Build fails on Python 3.5.2

[vbraun]

[sagelib-9.1] ************************************************************************
[sagelib-9.1] Traceback (most recent call last):
[sagelib-9.1]   File "setup.py", line 73, in <module>
[sagelib-9.1]     from sage_setup.find import find_extra_files
[sagelib-9.1]   File "/var/lib/buildbot/slave/sage2_git/build/src/sage_setup/find.py", line 15, in <module>
[sagelib-9.1]     import importlib.machinery
[sagelib-9.1] ImportError: No module named machinery
[sagelib-9.1] ************************************************************************
[sagelib-9.1] Error building the Sage library
[sagelib-9.1] ************************************************************************

Component: build

Reviewer: Jonathan Kliem

Issue created by migration from https://trac.sagemath.org/ticket/29753

[mkoeppe]

comment:1

What code is invoking this? Please post config.log

[mkoeppe]

comment:2

In particular, please clarify what you mean by building "on Python 3.5.2" in light of my explanation in https://groups.google.com/d/msg/sage-devel/fgsSsJmVVXo/n2BQBSaBAAAJ

[kliem]

comment:4

You probably already found out: This is caused by #27472.

I see this error on the standard-python2-gcc_spkg tests.

E.g. https://github.com/kliem/sage-test-27122/runs/722743058

[mkoeppe]

comment:5

Replying to @kliem:

You probably already found out: This is caused by #27472.

Could you elaborate? I don't understand this ticket, and unfortunately it seems that Volker does not see my requests for more information.

[kliem]

comment:6

Sorry, this has nothing to do with #27472. The line 15 in sage_setup/find.py it is complaining about was added in #28000 (remove python2 support from sage lib).

To me it looks like this also removed python2 support for building sage, which was not the plan I believe (I think the agreement was that python2 may still be used to build sage). The module importlib.machinery just doesn't exist in python2 and hence the error.

I think the title of this ticket here is not correct as it appears to have nothing to do with python 3.5.2. It appears that the kucalc buildbot just doesn't have python3 installed. It should then in the build process automatically install python3, but it doesn't even get there.

[mkoeppe]

comment:7

But why would sage_setup/find.py be invoked using system python2?

[kliem]

comment:8

That's pretty much the first thing it runs. So apparently rejecting python2 wasn't very successful. E.g. it tries to install scipy before python3, which doesn't look like it did really reject the wrong system python.

[mkoeppe]

comment:9

Are you referring to a build --with-python=2?

[mkoeppe]

comment:10

That is being removed in 9.2.

[kliem]

comment:11

I'm refering to the github testrun ubuntu-bionic, standard-python2-gcc_spkg (and likewise all ... standard-python2-gcc_spkg).

To my understanding this tries to install sage on a system, where all standard packages are installed but not python3 (instead python2). Is this correct?

[mkoeppe]

comment:12

No, the factor python2 means that ./configure --with-python=2 is called. And that is scheduled to be removed in 9.2 (the removal of this option is done in #29669).

[kliem]

comment:13

Ok. Using this option the build fails exactly as in the description of this ticket.

[mkoeppe]

comment:14

OK, then let's assume that this is also what Volker meant. Marking it as invalid.

[kliem]

Reviewer: Jonathan Kliem

[kliem]

comment:15

I agree, this import error really shouldn't happen in python 3.5.2, see https://docs.python.org/3.5/library/importlib.html. The module is present in 3.5.2 and an import error implies that not python 3.5.2 was used but probably some python2.