GNU bug report logs - #77708
[PATCH] gexp: ‘with-parameters‘ is respected by caches.

Previous Next

Package: guix-patches;

Reported by: David Elsing <david.elsing <at> posteo.net>

Date: Thu, 10 Apr 2025 14:50:02 UTC

Severity: normal

Tags: patch

Done: David Elsing <david.elsing <at> posteo.net>

Bug is archived. No further changes may be made.

Full log


Message #11 received at 77708 <at> debbugs.gnu.org (full text, mbox):

From: David Elsing <david.elsing <at> posteo.net>
To: Ludovic Courtès <ludovic.courtes <at> inria.fr>
Cc: dev <at> jpoiret.xyz, zimon.toutoune <at> gmail.com, othacehe <at> gnu.org, ludo <at> gnu.org,
 me <at> tobias.gr, 77708 <at> debbugs.gnu.org, guix <at> cbaines.net
Subject: Re: [bug#77708] [PATCH] gexp: ‘with-parameters‘ is respected by caches.
Date: Tue, 24 Jun 2025 17:38:25 +0000
Hi Ludo',

Ludovic Courtès <ludovic.courtes <at> inria.fr> writes:

> Apologies for the delay.

No problem, thanks for looking at the patch!

> Overall I have two main concerns: the added complexity and the run-time
> overhead.  I would be tempted to just not pretend supporting
> <parameterized> for arbitrary parameters than to pay this price.
>
> WDYT?

I think parameterized packages would be very useful. Something like the
previous approach [1] would of course be nicer, but I didn't find any
information why it was not continued. From what I can tell,
`with-parameters' is now working except for the caches, so in my opinion
it makes sense to fix that.

> The “counter” really determines the dynamic scope of parameters in fact,
> right?  Should it be renamed to “parameter scope” or similar?

I only defined it to avoid frequently recalculating the hash of the
dynamic parameters, so the caches only need to include the single
integer in the key.

>> +(define %parameterized-counter-next-value 0)
>> +
>> +(define %parameterized-counters (make-hash-table))
>
> This is very much like the regular dynamic environment (info "(guile)
> Fluids and Dynamic States").

Yes, this just assigns a unique integer to each unique set of
parameters.

>> +;; Add %parameterized-counter to PARAMETERS and its value,
>> +;; which depends on PARAMETERS and VALUES, to PARAMETER-VALUES.
>> +(define (add-parameterized-counter parameters parameter-values)
>> +  (let* ((key `(,parameters . ,parameter-values))
>> +         (counter
>> +          (match (hash-ref %parameterized-counters key)
>
> Note that this is quite expensive: hashing traverses part of KEY, and
> then all of it must be traversed for comparison; in practice there would
> probably be few parameters though.

Do you think using a cryptographic hash (to avoid collisions) would be
better? I think we need to look at the values of the parameters and not
just at object identity in case an object was modified or for objects
like strings, right?

> The ‘scm->pointer’ trick prevents the expensive <package> structure
> traversal, but I’m unsure about the performance of switching to ‘equal?’
> instead of ‘eq?’.

> I wonder about the performance impact of the extra lookup key.

I did some benchmarking with 10 runs of 'guix build --no-grafts libreoffice -d':

- Previously:
  3.684 s ± 0.019 s
- Using `scm->pointer' and `hash-set' and `hash-ref' in `cache!':
  3.712 s ± 0.027 s
- Using an additional key in `package->derivation':
  3.697 s ± 0.028 s
- With the full path applied:
  3.726 s ± 0.032 s

So there indeed seems to be a perfomance impact of the order of ~1% in
this case.

> Does the ‘scm->pointer’ trick work with weak-key hash tables like
> ‘%bag-cache’?  I think it does because ‘scm->pointer’ records the link
> between the pointer object and the package object in a weak hash table.
> But it’s still quite bumpy.

Oh that's a good point, is chaining weak hash tables a problem? I
couldn't think of another way to add the integer describing the
parameters to the key though.

Also problematic is the `%parameterized-counter' hash table, as it keeps
the keys (i.e. all parameters passed to `with-parameters') alive
forever, so I think it should also be a weak hash table, right? If the
parameters are dropped and later constructed again, the derivation would
just be recalculated, like for `%bag-cache'.

Cheers,

David

[1] https://guix.gnu.org/en/blog/2023/parameterized-packages-for-gnu-guix/




This bug report was last modified 7 days ago.

Previous Next


GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.