Stream: math-comp devs

Topic: Applying for GitLab for Open Source requires OSI license


view this post on Zulip Erik Martin-Dorel (Sep 29 2022 at 12:58):

Dear all, Cc @Cyril Cohen @Théo Zimmermann

[ as a follow-up of https://github.com/coq/coq/wiki/Coq-Call-2022-06-29 and https://github.com/math-comp/math-comp/pull/924 ]

It appears we can't apply for GitLab Open Source program yet (and benefit from 2×50,000 CI/CD minutes per month, vs. 25×400 currently) because math-comp does not have an OSI-approved license.

While BSD-3, for example, (the closest license from CeCILL-B (that might be viewed as a "French BSD-4" license)) would be OK.

What do you think?

(I believe another incentive for suggesting a license change was raised by the compatibility issue when combining CeCILL-B and GPL in infotheo, IIRC.)

view this post on Zulip Karl Palmskog (Sep 29 2022 at 13:51):

@Erik Martin-Dorel here are the relevant links that explain the incompatibility of CECILL-C and CECILL-B with GPL:

This is why one likely can't legally distribute .vo files that simultaneously use Mathematical Components and some other GPL-licensed library

view this post on Zulip Théo Zimmermann (Sep 29 2022 at 17:07):

I strongly encourage the MathComp team to consider doing the effort to relicense to an equivalent OSI-approved and GPL-compatible license. If that does not work, I would even suggest reaching out to the authors of CeCILL (they include Inria) to ask for a new CeCILL-B license that would be GPL-compatible and that they submit it for approval to the OSI (the CeCILL 2.1 license was already OSI-approved).

view this post on Zulip Théo Zimmermann (Sep 29 2022 at 17:10):

Of course, the MathComp authors also have the option to relicense their software under CeCILL 2.1, which is GPL-compatible and OSI-approved, with less effort (no need to get agreement of copyright holders since there is an explicit path for such a relicensing in the CeCILL-B license) but this is probably not something that anyone wants since it is a strong copyleft license.

view this post on Zulip Cyril Cohen (Sep 29 2022 at 17:52):

Noted. I added the topic to the next meeting on Wednesday October 5th. I am absolutely incompetent on licencing, so I would really appreciate if one (or several) of you @Erik Martin-Dorel, @Karl Palmskog and @Théo Zimmermann could join and contribute your knowledge to the meeting.

view this post on Zulip Théo Zimmermann (Sep 29 2022 at 17:55):

FTR, the GPL-compatible issue is a serious issue but it is separate from the OSI-approved / GitLab OSS plan thing.

view this post on Zulip Karl Palmskog (Sep 29 2022 at 17:56):

@Cyril Cohen I would be happy to join, but Théo is more experienced than me on licensing

view this post on Zulip Théo Zimmermann (Sep 29 2022 at 17:56):

Yeah, I should be able to join. I will just have to move another meeting. How long do you expect to spend on this issue?

view this post on Zulip Cyril Cohen (Sep 29 2022 at 17:57):

Théo Zimmermann said:

Yeah, I should be able to join. I will just have to move another meeting. How long do you expect to spend on this issue?

I'm not sure, what would be a reasonable duration to expose all the pros and cons and answer a few questions?

view this post on Zulip Karl Palmskog (Sep 29 2022 at 17:58):

as a big downstream consumer of MathComp (in Coq projects and research), I can at least give the licensing perspective from there

view this post on Zulip Théo Zimmermann (Sep 29 2022 at 17:59):

This kind of topic can often extend quite a lot, so it depends on how much time you think the MathComp team wants to spend on this.

view this post on Zulip Théo Zimmermann (Sep 29 2022 at 17:59):

It could be from 15 minutes to 45 minutes I guess.

view this post on Zulip Karl Palmskog (Sep 29 2022 at 18:02):

my Wednesday morning(s) are looking good, so I can stay a full hour if that's needed. But perhaps set a limit of 45 min.

view this post on Zulip Karl Palmskog (Sep 29 2022 at 18:13):

I guess we can link to our licensing advice from Coq-community in this context as well, which sort of represents what Théo and I could agree on was good free open source software licensing practices: https://github.com/coq-community/manifesto#best-practices

TLDR: we recommend MIT for the permissive licensing path and MPL-2.0 for the (weak) copyleft path, and otherwise another license that is both approved as an open source license by OSI and considered a free software license by FSF.

Nevertheless, Erik has a point that BSD-3-Clause is likely the closest OSI license to MathComp's current license.

view this post on Zulip Maxime Dénès (Sep 29 2022 at 18:24):

Doesn't relicensing mathcomp require approval from MSR? (and Inria, of course)

view this post on Zulip Karl Palmskog (Sep 29 2022 at 18:25):

as Théo says above, it depends on what you relicense to. But for relicensing to BSD-3-Clause, yes, very likely.

view this post on Zulip Karl Palmskog (Sep 29 2022 at 18:28):

this is why the path of asking for a new GPL-compatible, eventually-OSI-approved CECILL-B version may not look so bad after all. There seems to be an upgrade clause here: https://github.com/math-comp/math-comp/blob/master/CeCILL-B#L497-L499

view this post on Zulip Erik Martin-Dorel (Sep 29 2022 at 19:19):

I added the topic to the next meeting on Wednesday October 5th.

Thanks! AFAIAC, next Wednesday morning I just have one other meeting planned, but it can be moved, so I can join as well.
Just let me know of the starting time that looks good to all of you.

view this post on Zulip Erik Martin-Dorel (Sep 29 2022 at 19:20):

the path of asking for a new GPL-compatible, eventually-OSI-approved CECILL-B version may not look so bad after all. There seems to be an upgrade clause here.

Good point, and BTW (independently of the GitLab Open Source program application for mathcomp),
CeCILL-C (the license of coq-interval and mathcomp-analysis, for instance) could/should just as well be revised by CEA/CNRS/Inria legal teams… so that the v2 is OSI-approved :fingers_crossed:

view this post on Zulip Enrico Tassi (Sep 29 2022 at 19:36):

Maxime Dénès said:

Doesn't relicensing mathcomp require approval from MSR? (and Inria, of course)

Yes. I think the amount of code which did not move after 2012 is substantial.
IMO the best option would be a CECILL-Bv2 which is not GPL incompatible, but this is not in our hands.

view this post on Zulip Enrico Tassi (Sep 29 2022 at 19:38):

I did recall MC being double licensed, like LGPL and CECILL-B, but I can't find any trace of it in the historical archives

view this post on Zulip Erik Martin-Dorel (Oct 02 2022 at 21:02):

Erik Martin-Dorel said:

I added the topic to the next meeting on Wednesday October 5th.

Thanks! AFAIAC, next Wednesday morning I just have one other meeting planned, but it can be moved, so I can join as well.
Just let me know of the starting time that looks good to all of you.

Actually @Théo Zimmermann @Karl Palmskog I've just seen the MC meeting schedule is documented in https://github.com/math-comp/math-comp/wiki#meetings
→ on Wednesday 10AM (Paris Time)

See you soon.


Last updated: Jan 29 2023 at 16:02 UTC