bug#73599: packages from store paths don't propagate propagated-inputs

bug#73599: packages from store paths don't propagate propagated-inputs

[MSavoritias]

With a package definition adding a propagated input.

Using guix build to build a package like so:

guix build --rounds=3 --no-substitutes --no-grafts --keep-failed 
--load-path=. <the package>


and then taking the store path at the end for example:

phase `compress-documentation' succeeded after 0.0 seconds
successfully built /gnu/store/vcj7bq0igj2354m5pqk9zf1x0cnwgjzd-po4a-0.73.drv
/gnu/store/j37fygmv7p69nmw0md916acinc8c8m4y-po4a-0.73


and installing it in the profile with

guix package --install /gnu/store/j37fygmv7p69nmw0md916acinc8c8m4y-po4a-0.73


will not install the propagated input into the profile. It will only 
have po4a installed.

$ guix package --install 
/gnu/store/j37fygmv7p69nmw0md916acinc8c8m4y-po4a-0.73
The following package will be installed:
    po4a 0.73

The following derivation will be built:
   /gnu/store/mc57jjaqlrmp0dw8sy2vlcv3c5giblk9-profile.drv

building CA certificate bundle...
listing Emacs sub-directories...
building fonts directory...
building directory of Info manuals...
building profile with 1 package...
$ guix package --list-installed
po4a    0.73    out /gnu/store/j37fygmv7p69nmw0md916acinc8c8m4y-po4a-0.73
$


;; This file was automatically generated and is for internal use only.
;; It cannot be passed to the '--manifest' option.
;; Run 'guix package --export-manifest' if you want to export a file
;; suitable for '--manifest'.

(manifest
   (version 4)
   (packages
     (("po4a"
       "0.73"
       "out"
       "/gnu/store/j37fygmv7p69nmw0md916acinc8c8m4y-po4a-0.73"))))

bug#73599: packages from store paths don't propagate propagated-inputs

[Tobias Geerinckx-Rice via Bug reports for GNU Guix]

Hi [explicitly CC'ing Ludo', I hope that's OK],

I don't think

  guix install $(guix build foo)

is, or is expected to be, a supported way to install packages?  Packages have more metadata attached to them than store items.  I don't think that propagation is recorded in the Nix database.  Nor do I think it should be just to support this hack.

The manual doesn't explicitly say that it's semi-supported, which could be improved, but I think it's quietly understood...

It happens to work most of the time, it can be nice to have (I install most firefox versions this way), we won't deliberately break the way it happens to work, but it's not a synonym of

  guix install foo

Kind regards,

T G-R

Sent on the go.  Excuse or enjoy my brevity.

bug#73599: packages from store paths don't propagate propagated-inputs

[MSavoritias]

[-- Attachment #1: Type: text/plain, Size: 1476 bytes --]


Tobias Geerinckx-Rice kirjoitti 2.10.2024 klo 21.22:
> Hi [explicitly CC'ing Ludo', I hope that's OK],
>
> I don't think
>
>    guix install $(guix build foo)
>
> is, or is expected to be, a supported way to install packages?  Packages have more metadata attached to them than store items.  I don't think that propagation is recorded in the Nix database.  Nor do I think it should be just to support this hack.

It actually is mentioned explicitly in the manual. 
https://guix.gnu.org/manual/devel/en/html_node/Invoking-guix-package.html

from the link above:

Alternatively, a package can directly specify a store file name such as 
/gnu/store/...-guile-3.0.7, as produced by, e.g., |guix build|.

Sometimes packages have /propagated inputs/: these are dependencies that 
automatically get installed along with the required package (see 
|propagated-inputs| in |package| objects 
<https://guix.gnu.org/manual/devel/en/html_node/package-Reference.html#package_002dpropagated_002dinputs>, 
for information about propagated inputs in package definitions).


There was an investigation yesterday at the xmpp room and that is how we 
discovered that the documentation has a bug. If we do not want to 
support it then this is just a documentation bug.

