package-input-rewriting causes different ordering of grafts

  • Open
  • quality assurance status badge
Details
3 participants
  • Akira Kyle
  • Josselin Poiret
  • Ludovic Courtès
Owner
unassigned
Submitted by
Akira Kyle
Severity
important
A
A
Akira Kyle wrote on 29 Jan 2023 16:59
(address . bug-guix@gnu.org)
875ycox1l8.fsf@akirakyle.com
I think this may be a strange, subtle bug and hopefully this is
sufficient to reproduce.
This was all done with guix commit
da9af8c72fe427e3eeb718b53c599139a8958d61.

Consider the following:

(use-modules (guix packages)
(guix gexp)
(guix monads)
(guix store)
(guix git)
(guix download)
(guix git-download)
(gnu packages compression)
(gnu packages emacs-xyz))

(define (package-commit pkg commit checksum)
"Return a package variant using the given commit and sha256."
(package
(inherit pkg)
(name (package-name pkg))
(version (substring commit 0 7))
(source
(origin
(method git-fetch)
(uri (git-reference
(url (git-checkout-url (git-reference->git-checkout
(origin-uri (package-source
pkg)))))
(commit commit)))
(sha256 (base32 checksum))
(file-name (git-file-name name version))))))

(define emacs-magit-latest
(package-commit emacs-magit
"45be64e5f1ef0d03156c53543eb198829177a934"
"19b1d4200x1rhffmxf834kh55nj2zjps09hkkhw9hrsdig3nlfl9"))

(define emacs-compat-latest
(package
(inherit emacs-compat)
(version "29.1.3.0")
(source (origin
(method url-fetch)
(uri (string-append "https://elpa.gnu.org/packages/"
"compat-" version ".tar.lz"))
(sha256
(base32
"0jnk81rg5w6zhf413nf3j72i2mr8vfdms55gahrdx28727w8gfj8"))))
(native-inputs (modify-inputs (package-native-inputs
emacs-compat)
(prepend lzip)))))

(define with-replacements
(package-input-rewriting `((,emacs-compat
. ,emacs-compat-latest))))


(list
emacs-magit-latest
(with-replacements emacs-magit-latest)
)

emacs-magit-latest depends on
/gnu/store/xrwyv6s8gsazgb2hqxxvh3h6d9cvkxal-git-2.39.1.drv

whereas (with-replacements emacs-magit-latest)` depends on
/gnu/store/x2gd1dsqvpas4rwvlxdmj7p9w2d2ba49-git-2.39.1.drv

Which is surprising since I didn't change anything about the git
package.

Following this a bit further reveals that the difference comes
down to the grafts being applied to python where
emacs-magit-latest depends on
/gnu/store/848hfdz4xj91jail0y8lfj22dkgb3d12-python-3.9.9-builder


whereas (with-replacements emacs-magit-latest)` depends on
/gnu/store/f3b6wlkwnnp2zfm20br6hbc9i0pipi24-python-3.9.9-builder


Inspecting these derivations, the difference between them seems to
be that in the former has zlib first in the `mapping` field of the
graft while in the latter zlib is last. My guess is this is due to
the fact that the emacs-compat override adds lzip to its
native-inputs and causes a different traversal order for grafting
zlib. But this is my first time attempting to look at the grafting
machinery in detail, so hopefully the experts can figure out if
this really is a bug and how to fix it!

I don't have a concrete example where this causes issues in
practice, since here the two packages would conflict anyways due
to different compat versions, but it seems like that's mostly a
consequence of emacs-packages having propagated dependencies. At
the very least it would seem to cause unnecessary duplication in
the store.
L
L
Ludovic Courtès wrote on 30 Jan 2023 08:24
control message for bug #61161
(address . control@debbugs.gnu.org)
87mt60yoqx.fsf@gnu.org
severity 61161 important
quit
J
J
Josselin Poiret wrote on 1 Feb 2023 09:23
Re: bug#61161: package-input-rewriting causes different ordering of grafts
87a61xb8qe.fsf@jpoiret.xyz
Hi Akira,

Akira Kyle <akira@akirakyle.com> writes:

Toggle quote (16 lines)
> Inspecting these derivations, the difference between them seems to
> be that in the former has zlib first in the `mapping` field of the
> graft while in the latter zlib is last. My guess is this is due to
> the fact that the emacs-compat override adds lzip to its
> native-inputs and causes a different traversal order for grafting
> zlib. But this is my first time attempting to look at the grafting
> machinery in detail, so hopefully the experts can figure out if
> this really is a bug and how to fix it!
>
> I don't have a concrete example where this causes issues in
> practice, since here the two packages would conflict anyways due
> to different compat versions, but it seems like that's mostly a
> consequence of emacs-packages having propagated dependencies. At
> the very least it would seem to cause unnecessary duplication in
> the store.

This makes me think of [1], which was supposedly fixed, although from
what I remember we didn't actually find out what was causing this.


Best,
--
Josselin Poiret
?