Add --with-python=3 configure flag to replace SAGE_PYTHON3=yes

[jdemeyer]

CC: @embray @fchapoton @mkoeppe

Component: build: configure

Author: Jeroen Demeyer

Branch: 8af9c37

Reviewer: Erik Bray

Issue created by migration from https://trac.sagemath.org/ticket/24729

[jdemeyer]

Branch: u/jdemeyer/add___with_python_3_configure_flag_to_replace_sage_python3_yes

[jdemeyer]

Commit: 4b4f40e

[jdemeyer]

New commits:

4b4f40e

Add --with-python=3 configure flag to replace SAGE_PYTHON3=yes

[embray]

comment:4

+1 Been meaning to do the same for a while (actually already did once in an old branch that I abandoned). Though as written this will obviously conflict with #21524.

[fchapoton]

comment:5

what is the advantage ?

[jdemeyer]

comment:6

Replying to @fchapoton:

what is the advantage ?

First of all, it's a more standard way to configure things. But one real advantage is that you only need to set this once when running ./configure initially. With this ticket, once you ran ./configure --with-python=2 you can just run make without needing to remember to set SAGE_PYTHON3=yes.

[fchapoton]

comment:7

ok. Great.

Stupid question maybe: in

+export SAGE_PYTHON_VERSION=@SAGE_PYTHON_VERSION@

should there be " around @SAGE_PYTHON_VERSION@ ?

[jdemeyer]

comment:8

Replying to @fchapoton:

ok. Great.

Stupid question maybe: in

+export SAGE_PYTHON_VERSION=@SAGE_PYTHON_VERSION@

should there be " around @SAGE_PYTHON_VERSION@ ?

That's not needed in shell scripts.

[mkoeppe]

comment:9

When I see a configure option --with-python=..., I would expect that I can do ./configure --with-python=/usr/bin/python3.5.77. How about having options --with-python2, --with-python3 instead?

[jdemeyer]

comment:10

That's a fair comment. Let me just say that I chose the option analogous to --with-mp=gmp|mpir

[jdemeyer]

comment:11

Actually, this shouldn't be a --with option at all since those typically indicate components of the system which should be used. For example, mpfr has a --with-gmp option to indicate which GMP it should use. So it should be an --enable option.

I don't like --enable-python2 and --enable-python3 so much because it's not obvious that they are mutually exclusive. So I would prefer --enable-python=2 and --enable-python=3.

Are you willing to give positive review if I change the option to --enable-python instead of --with-python?

[mkoeppe]

comment:13

No, I would say that --with... is the right thing to use, because Python is a dependency, rather than a feature of Sage. --with... does not require an argument. And we can reserve the use of an argument for future use (such as working with a distribution's Python rather than our own).

And a help string can make clear that two options are mutually exclusive (as well as signalling an error if both are provided.)

[jdemeyer]

comment:14

Replying to @mkoeppe:

No, I would say that --with... is the right thing to use, because Python is a dependency, rather than a feature of Sage.

Huh? Maybe you are confusing Sage-the-library with Sage-the-distribution. I would argue that ./configure is really the configuration script for Sage-the-distribution and that the Python-in-Sage is a feature (not a dependency) of Sage-the-distribution. Python is a dependency of Sage-the-library but I don't think that is relevant here.

Apart from this, there is also the unrelated Python-used-by-the-build-system which is a dependency. In fact, that is what we should use the --with-python option for.

[mkoeppe]

comment:15

Replying to @jdemeyer:

Replying to @mkoeppe:

No, I would say that --with... is the right thing to use, because Python is a dependency, rather than a feature of Sage.

Huh? Maybe you are confusing Sage-the-library with Sage-the-distribution. I would argue that ./configure is really the configuration script for Sage-the-distribution and that the Python-in-Sage is a feature (not a dependency) of Sage-the-distribution. Python is a dependency of Sage-the-library but I don't think that is relevant here.

Hm, well, I thought you might say that. But the configuration of sage-the-distribution does also affect sagelib (configuration is somehow passed through); it does not just make one or the other version of python available for sagelib to find.

Also, in light of various efforts to use system packages when available I could imagine that one day sage-the-distribution could skip installing its copy of Python and then --with-python3=/usr/bin/python3.5.77 would make sense.

Apart from this, there is also the unrelated Python-used-by-the-build-system which is a dependency. In fact, that is what we should use the --with-python option for.

For that one I would suggest something like --with-build-python to disambiguate.

[jdemeyer]

comment:16

OK, I see your point of using --with-python instead of --enable-python. Still, I see no reason for separate --with-python2 and --with-python3 options. Why do you prefer that over --with-python? We already have --with-blas=atlas and not --with-atlas, so an option --with-python=2 seems more consistent.

[mkoeppe]

comment:17

First of all, I'd say --with-python=2 just looks odd to the trained configure-user's eye, one would expect to see --with-python=python2 or perhaps --with-python-version=2.
And as I explained, --with-python2 and --with-python3 allows for future, compatible extensions.

Not sure if the following will make sense, but perhaps a difference to BLAS is that Python exists as an executable. So --with-python3=/usr/bin/python3.5.77 makes sense; whereas one would perhaps see --with-blas=atlas --with-atlas-lib=/usr/lib/dfdkfldkflkdlfk --with-atlas-include=/usr/include/fjdkfjkdjfkjd.

But obviously all of this is not exact science; so I don't insist.

[embray]

comment:18

I like --with-python=2. I think it would be more consistent to be --with-python=python2 but that's obviously redundant and more likely to clash with a future use of --with-python=<path-to-python-executabe>. So I think in this case we could make an exception.

[fchapoton]

comment:19

Not an expert at all on this kind of things, but I agree with --with-python=2.

[jdemeyer]

comment:20

Replying to @mkoeppe:

First of all, I'd say --with-python=2 just looks odd to the trained configure-user's eye, one would expect to see --with-python=python2 or perhaps --with-python-version=2.

I mainly used --with-python=2 because it's the shortest of those alternatives.

[jdemeyer]

comment:21

Given that nobody really objects, is somebody willing to give positive review?

[embray]

comment:23

Unless mkoeppe still objects, +1 from me. I will just make this a dependency of #21524 (even though it came first, I don't mind).

[embray]

Reviewer: Erik Bray

[vbraun]

comment:25

Running the test suite for scipy-0.19.1...
  File "spkg-check.py", line 9
    print r
          ^
SyntaxError: Missing parentheses in call to 'print'

[sagetrac-git]

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

8af9c37

Safer enabling of SAGE_PYTHON3=yes

[sagetrac-git]

Changed commit from 4b4f40e to 8af9c37

[jdemeyer]

comment:27

Right, let's play the make-this-work-without-patching-configure game again :-)

[fchapoton]

comment:29

shouldn't we fix the scipy print also, in passing ?

[jdemeyer]

comment:30

Replying to @fchapoton:

shouldn't we fix the scipy print also, in passing ?

See #24766 for that.

[vbraun]

Changed branch from u/jdemeyer/add___with_python_3_configure_flag_to_replace_sage_python3_yes to 8af9c37

[fchapoton]

Changed commit from 8af9c37 to none

[fchapoton]

comment:33

Well.. On 8.2.b7 I get

./configure --with-python=3
configure: WARNING: unrecognized options: --with-python

[jdemeyer]

comment:34

Try running make configure first.