Personally i could go either way. I already updated the project 
documentation I have to reflect not to do this 
https://codeberg.org/Guix_Bechamel/collective/wiki/Updating-Packages#testing-the-package-at-runtime

MSavoritias

[-- Attachment #2: Type: text/html, Size: 2597 bytes --]

bug#73599: packages from store paths don't propagate propagated-inputs

[MSavoritias]

[-- Attachment #1: Type: text/plain, Size: 3142 bytes --]


MSavoritias kirjoitti 3.10.2024 klo 10.39:
>
>
> Tobias Geerinckx-Rice kirjoitti 2.10.2024 klo 21.22:
>> Hi [explicitly CC'ing Ludo', I hope that's OK],
>>
>> I don't think
>>
>>    guix install $(guix build foo)
>>
>> is, or is expected to be, a supported way to install packages?  Packages have more metadata attached to them than store items.  I don't think that propagation is recorded in the Nix database.  Nor do I think it should be just to support this hack.
>
> It actually is mentioned explicitly in the manual. 
> https://guix.gnu.org/manual/devel/en/html_node/Invoking-guix-package.html
>
> from the link above:
>
> Alternatively, a package can directly specify a store file name such 
> as /gnu/store/...-guile-3.0.7, as produced by, e.g., |guix build|.
>
> Sometimes packages have /propagated inputs/: these are dependencies 
> that automatically get installed along with the required package (see 
> |propagated-inputs| in |package| objects 
> <https://guix.gnu.org/manual/devel/en/html_node/package-Reference.html#package_002dpropagated_002dinputs>, 
> for information about propagated inputs in package definitions).
>
>
> There was an investigation yesterday at the xmpp room and that is how 
> we discovered that the documentation has a bug. If we do not want to 
> support it then this is just a documentation bug.
>
> Personally i could go either way. I already updated the project 
> documentation I have to reflect not to do this 
> https://codeberg.org/Guix_Bechamel/collective/wiki/Updating-Packages#testing-the-package-at-runtime
>
> MSavoritias
>

Sorry for the second reply just wanted to add some additional things here.

 > Packages have more metadata attached to them than store items.

I was very surprised to discover this yesterday when somebody pointed it 
out. The manual does not mention this in the store section. Is this 
because what you want is a nar that is retrieved only when doing guix 
export/import? at least that is my guess doing a search for metadata in 
the guix manual.

Aside from that as a separate issue of concern

it was mentioned that the Gnu Guix channel has solved this by doing the 
whole pre-inst-env architecture thing but i found little documentation 
to go on in the manual aside from "keeping things separate". There is 
also no mention of why it is needed, why these tools where picked, how 
the architecture is arranged and interacts or what each tool actually 
does. I tried to read the source files but there is not enough comments 
to go on. I assume it is expected to know pathing, unix, autotools, bash 
scripting among others but that is not mentioned and shouldn't be needed 
imo.

Some clarifications of this would be nice but as I said these are out of 
scope of this bug report probably.

Also some guide on the Guix manual by hopefully more knowledgeable 
people on how you are supposed to test packages at runtime would be 
nice. The guide only mentions to build packages but that doesn't make 
sure that packages actually work. namely:

- po4a in the gnu channel of guix is atm broken. It misses gettext at 
runtime.

- gajim doesn't show emojis

MSavoritias

[-- Attachment #2: Type: text/html, Size: 5287 bytes --]

bug#73599: packages from store paths don't propagate propagated-inputs

[Tobias Geerinckx-Rice via Bug reports for GNU Guix]

Heyo,

On 3 October 2024 07:39:27 UTC, MSavoritias <email@msavoritias.me> wrote:
>Tobias Geerinckx-Rice kirjoitti 2.10.2024 klo 21.22:
>> is, or is expected to be, a supported way to install packages?  Packages have more metadata attached to them than store items.  I don't think that propagation is recorded in the Nix database.  Nor do I think it should be just to support this hack.
>
>It actually is mentioned explicitly in the manual. https://guix.gnu.org/manual/devel/en/html_node/Invoking-guix-package.html

I think I was misunderstood.  That is the part that needs improvement.  It may not 'explicitly' say that propagation will work in both cases, but the flow strongly implies it.

