Re: Python 3.5 start of update

[Ludovic Courtès]

Christopher Allan Webber <address@hidden> skribis:

> Building on both the work that Diane did and my recent patch for python
> 3.4.4, here's two patches which do both consecutively: the first patch
> updates python-3.4 to 3.4.4, and the second patch adds python-3.5 and
> sets this to be the default python.
>
> I'm interested in packaging Pulsar, but this requires Python 3.5 or
> higher, so these patches are a step towards it.
>
> Most of the work is Diane's.  Thank you Diane!  Great work :)

Nice!

> From bbc1f68204318d79ece3fa46ed71d7c21d6ec8a2 Mon Sep 17 00:00:00 2001
> From: Christopher Allan Webber <address@hidden>
> Date: Fri, 29 Apr 2016 17:53:50 -0500
> Subject: [PATCH] gnu: Add geiser-next.

Oops.  ;-)

> From ec0797ceaed8f20b4dbbfb8fb96b2c4752e93f2e Mon Sep 17 00:00:00 2001
> From: Diane Trout <address@hidden>
> Date: Tue, 21 Jun 2016 09:22:43 -0500
> Subject: [PATCH 2/2] gnu: Add python-3.5.
>
> * gnu/packages/python.scm (python-3.5): New variable.
>   (python-3.4): Inherit from python-3.5.
>   (python-3): Switched alias to point to python-3.5.
>   (python-minimal): Use the system-ffi as without-system-ffi caused the build
>   to fail when trying to import _ctypes.
> * gnu/packages/patches/python-fix-tests-py3.5.patch:
>   New file.  Patch Lib/test/test_pathlib.py to skip test_expanduser.
>   Patch Lib/test/test_tarfile.py to ignore pwd.getpwuid and grp.getgrgid tests
>   for user root being of id 0.

Feel free to add a “Co-authored-by” tag if you think it’s relevant.

> diff --git a/gnu/packages/patches/python-fix-tests-py3.5.patch 
> b/gnu/packages/patches/python-fix-tests-py3.5.patch
> new file mode 100644
> index 0000000..47dded4
> --- /dev/null
> +++ b/gnu/packages/patches/python-fix-tests-py3.5.patch
> @@ -0,0 +1,35 @@
> +Additional test fixes which affect Python 3.5 (and presumably later) but not
> +prior revisions of Python.
> +
> +--- Lib/test/test_pathlib.py     2014-03-01 03:02:36.088311000 +0100
> ++++ Lib/test/test_pathlib.py     2014-03-01 04:56:37.768311000 +0100
> +@@ -1986,8 +1986,9 @@
> +         expect = set() if not support.fs_is_case_insensitive(BASE) else 
> given
> +         self.assertEqual(given, expect)
> +         self.assertEqual(set(p.rglob("FILEd*")), set())
> + 
> ++    @unittest.skipIf(True, "Guix builder home is '/' which causes trouble 
> for these tests")

Could it be that adding a phase that does (setenv "HOME" (getcwd)) would
help?

> +--- Lib/test/test_tarfile.py        2016-02-24 19:22:52.597208055 +0000
> ++++ Lib/test/test_tarfile.py     2016-02-24 20:50:48.941950135 +0000
> +@@ -2305,11 +2305,14 @@
> +     try:
> +         import pwd, grp
> +     except ImportError:
> +         return False
> +-    if pwd.getpwuid(0)[0] != 'root':
> +-        return False
> +-    if grp.getgrgid(0)[0] != 'root':
> ++    try:
> ++        if pwd.getpwuid(0)[0] != 'root':
> ++            return False
> ++        if grp.getgrgid(0)[0] != 'root':
> ++            return False
> ++    except KeyError:
> +         return False
> +     return True

(I was skeptical about this bit so I tried this:

--8<---------------cut here---------------start------------->8---
store-monad@(guile-user) [1]> (gexp->derivation "foo"
                                                #~(call-with-output-file 
#$output
                                                    (lambda (port)
                                                      (write (list 
(false-if-exception (getpwuid 0))
                                                                   
(false-if-exception (getgrgid 0)))
                                                             port))))
$6 = #<derivation /gnu/store/zxlirg8wbl8r5pkkhsc89gxvx6lzmwd7-foo.drv => 
/gnu/store/fr9klqdkfqz8ayiq1fjq5i468qjx39mc-foo 42843c0>
store-monad@(guile-user) [1]> (built-derivations (list $6))
substitute: updating list of substitutes from 'https://mirror.hydra.gnu.org'... 
100.0%
building path(s) `/gnu/store/fr9klqdkfqz8ayiq1fjq5i468qjx39mc-foo'
$7 = #t
--8<---------------cut here---------------end--------------->8---

… and indeed, we get (#f #f).  No wonder because libstore/build.cc
does not add any entry for “root” in /etc/passwd.  Probably worth fixing
someday.)

Other than that, LGTM!

The big frustration is that we’ll have to open a ‘python-updates’
branch, where we can also add Hartmut’s changes, but we’ll have to wait
until ‘core-updates’ is merged before we can start building it, which
could take another week or so.

Thank you!

Ludo’.