gexps

gexps

[David Craven]

Hi, I have a couple of questions:

How do gexps relate to packages? Are they used in the
package->bag->derivation process or is it just for writing services?

gexp->script and gexp->file only dump text into the store. How can I
make them ignore the system/target architectures?

Re: gexps

[Ludovic Courtès]

David Craven <david@craven.ch> skribis:

> Hi, I have a couple of questions:
>
> How do gexps relate to packages? Are they used in the
> package->bag->derivation process or is it just for writing services?

Yes, currently.

Currently the bag->derivation process uses
‘build-expression->derivation’, which is the (ugly) predecessor of
‘gexp->derivation’.

The goal is to switch over to gexps eventually; ‘wip-build-systems-gexp’
does exactly that, but there were performance regressions that I never
took the time to address completely.

> gexp->script and gexp->file only dump text into the store. How can I
> make them ignore the system/target architectures?

It’s not really possible, though I agree it would be nice!

If we force use of the host system type (say x86_64-linux), independent
of (%current-system) (say mips64el-linux), we approach what you want.
However, that would give us different derivations depending on the host
system type.  That is, ‘guix system build -s mips64el-linux -d’ on
x86_64 would return something different from ‘guix system build -d’ on
mips64.  This is not desirable.

An option would to have ‘gexp->file’ produce a fixed-output derivation
(it can compute the hash of the output on the host side).

Or maybe we need another mechanism similar to fixed-output derivations
to define equivalence classes of derivations.

Food for thought…

Ludo’.

Re: gexps

[David Craven]

> If we force use of the host system type (say x86_64-linux), independent
> of (%current-system) (say mips64el-linux), we approach what you want.
> However, that would give us different derivations depending on the host
> system type.  That is, ‘guix system build -s mips64el-linux -d’ on
> x86_64 would return something different from ‘guix system build -d’ on
> mips64.  This is not desirable.

Maybe an option would be if %current-system != host-system run guile
using qemu-arm?

> An option would to have ‘gexp->file’ produce a fixed-output derivation
> (it can compute the hash of the output on the host side).
>
> Or maybe we need another mechanism similar to fixed-output derivations
> to define equivalence classes of derivations.

I'll have to read up on fixed-output derivations. Can you give me an
example of where they are used currently?

Re: gexps

[Ludovic Courtès]

David Craven <david@craven.ch> skribis:

>> If we force use of the host system type (say x86_64-linux), independent
>> of (%current-system) (say mips64el-linux), we approach what you want.
>> However, that would give us different derivations depending on the host
>> system type.  That is, ‘guix system build -s mips64el-linux -d’ on
>> x86_64 would return something different from ‘guix system build -d’ on
>> mips64.  This is not desirable.
>
> Maybe an option would be if %current-system != host-system run guile
> using qemu-arm?

That would make it a different derivation.

>> An option would to have ‘gexp->file’ produce a fixed-output derivation
>> (it can compute the hash of the output on the host side).
>>
>> Or maybe we need another mechanism similar to fixed-output derivations
>> to define equivalence classes of derivations.
>
> I'll have to read up on fixed-output derivations. Can you give me an
> example of where they are used currently?

For downloads (origins).  The manual mentions it briefly.

Ludo’.