What I'm not aware of is any documentation that 'installing a store item'  is not the same as 'installing a package'.  It's a lossy transformation.

See also:

  $ guix package -I sbcl-typo
  $ guix install sbcl-typo &>/dev/null
  $ guix package -I sbcl-typo
  sbcl-typo       0.0.0-0.0e88349 out     /gnu/store/67nkzm5mw4ma39na24l9rv98gkfh3nbg-sbcl-typo-0.0.0-0.0e88349
  $ guix package -r sbcl-typo -i $(guix build sbcl-typo) &>/dev/null
  $ guix package -I sbcl-typo
  sbcl-typo-0.0.0 0.0e88349       out     /gnu/store/67nkzm5mw4ma39na24l9rv98gkfh3nbg-sbcl-typo-0.0.0-0.0e88349
  $ guix package -r sbcl-typo
  guix package: error: package 'sbcl-typo' not found in profile

Because  <https://gitlab.oit.duke.edu/duke-guix/guix/-/blob/master/guix/scripts/package.scm?ref_type=heads#L619> — I guess Savannah's down again.

Store items just aren't packages.

>Personally i could go either way.

I don't think you can, not *without* extending the database schema or otherwise adding the missing metadata back to the store item though, e.g., a hidden Scheme file, i.e., overkill.  I'd love to be proven wrong.

Otherwise, the documentation-bug way is the only realistic way.



Kind regards,

T G-R

Sent on the go.  Excuse or enjoy my brevity.

bug#73599: packages from store paths don't propagate propagated-inputs

[MSavoritias]

[-- Attachment #1: Type: text/plain, Size: 2283 bytes --]

Tobias Geerinckx-Rice kirjoitti 3.10.2024 klo 12.29:

> I think I was misunderstood.  That is the part that needs improvement.  It may not 'explicitly' say that propagation will work in both cases, but the flow strongly implies it.
Agreed.
>> Personally i could go either way.
> I don't think you can, not *without* extending the database schema or otherwise adding the missing metadata back to the store item though, e.g., a hidden Scheme file, i.e., overkill.  I'd love to be proven wrong.
>
> Otherwise, the documentation-bug way is the only realistic way.
Documentation bug sounds good to me. It would also ideally be handled 
the same with the Documentation bugs i sent in the other email. I didn't 
see it reflected back to me so adding it here. Maybe email lists are 
having problems.


 > Packages have more metadata attached to them than store items.

I was very surprised to discover this yesterday when somebody pointed it 
out. The manual does not mention this in the store section. Is this 
because what you want is a nar that is retrieved only when doing guix 
export/import? at least that is my guess doing a search for metadata in 
the guix manual.

Aside from that as a separate issue of concern

it was mentioned that the Gnu Guix channel has solved this by doing the 
whole pre-inst-env architecture thing but i found little documentation 
to go on in the manual aside from "keeping things separate". There is 
also no mention of why it is needed, why these tools where picked, how 
the architecture is arranged and interacts or what each tool actually 
does. I tried to read the source files but there is not enough comments 
to go on. I assume it is expected to know pathing, unix, autotools, bash 
scripting among others but that is not mentioned and shouldn't be needed 
imo.

Some clarifications of this would be nice but as I said these are out of 
scope of this bug report probably.

Also some guide on the Guix manual by hopefully more knowledgeable 
people on how you are supposed to test packages at runtime would be 
nice. The guide only mentions to build packages but that doesn't make 
sure that packages actually work. namely:

- po4a in the gnu channel of guix is atm broken. It misses gettext at 
runtime.

- gajim doesn't show emojis

MSavoritias

[-- Attachment #2: Type: text/html, Size: 3672 bytes --]