bug#45200: Memory leaks: (garbage-collect) fails to reclaim memory

[Konstantin Kharlamov <hi-angel@yandex.ru>]

On Sat, 2020-12-12 at 22:15 +0200, Eli Zaretskii wrote:
> > From: Konstantin Kharlamov <hi-angel@yandex.ru>
> > Date: Sat, 12 Dec 2020 21:43:10 +0300
> >
> > # Steps to reproduce:
> >
> > 1. Run `mkdir /tmp/.emacs.d`
> > 2. Run emacs as `HOME=/tmp/ emacs`, and measure its PSS
> > 3. Create a file /tmp/.emacs.d/early-init.el with content:
> >
> >     ;; only run garbage collection on idle
> >     (setq gc-cons-threshold most-positive-fixnum)
> >     (run-with-idle-timer 2 t (lambda () (garbage-collect)))
> >
> > 4. Run emacs as `HOME=/tmp/ emacs`, evaluate (garbage-collect), then measure
> > its PSS
> >
> > ## Expected
> >
> > Size has no statistically-significant difference, because in both cases we
> > garbage-collected memory.
> >
> > ## Actual
> >
> > Size without calling explicit garbage-collect, from 3 runs, varied around
> > 41.2..41.7 MB.
> >
> > Size afterwards, also 3 runs, varied around 45.4..45.5 MB.
>
> Could be simply the effect of different stack size, since Emacs's GC
> is conservative, and when there's doubt whether something is a live
> object, it won't GC it.
>
> I think more specific and detailed evidence is needed to prove your
> case: which objects were not GC'ed and why.

Alright, fair enough. I crafted up another testcase, it may be better. The following code first temporarily disables GC, then it prints "hello" 1000000 times, and finally it calls GC manually.

I call `emacs -Q`, then measure PSS, then evaluate the code below, then again measure PSS.

    (let ((i 1000000))
      (setq gc-cons-threshold most-positive-fixnum)
      (while (> i 0)
        (print "hello")
        (setq i (- i 1)))
      (garbage-collect))

The loop takes 20-30 seconds for me, I think. PSS before is ≈41M, and PSS after is 266.3M. That is ≈200M of memory just vanished.

Regarding, whether it is stack size:

     λ grep VmStk /proc/283047/status
    VmStk:       132 kB

Apparently, it is not stack size.