: 29.0.50; Compilation failure since recent loaddefs speed up

[Lars Ingebrigtsen]

Sean Whitton <spwhitton@spwhitton.name> writes:

> When I try to byte compile bongo.el[1] with recent master I get this
> strange error:
>
>     ~/tmp/bongo.el:5341:35: Error: Wrong type argument: sequencep, 0

Do you have a backtrace for this error?

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

bug#56292: 29.0.50; Compilation failure since recent loaddefs speed up

[Sean Whitton]

Hello,

On Wed 29 Jun 2022 at 12:23pm +02, Lars Ingebrigtsen wrote:

> Sean Whitton <spwhitton@spwhitton.name> writes:
>
>> When I try to byte compile bongo.el[1] with recent master I get this
>> strange error:
>>
>>     ~/tmp/bongo.el:5341:35: Error: Wrong type argument: sequencep, 0
>
> Do you have a backtrace for this error?

No, as it doesn't happen interactively.  What's the easiest way to get
one out of --batch?

-- 
Sean Whitton

bug#56292: 29.0.50; Compilation failure since recent loaddefs speed up

[Lars Ingebrigtsen]

Sean Whitton <spwhitton@spwhitton.name> writes:

>>>     ~/tmp/bongo.el:5341:35: Error: Wrong type argument: sequencep, 0
>>
>> Do you have a backtrace for this error?
>
> No, as it doesn't happen interactively.  What's the easiest way to get
> one out of --batch?

Normal errors should give you a backtrace by default in --batch, but I
guess whatever is generating the message here is catching the error.
Try perhaps (setq debug-on-signal t) and see whether that gives a
backtrace?

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

bug#56292: 29.0.50; Compilation failure since recent loaddefs speed up

[Sean Whitton]

retitle 56292 29.0.50; Problem with define-obsolete-function-alias in loaddefs.el
thanks

On Thu 30 Jun 2022 at 11:15am +02, Lars Ingebrigtsen wrote:

> Normal errors should give you a backtrace by default in --batch, but I
> guess whatever is generating the message here is catching the error.
> Try perhaps (setq debug-on-signal t) and see whether that gives a
> backtrace?

Thanks, I eventually managed to get some useful output.

After 1d4e903417, before time-date is loaded, (get 'subtract-time
'byte-obsolete-info) yields (time-subtract nil 0).  Previously, whether
or not time-date was loaded, it yielded (time-substract nil "26.1").
The wrong-type-argument is when macroexp--obsolete-warning tries to pass
0 as an argument to concat.

If you replace

    (define-obsolete-function-alias 'subtract-time 'time-subtract "\
26.1")

with

    (define-obsolete-function-alias 'subtract-time 'time-subtract "26.1")

in loaddefs.el then the problem goes away.  I hacked loadup.el and
determined that when loadup.el is loading loaddefs.el,

    (macroexpand-1
     '(define-obsolete-function-alias 'subtract-time 'time-subtract "\
26.1"))

yields

(progn (defalias 'subtract-time 'time-subtract nil)
       (make-obsolete 'subtract-time 'time-subtract 0))

which is not the correct expansion according to the definition of
define-obsolete-function-alias ..

-- 
Sean Whitton

bug#56292: 29.0.50; Compilation failure since recent loaddefs speed up

[Lars Ingebrigtsen]

Sean Whitton <spwhitton@spwhitton.name> writes:

> If you replace
>
>     (define-obsolete-function-alias 'subtract-time 'time-subtract "\
> 26.1")
>
> with
>
>     (define-obsolete-function-alias 'subtract-time 'time-subtract "26.1")
>
> in loaddefs.el then the problem goes away.

Hm.  Yes, I'll change it to format the string without the \ (because
only doc strings should be formatted that way), but this is pretty bizarre.

> I hacked loadup.el and
> determined that when loadup.el is loading loaddefs.el,
>
>     (macroexpand-1
>      '(define-obsolete-function-alias 'subtract-time 'time-subtract "\
> 26.1"))
>
> yields
>
> (progn (defalias 'subtract-time 'time-subtract nil)
>        (make-obsolete 'subtract-time 'time-subtract 0))
>
> which is not the correct expansion according to the definition of
> define-obsolete-function-alias ..

"\
26.1"

and

"26.1"

should yield an identical string after it's been read by the lisp
reader, so I don't understand how it could be making a difference here.  

Unless makedoc is involved somehow...

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

bug#56292: 29.0.50; Problem with define-obsolete-function-alias in loaddefs.el

[Lars Ingebrigtsen]

Sean Whitton <spwhitton@spwhitton.name> writes:

> The wrong-type-argument is when macroexp--obsolete-warning tries to pass
> 0 as an argument to concat.
>
> If you replace
>
>     (define-obsolete-function-alias 'subtract-time 'time-subtract "\
> 26.1")
>
> with
>
>     (define-obsolete-function-alias 'subtract-time 'time-subtract "26.1")
>
> in loaddefs.el then the problem goes away.

Aha!  This is due to this code:

  /* If purifying, and string starts with \ newline,
     return zero instead.  This is for doc strings
     that we are really going to find in etc/DOC.nn.nn.  */
  if (!NILP (Vpurify_flag) && NILP (Vdoc_file_name) && cancel)
    {
      unbind_to (count, Qnil);
      return make_fixnum (0);
    }

Which will be removed after bug#53024 is done.

But the fixes I did to loaddefs-gen in this bug report fixed this bug
report, so I'm closing it.