Various python 3 issues

[ohanar]

There are a number of really small issues when trying to build with python 3:

1. sagenb is a marked as dependency of jmol, even though this is no longer the case
2. sagenb is marked as a runtime dependency of sage, again this is no longer the case
3. scons and sagenb don't work with python 3, so they should only be a standard package for python 2
4. atlas needs a small change to make its build script work with python 3

CC: @vbraun @nathanncohen

Component: build

Author: R. Andrew Ohana

Branch/Commit: 83e849e

Reviewer: Jeroen Demeyer

Issue created by migration from https://trac.sagemath.org/ticket/18508

[ohanar]

Dependencies: #17607

[vbraun]

comment:3

The notebook() command uses sagenb, so its a runtime dependency. See also #18512

I'm fine with EOL'ing SageNB with Python2, but you should probably post to sage-devel to see if anybody is interested in maintaining it.

[jdemeyer]

comment:4

Replying to @vbraun:

The notebook() command uses sagenb, so its a runtime dependency.

No, it's not a run-time dependency, since Sage works without the Sage notebook. Of course, the notebook() command doesn't work, but that doesn't matter.

[jdemeyer]

comment:5

Unless you add support for the python2standard package type, this branch isn't going to work.

[vbraun]

comment:6

There are other packages that could be removed yet Sage would still starts up; of course some functionality is then missing. I don't see how sagenb is different. Define "works". If you mean "works enough to create the conway polynomial pickles" then I guess so.

The python2standard type is from #17607

[jdemeyer]

comment:7

Replying to @vbraun:

Define "works".

This is actually documented in build/deps:

# Everything needed to start up Sage using "./sage".  Of course, not
# every part of Sage will work.  It does not include Maxima for example.

[jdemeyer]

comment:8

I guess a formal definition of "works" is that

./sage -c ""

has exit status 0.

[jdemeyer]

comment:10

I don't like this from #17607:

    if [ "$1" = all ]; then
        PKG_VAR=PYTHON
        if [ "$SAGE_PYTHON3" = yes ]; then
            PKG_NAME=python3
        else
            PKG_NAME=python2
        fi
        PKG_VERSION=$(newest_version $PKG_NAME)
        echo "$PKG_NAME $PKG_VERSION $PKG_VAR"
    fi

Just define PYTHON as alias of PYTHON2 or PYTHON3, analogous how build/install deals with GMP/MPIR and the toolchain. See the SAGE_MP_LIBRARY and TOOLCHAIN variables.

[vbraun]

comment:11

We will have to remove python2 in the not so distant future so I don't see the value in winning any beauty contests in the meantime.

[jdemeyer]

comment:12

Replying to @vbraun:

We will have to remove python2 in the not so distant future

I think that future might be more distant than you think... but that is besides the point anyway.

I don't see the value in winning any beauty contests in the meantime.

This isn't just about beauty. It's about writing code which is easy to understand by doing things in a consistent way.

[nathanncohen]

comment:13

This isn't just about beauty. It's about writing code which is easy to understand by doing things in a consistent way.

+1

Plus we might also want to work on the build/ files to make them more readable, and that task is not made simpler if we have to clean "workarounds" on the way.

Nathann

P.S.: this does not exactly give the impression of a 'clean job'.

(GRRR|…)~/.Sage/build/pkgs$ head python3/checksums.ini -n 1
tarball=Python-VERSION.tar.gz
(GRRR|…)~/.Sage/build/pkgs$ head python2/checksums.ini -n 1
tarball=python-VERSION.tar.gz

[vbraun]

comment:14

Replying to @nathanncohen:

Plus we might also want to work on the build/ files to make them more readable

Like strip out the Python 2.x stuff in a little bit over 4 years?

[jdemeyer]

comment:15

Replying to @vbraun:

The python2standard type is from #17607

...and I will remove it in #18517 since it's a very ugly hack to add to the build system.

[jdemeyer]

Changed dependencies from #17607 to #17607, #18517

[jdemeyer]

comment:16

What exactly are you trying to achieve with making scons build on Python 2 only? How do you solve the problem that c_lib still requires scons to build?

[jdemeyer]

comment:17

For sagenb, you could instead edit build/pkgs/sagenb/spkg-install to skip the package on Python 3.

[vbraun]

comment:18

scons doesn't support Python 3 at this point.

[jdemeyer]

comment:19

Replying to @vbraun:

scons doesn't support Python 3 at this point.

I know, but read my question: How do you solve the problem that c_lib still requires scons to build? In other words: what's the gain from removing SCONS as standard package if it's still a dependency of csage?

[vbraun]

comment:20

I don't claim to have solved that, but we might want to get rid of scons. I think Andrew did some work towards autotoolizing polybori. We might want to do the same for c_lib, though its not the only option.

[jdemeyer]

comment:21

Replying to @vbraun:

I don't claim to have solved that, but we might want to get rid of scons.

If it helps you, I propose to make scons an optional package. It will still be built by default since it's needed as dependency of polybori and c_lib.

We might want to do the same for c_lib, though its not the only option.

I assume you are aware of #17854?

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

1628a11

fix scons and sagenb to work with #18517

[sagetrac-git]

Changed commit from 9d2af12 to 1628a11

[kcrisman]

comment:25

Related: sagemath/sagenb#343

[jdemeyer]

comment:26

Replying to @ohanar:

I've been basically assuming that #17854 will be done in the not too distant future

It's only ntl_wrap which remains (#18367). Moving ntl_wrap to the Sage library is relatively easy I think but a lot of boring work.

[jdemeyer]

comment:27

The content of "/usr/local/src/sage-config/build/pkgs/sagenb/type" must be 'base', 'standard', 'optional' or 'experimental'

(just set back the type to standard, see #18517)

[sagetrac-git]

Changed commit from 1628a11 to 6fe3896

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

6fe3896

make sagenb a standard package again

[jdemeyer]

comment:30

This is probably good, but I'd rather wait for Sage 6.8.beta1 to formally review this.

[jdemeyer]

Changed branch from u/ohanar/python3smallfixes to u/jdemeyer/python3smallfixes

[jdemeyer]

Changed commit from 6fe3896 to 83e849e

[jdemeyer]

New commits:

83e849e

Merge tag '6.8.beta2' into t/18508/python3smallfixes

[jdemeyer]

Changed dependencies from #17607, #18517 to none

[vbraun]

comment:34

Reviewer name is missing

[jdemeyer]

Reviewer: Jeroen Demeyer

[vbraun]

Changed branch from u/jdemeyer/python3smallfixes to 83e849e