On 2024-05-22 17:33, Richard Sent wrote:

> Andrew Tropin <andrew@trop.in> writes:
>
>> After rewriting from car/cdr to match-lambda in v2 of this patch:
>> https://yhetil.org/guix-patches/3394b0b51f6a5a608ebcfb7a63fdc34e52fe928e.1711046203.git.richard@freakingpenguin.com/
>>
>> the format changed from pairs to lists, I didn't noticed this nuance
>> during review because the documentation still says that service should
>> be configured and extended with pairs.  Also, pairs are more
>> apropriate data type here.  And this match-lambda rewrite will break
>> downstream RDE user's setups after migrating to upstreamed version of
>> service.
>>
>> That's why I propose to go back to pairs.
>>
>
> I'm not opposed to going back to cons cell pairs. I didn't put too much
> thought in a "list of two elements" vs. "cons cell" besides the match
> statement being easier to handle with a list.
>
> Would this patch have unintended side effects? I thought the . in match
> conditions had a different behavior.
>
> --8<---------------cut here---------------start------------->8---
> scheme@(guile-user)> (match '(1 2 3 4)
>                        ((a . b) b))
> $5 = (2 3 4)
> scheme@(guile-user)> (match '(1 2)
>                        ((a . b) b))
> $6 = (2)
> scheme@(guile-user)> (match '(1 . 2)
>                        ((a . b) b))
> $7 = 2
> --8<---------------cut here---------------end--------------->8---
>
> So changing to this would allow for a home-service entry like the
> following to match:
>
> --8<---------------cut here---------------start------------->8---
> (simple-service 'my-extra-home home-service-type
>                 `(("bob" "bill" "bertha" "billyanne" ,my-extra-home ,my-home-2)))
> --8<---------------cut here---------------end--------------->8---
>
> From my testing, this /will/ error (yay), but not as soon as the match
> would with the current code. Instead of being caught before being passed
> to the daemon, it seems to be caught while lowering the invalid
> file-append object.
>
> Personally I would prefer to catch as many errors as possible before
> beginning to pass code off to the daemon where possible. Generally
> speaking it feels like pre-daemon errors are easier to mentally parse.
>
> In fairness, the current code also isn't trying particularly hard to
> check that "user" is a string and "he" is a home-environment or printing
> a fancy error message.
>
> Perhaps this would work?
>
> --8<---------------cut here---------------start------------->8---
> (define-module (gnu services guix)
>   ...
>   #:use-module (gnu home))
> ...
> (define (guix-home-shepherd-service config)
>   (map (match-lambda
>          (((? string? user) . (? home-environment? he))
>           (shepherd-service
>            ...
>            ))
>          (e (error "Invalid value for guix-home-shepherd-service: " e)))
>        config))
> --8<---------------cut here---------------end--------------->8---

This idea is good, I'll incorporate this into v2.

>
> Maybe I'm just being silly. 🙂

-- 
Best regards,
Andrew Tropin