`map-derivation' is too slow for meaningful usage

`map-derivation' is too slow for meaningful usage

[Sergio Pastor Pérez]

Hello!

Today I've spent some time understanding the issue with the
`map-derivation` procedure[1]. I've submitted a fix to the issue I
previously opened.

As I describe in the answer, the procedure seems too slow for any
meaningful usage; specially what I was describing in this thread[2].

What do you think?

[1] https://issues.guix.gnu.org/71941
[2] https://lists.gnu.org/archive/html/guix-devel/2024-06/msg00275.html


Regards,
Sergio

Re: `map-derivation' is too slow for meaningful usage

[Ludovic Courtès]

Hi Sergio,

Sergio Pastor Pérez <sergio.pastorperez@outlook.es> skribis:

> Today I've spent some time understanding the issue with the
> `map-derivation` procedure[1]. I've submitted a fix to the issue I
> previously opened.
>
> As I describe in the answer, the procedure seems too slow for any
> meaningful usage; specially what I was describing in this thread[2].
>
> What do you think?

I haven’t taken the time to read that earlier thread and the patch you
submitted yet, sorry.

‘map-derivation’ as you noticed is not used anywhere so far and remains
kind of an experiment.  It predates package rewriting and package
transformations.  The idea was to test the feasibility of a sledgehammer
approach to dependency rewriting; from what you write, I gather there’s
room for improvement.

I prefer package transformations because they work at a higher level,
which make them easier to use.

But I’m also well aware that there are circumstances where they’re not
really applicable, such as when trying to build a variant of an entire
operating system.  I’m not sure whether operating at the derivation
level is the right approach, so your tests here are much welcome.

Perhaps we could try and address those performance issues, and then see
if we could come up maybe with a “derivation transformation” framework
to make it more approachable?

Thanks,
Ludo’.

Re: `map-derivation' is too slow for meaningful usage

[Sergio Pastor Pérez]

Hello, Ludo.

On 06-09-2024 12:17, Ludovic Courtès wrote:
> Perhaps we could try and address those performance issues, and then see
> if we could come up maybe with a “derivation transformation” framework
> to make it more approachable?

That sound good to me. Do you have any idea on what could be the bottleneck?

Regards,

Sergio.

Re: `map-derivation' is too slow for meaningful usage

[Ludovic Courtès]

Hi,

Sergio Pastor Pérez <sergio.pastorperez@outlook.es> skribis:

> On 06-09-2024 12:17, Ludovic Courtès wrote:
>> Perhaps we could try and address those performance issues, and then see
>> if we could come up maybe with a “derivation transformation” framework
>> to make it more approachable?
>
> That sound good to me. Do you have any idea on what could be the bottleneck?

No idea, but the implementation is likely naïve: I wrote once, tested it
on simple cases, but it never got any use beyond that.

I would start by profiling it with (statprof), and then also look at
caching issues (see the ‘GUIX_PROFILING’ environment variable).

HTH!

Ludo’.