Re: [Maxima-discuss] Trying to accommodate GCL is keeping us from making progress

[Robert S. <ro...@st...>]

Wolfgang:

Awesome and amazingly valuable work on the daily testing!

Scieneer indeed should be dropped.

I think we could reconsider GCL if it becomes ANSI compliant. That is to say, Maxima should motivate changes to GCL, not vice versa.

As for Allegro, I don't think it should be dropped necessarily, since it is an active commercial implementation of ANSI CL, but it probably ought to come with a warning label saying that it does not receive regular or first-class support/testing.

Some projects classify their support for operating systems and implementations based on "tiers". For example, the Rust programming language classifies (operating system, processor architecture) compiler targets as:

Tier 3: The target may exist in the code base, but it is not officially designated as "supported". No guarantees about basically anything, even the ability to build.

Tier 2: The target is ensured to be able to be built, but there are no guarantees that the tests will pass.

Tier 1: Continuous integration tests that the target builds and the tests pass.

(Details: https://doc.rust-lang.org/beta/rustc/target-tier-policy.html)

We might adopt a similar, albeit less rigorous standard for communicating our support. To illustrate:

Unsupported: Implementations that do not have any mention or support in the code.

* GCL, JSCL, Allegro Trial Edition, LispWorks Trial Edition

Tier 3: Implementations that are mentioned in the code base, that may or may not build Maxima, and are known to be buggy or fail tests, and are not a priority to core developers.

* Allegro (all platforms)

Tier 2: Implementations that are mentioned in the code base, that also can build Maxima, pass a majority of the tests, but are not a priority to core developers.

* LispWorks (all platforms), ABCL (all platforms), ECL (macOS/arm64, Linux/x86)

Tier 1: Implementations that developers strive to haveworking at all times.

* SBCL (all platforms), CMUCL (Linux/x64), CLISP (all platforms), ECL (Linux/x64), CCL (all platforms)

(Again, this is just an example to illustrate, it is not a proposal.)
It might be over-complicating things a bit. I think most Maxima users care more about having available binaries to run (with e.g., wxMaxima) and less about whether their favorite Lisp is or is not supported.

Cheers,

Robert

On Mon, Oct 17, 2022 at 13:49, Wolfgang Dautermann <wol...@da...> wrote:

> Am 17.10.22 um 18:46 schrieb Robert Dodier:
>> At this point I'd like to discontinue support for GCL in Maxima, and
>> discourage package maintainers from trying to use it, since that will
>> inevitably lead to users reporting non-working features as bugs. Can I
>> get anybody on board with that?
>
> Hi,
>
> I compile Maxima every day using 8 different Lisp's - and run the
> testsuite (main and share) with every version:
>
> https://wolfgang.dautermann.at/maxima/nightlybuild/logfile-summary.txt
> https://wolfgang.dautermann.at/maxima/nightlybuild/logfile-share-summary.txt
>
> You get all build logs at:
> https://wolfgang.dautermann.at/maxima/nightlybuild/
>
> GCL does not look that bad im my 'statistics'.
>
> Maybe - if Camm Maguire reads that list - it would make sense to create
> a real new *release* - 2.6.13 - in the near future, not just 124
> 2_6_13pre-versions? Gcl seemms to be actively maintained (although there
> seems to be just one developer for several years, a pity...).
>
> If "code cleanup" (removal of current configure-options, and currently
> working code in the Maxima code base) is the target, maybe the "support"
> for "Scieneer Common Lisp (SCL)" should be removed - it is/was
> commercial software and their website does no longer exist. I changed it
> to their 'archive.org'-Backup some time ago in our Readme (commit 99a91e9).
>
> And maybe too the support for Allegro Common Lisp too - also a
> commercial Lisp, where one could download a free version for NON
> COMMERCIAL PURPOSES only. I assume, that I am the only guy, which
> compiles Maxima using ACL every day...
> Most tests - of all tested Lisp compilers - fail with ACL (with the main
> testsuite):
> 79 tests failed out of 12,805
> Much more than other Lisp compilers - 7 failed tests (ABCL) is the second.
>
> And maybe for GCL the current code (code for ./configure-options, ...)
> for GCL should be kept, but we might call it a 'not recommended (or not
> very well tested) compiler'? Or somethin similar?
>
> Other Open source projects - e.g. wxMaxima, where I do some
> contributions in my spare time - are usually just tested with the Gnu
> C++ compiler - well, every compiler which supports C++14 *should* work,
> but not every compiler is tested...
>
> Best regards, Wolfgang
>
> _______________________________________________
> Maxima-discuss mailing list
> Max...@li...
> https://lists.sourceforge.net/lists/listinfo/maxima-discuss