Re: Failing to GC killed buffers considered harmful

[Daniel Colascione <dancol@dancol.org>]

On 3/29/20 12:07 PM, Pip Cet wrote:
> On Sun, Mar 29, 2020 at 6:48 PM Eli Zaretskii <eliz@gnu.org> wrote:
>>> From: Pip Cet <pipcet@gmail.com>
>>> Date: Sun, 29 Mar 2020 16:46:03 +0000
>>> Cc: Daniel Colascione <dancol@dancol.org>, emacs-devel@gnu.org
>>>
>>>> We could, of course, remove the assertion, but then we are left with a
>>>> dead buffer that will never be GC'ed, AFAIU, because objects that come
>>>> from the portable dump are considered constantly marked.
>>>
>>> The same is true of all other objects, though, right?
>>
>> You mean, that they are considered marked?  Yes.
> 
> Well, if my initial experiments are correct, we correctly remove the
> killed buffer from the all_buffers list; we simply fail to free the
> memory it occupied. That seems like an acceptable loss to me.
> 
>>> It seems like it's simply a bad assert to me.
>>
>> Which one?
> 
> The one in buffer.c:
> 
>        /* Only buffers with allocated buffer text should be present at
>       this point in temacs.  */
>        FOR_EACH_BUFFER (b)
>          {
>        eassert (b->text->beg != NULL);
>      }
> 
>>> The problem appears to be that "all_buffers" points to a killed
>>> buffer, and is remembered by pdumper.
>>
>> Yes, because pdumper dumps all the Lisp objects present at that point,
>> after performing GC.  What else can it do?
> 
> My understanding is that pdumper only dumps Lisp objects reachable
> from the heap (not the stack). Buffers are an exception only because
> of all_buffers, which we could rebuild from the dumped buffers.
> 
> The call to garbage_collect in Fdump_emacs_portable doesn't actually
> reduce image size.

It runs finalizers, which can free additional memory.