significance of collisions in 'guix environment'?

significance of collisions in 'guix environment'?

[Paul Garlick]

[-- Attachment #1: Type: text/plain, Size: 1156 bytes --]

Hi All,

I am in the middle of upgrading##SELECTION_END## the Gmsh package in
Guix.  I have cloned the Guix repository, set up a profile for
development, and run bootstrap, configure and make.  In setting up the
build environment for Gmsh, using 'guix environmnet ...', there are
surprising collision messages generated:

I do:

$ ./pre-inst-env guix environment --pure gmsh

I get:

.
.
.
warning: collision encountered:
/gnu/store/m6s3bz6a2sv7015z78sfzdk5ngkzdwwg-gfortran-
4.9.3/libexec/gcc/x86_64-unknown-linux-gnu/4.9.3/liblto_plugin.so.0.0.0 
/gnu/store/frrj3bfbmg5vrd0flh9cf8j64h7cr2v4-gcc-
4.9.3/libexec/gcc/x86_64-unknown-linux-gnu/4.9.3/liblto_plugin.so.0.0.0
warning: arbitrarily choosing
/gnu/store/m6s3bz6a2sv7015z78sfzdk5ngkzdwwg-gfortran-
4.9.3/libexec/gcc/x86_64-unknown-linux-gnu/4.9.3/liblto_plugin.so.0.0.0
.
.
.

There are 14 gfortran-related messages and a message regarding ld-
wrapper and binutils.

How should these messages be interpreted?  Could there be duplication
in the package inputs?  

I have noticed that Gmsh lists gfortran and lapack as propagated-inputs 
and lapack also lists gfortran as an input.

Regards,

Paul.

[-- Attachment #2: Type: text/html, Size: 1509 bytes --]

Re: significance of collisions in 'guix environment'?

[Leo Famulari]

On Tue, Nov 08, 2016 at 05:09:11PM +0000, Paul Garlick wrote:
> $ ./pre-inst-env guix environment --pure gmsh
> warning: collision encountered:
> /gnu/store/m6s3bz6a2sv7015z78sfzdk5ngkzdwwg-gfortran-
> 4.9.3/libexec/gcc/x86_64-unknown-linux-gnu/4.9.3/liblto_plugin.so.0.0.0 
> /gnu/store/frrj3bfbmg5vrd0flh9cf8j64h7cr2v4-gcc-
> 4.9.3/libexec/gcc/x86_64-unknown-linux-gnu/4.9.3/liblto_plugin.so.0.0.0
> warning: arbitrarily choosing
> /gnu/store/m6s3bz6a2sv7015z78sfzdk5ngkzdwwg-gfortran-
> 4.9.3/libexec/gcc/x86_64-unknown-linux-gnu/4.9.3/liblto_plugin.so.0.0.0

[...]

> There are 14 gfortran-related messages and a message regarding ld-
> wrapper and binutils.

The ld-wrapper collision is harmless. See http://bugs.gnu.org/22659

> How should these messages be interpreted?  Could there be duplication
> in the package inputs?  

These are filesystem paths that collide when generating a profile. A
profile is a bunch of symlinks, forming a union of various directories
in /gnu/store.

The collision warnings mean that of the store directories linked to from
your profile, more than one provide a given path.

For example, GNU Parallel and moreutils both have a path "bin/parallel":
/gnu/store/7ggbvsql79s5i05a3v8cbn48kdrzf7nc-parallel-20151222/bin/parallel
/gnu/store/ax4ppgzmx5cjqr813l3djgzhiniam1yz-moreutils-0.57/bin/parallel

If you wanted to install both in your profile, those two paths would
both become: ~/.guix-profile/bin/parallel

To resolve the conflict, the first one is chosen and the rest are not
included in the profile.

In general, we have to investigate collisions on a case by case basis;
some are harmless, some are not.

Avoiding harmful collisions is a use case for creating profiles with
`guix package --profiles`.

> I have noticed that Gmsh lists gfortran and lapack as propagated-inputs 
> and lapack also lists gfortran as an input.

I wonder if this could be improved to eliminate the collisions, but I
don't have experience with those packages.

Re: significance of collisions in 'guix environment'?

[Ludovic Courtès]

Paul Garlick <pgarlick@tourbillion-technology.com> skribis:

> $ ./pre-inst-env guix environment --pure gmsh
>
> I get:
>
> .
> .
> .
> warning: collision encountered:
> /gnu/store/m6s3bz6a2sv7015z78sfzdk5ngkzdwwg-gfortran-4.9.3/libexec/gcc/x86_
> 64-unknown-linux-gnu/4.9.3/liblto_plugin.so.0.0.0
> /gnu/store/frrj3bfbmg5vrd0flh9cf8j64h7cr2v4-gcc-4.9.3/libexec/gcc/x86_64-
> unknown-linux-gnu/4.9.3/liblto_plugin.so.0.0.0
> warning: arbitrarily choosing
> /gnu/store/m6s3bz6a2sv7015z78sfzdk5ngkzdwwg-gfortran-4.9.3/libexec/gcc/x86_
> 64-unknown-linux-gnu/4.9.3/liblto_plugin.so.0.0.0
> .
> .
> .
>
> There are 14 gfortran-related messages and a message regarding
> ld-wrapper and binutils.
>
> How should these messages be interpreted? Could there be duplication
> in the package inputs? 

They are harmless in this case: we purposefully want ld-wrapper to
override binutils, and the gfortran/gcc collisions have no effect.
So you can safely keep going.  :-)

However, these messages are clearly suboptimal, precisely because as a
user one doesn’t what to do with them.  There are ideas about avoiding
these messages and instead getting a proper error when there are really
problematic collisions (for instance, collisions due to two different
versions of a package ending up in the profile.)

HTH!

Ludo’.

Re: significance of collisions in 'guix environment'?

[Paul Garlick]

[-- Attachment #1: Type: text/plain, Size: 682 bytes --]

> 
> 
> So you can safely keep going.  :-)
> 
> 
Thank you Ludo; I pressed on within the environment, ignoring the
collisions for now:
$ ./pre-inst-env guix build -K gmsh
The build succeeds.  Then, to check:
$ GUILE_LOAD_PATH=$HOME/.guix-profile/share/guile/site/2.0 ./pre-inst-
env guix lint gmsh
No errors were reported.  
I have installed the upgraded version of Gmsh for testing:
$ ./pre-inst-env guix package -i gmsh
All looks good; I will submit a patch.
Thank you Leo for the suggestion to use a dedicated profile.  My
(first) impression is that the 'guix environment' command helps
automate this task, at the cost of allowing some duplication of paths.
Regards,
Paul.

[-- Attachment #2: Type: text/html, Size: 1114 bytes --]