Hash table GC

Hash table GC

[Anand Tamariya]

[-- Attachment #1: Type: text/plain, Size: 125 bytes --]

Hi,
If I've created a hash table with :weakness nil, how do I release it for GC
when it is no longer needed?

Regards,
Anand

[-- Attachment #2: Type: text/html, Size: 209 bytes --]

Re: Hash table GC

[Eli Zaretskii]

> From: Anand Tamariya <atamariya@gmail.com>
> Date: Mon, 14 Oct 2024 10:02:24 +0530
> 
> If I've created a hash table with :weakness nil, how do I release it for GC when it is no longer needed?

Why are you asking specifically about hash-tables?  AFAIK, they are
not different from any other Lisp object in this regard.  Weakness
affects GC of the keys and values, not of the table itself, AFAIU.

Re: Hash table GC

[Anand Tamariya]

[-- Attachment #1: Type: text/plain, Size: 1152 bytes --]

I'm trying to use a hash table for storing parsed CSS values. I'm seeing a
degradation in performance (memory and CPU) over a period of time. Trying
to investigate how to fix the same. Below is a minimal example:

(dotimes (i 10000)
  (let* ((a (make-hash-table :test 'equal :weakness 'key-or-value))
(j 0))
    (dolist (prop '(padding margin margin-top border
                            color background background-color display
                            width height clear
                            font font-size font-style font-weight))
      (puthash j (list prop (setq j (1+ j))) a))
    (pp a)
    ))
(Emacs 28.2 on Debian)

On Mon, Oct 14, 2024 at 7:30 PM Eli Zaretskii <eliz@gnu.org> wrote:

> > From: Anand Tamariya <atamariya@gmail.com>
> > Date: Mon, 14 Oct 2024 10:02:24 +0530
> >
> > If I've created a hash table with :weakness nil, how do I release it for
> GC when it is no longer needed?
>
> Why are you asking specifically about hash-tables?  AFAIK, they are
> not different from any other Lisp object in this regard.  Weakness
> affects GC of the keys and values, not of the table itself, AFAIU.
>

[-- Attachment #2: Type: text/html, Size: 1635 bytes --]

Re: Hash table GC

[Eli Zaretskii]

> From: Anand Tamariya <atamariya@gmail.com>
> Date: Tue, 15 Oct 2024 13:18:10 +0530
> Cc: emacs-devel@gnu.org
> 
> I'm trying to use a hash table for storing parsed CSS values. I'm seeing a degradation in performance
> (memory and CPU) over a period of time. Trying to investigate how to fix the same. Below is a minimal
> example:
> 
> (dotimes (i 10000)
>   (let* ((a (make-hash-table :test 'equal :weakness 'key-or-value))
> (j 0))
>     (dolist (prop '(padding margin margin-top border
>                             color background background-color display
>                             width height clear
>                             font font-size font-style font-weight))
>       (puthash j (list prop (setq j (1+ j))) a))
>     (pp a)
>     ))

You are using 'key-or-value, which prevents the table from being GCed
of _either_ key or value cannot be GCed.  And your keys are symbols,
so they are not GCed.

So I think you should not use 'key-or-value if you want the table to
be GCed.  E.g., why not use t instead?

Re: Hash table GC

[Tassilo Horn]

Eli Zaretskii <eliz@gnu.org> writes:

>> (dotimes (i 10000)
>>   (let* ((a (make-hash-table :test 'equal :weakness 'key-or-value))
>> (j 0))
>>     (dolist (prop '(padding margin margin-top border
>>                             color background background-color display
>>                             width height clear
>>                             font font-size font-style font-weight))
>>       (puthash j (list prop (setq j (1+ j))) a))
>>     (pp a)
>>     ))
>
> You are using 'key-or-value, which prevents the table from being GCed
> of _either_ key or value cannot be GCed.  And your keys are symbols,
> so they are not GCed.

Really?  I would assume that in a table with :weakness 'key-or-value the
*entries* can be removed/GCed if there are no references to either the
entry's key or value.  But the whole table should still be GCed if
there's no reference to the table itself.  So in the above example, I'd
think that `a' is GC-able immediately after the `let*`.

Running the above many times doesn't seem to increase my Emacs memory
footprint so my statement feels right.

Bye,
  Tassilo