Re: Early feedback on Guix Home

[Ludovic Courtès]

Hi,

(Sorry for the late reply…)

Andrew Tropin <andrew@trop.in> skribis:

> Ludovic Courtès <ludo@gnu.org> writes:

[...]

>>>> Possible action:
>>>>
>>>>   1. Change config records to accept file-like objects instead of
>>>>      strings.  That way, users can choose to have snippets inlined (in a
>>>>      ‘plain-file’ object) or separate (via ‘local-file’).  See for
>>>>      example how ‘tor-configuration->torrc’ does it.
>>>
>>> Yeah, there is a ‘slurp-file-gexp’ procedure that let’s one read an
>>> extenal file, but using existing APIs like ‘local-file’ is probably a
>>> better idea.
>>
>> Yes, it feels more natural.  Also, ‘slurp-file-gexp’ returns a gexp (a
>> code snippet), but as a user you don’t know where that snippet is going
>> to be inserted; it may not work in some contexts.
>
> Acually, the idea behind `slurp-file-gexp` is that you always know where
> it will be inserted, because it will be inside the specific section of
> the configuration.  Take a look at emacs home-service example:
> https://git.sr.ht/~abcdw/rde/tree/master/item/gnu/home-services/emacs.scm#L91

What I mean is that, in a general sense, one cannot know whether the
gexp will be inserted in a place where it’s “valid”.  Consider the
following examples, where the gexp is meant to be inserted in lieu of
“PLACEHOLDER”:

  #~(list '(#$PLACEHOLDER))  ;it’s quoted

  #~(let ((call-with-input-file (const #f)))
      #$PLACEHOLDER)

These are “hygiene” problems discussed in the “Code Staging in GNU Guix”
paper.

All this to say that, from an API viewpoint, I think it’s (1) more
robust, as I wrote, and (2) clearer to expect file-like objects in such
places.  It’s clearer because users can be expected to have an
understanding of what ‘local-file’ does, whereas ‘slurp-file-gexp’ is
more involved.

I hope this clarifies what I had in mind!

Ludo’.