Python 3 native-search-paths

Python 3 native-search-paths

[Andreas Enge]

I just noticed that there is a variable native-search-paths in python-2,
which looks like it should be updated for python 3. However, when I modify
the package accordingly, it is not recompiled. How come? Is this variable
really needed?

Andreas

Re: Python 3 native-search-paths

[Ludovic Courtès]

Andreas Enge <andreas@enge.fr> skribis:

> I just noticed that there is a variable native-search-paths in python-2,
> which looks like it should be updated for python 3. However, when I modify
> the package accordingly, it is not recompiled. How come? Is this variable
> really needed?

Yes it is [0].  :-)

Changing it triggers a recompilation of anything that depends on Python,
but not of Python itself.

Ludo’.

[0] http://lists.gnu.org/archive/html/bug-guix/2013-03/msg00158.html

Re: Python 3 native-search-paths

[Andreas Enge]

On Sat, Aug 31, 2013 at 06:07:35PM +0200, Ludovic Courtès wrote:
> Andreas Enge <andreas@enge.fr> skribis:
> > I just noticed that there is a variable native-search-paths in python-2,
> > which looks like it should be updated for python 3. However, when I modify
> > the package accordingly, it is not recompiled. How come? Is this variable
> > really needed?
> 
> Yes it is [0].  :-)
> Changing it triggers a recompilation of anything that depends on Python,
> but not of Python itself.

How can it trigger anything if the output path is identical
(here: /nix/store/0hf9lvd48q1l98k5jf98z17zq29955c4-python-3.3.2)?

Andreas

Re: Python 3 native-search-paths

[Ludovic Courtès]

Andreas Enge <andreas@enge.fr> skribis:

> On Sat, Aug 31, 2013 at 06:07:35PM +0200, Ludovic Courtès wrote:
>> Andreas Enge <andreas@enge.fr> skribis:
>> > I just noticed that there is a variable native-search-paths in python-2,
>> > which looks like it should be updated for python 3. However, when I modify
>> > the package accordingly, it is not recompiled. How come? Is this variable
>> > really needed?
>> 
>> Yes it is [0].  :-)
>> Changing it triggers a recompilation of anything that depends on Python,
>> but not of Python itself.
>
> How can it trigger anything if the output path is identical
> (here: /nix/store/0hf9lvd48q1l98k5jf98z17zq29955c4-python-3.3.2)?

It has no effect in the build of Python itself; its effect is only when
building Python code that uses Python and one or more Python libs.

Ludo’.

Re: Python 3 native-search-paths

[Andreas Enge]

On Sat, Aug 31, 2013 at 08:06:34PM +0200, Ludovic Courtès wrote:
> It has no effect in the build of Python itself; its effect is only when
> building Python code that uses Python and one or more Python libs.

I still think something is wrong here and that this variable should be
included into the computation of the hash. When I now recompile a
package depending on python, it gets the same hash, but it is different.

Or more easily: The python package with the new native-search-paths behaves
differently from the previous one, so it should have a different hash.

Andreas

Re: Python 3 native-search-paths

[Ludovic Courtès]

Andreas Enge <andreas@enge.fr> skribis:

> On Sat, Aug 31, 2013 at 08:06:34PM +0200, Ludovic Courtès wrote:
>> It has no effect in the build of Python itself; its effect is only when
>> building Python code that uses Python and one or more Python libs.
>
> I still think something is wrong here and that this variable should be
> included into the computation of the hash. When I now recompile a
> package depending on python, it gets the same hash, but it is different.
>
> Or more easily: The python package with the new native-search-paths behaves
> differently from the previous one, so it should have a different hash.

The hash is the hash of all the inputs to the build process,
recursively.  If a single bit changes somewhere, the hash changes.

But here, search path specifications do not affect the build process of
packages that carry them in any way (an improvement over Nix where
everything is an environment variable, and thus an input to the hash,
even if it is not meant to influence the build process.)

When ‘native-search-paths’ changes in python3, it’s not python3 that
behaves differently: it’s the build process of packages using it that
does.

Hope this clarify things.

Thanks,
Ludo’.

Re: Python 3 native-search-paths

[Andreas Enge]

On Sun, Sep 01, 2013 at 03:31:45PM +0200, Ludovic Courtès wrote:
> When ‘native-search-paths’ changes in python3, it’s not python3 that
> behaves differently: it’s the build process of packages using it that
> does.

Not really. Then the output of, say, pytz, will change when the
native-search-paths of python changes, no? So we would obtain a pytz package
which is binary different, but has the same hash?

Andreas