Add oct, a quad-double library

[rtoy]

A long time ago oct used to be part of quicklisp but was removed because it did build at some point.

I finally got around to updating it and I've updated the code and tests so that sbcl can now build and run the tests. Some tests fail, but they're marked as expected failures, so tests pass.

Use (asdf:test-system :oct) to run the tests.

Also, I had to set a dynamic-space size of 2048 (!) to allow sbcl to build oct. Not sure why that is. Cmucl only needs 384 MB (or less), and ccl64 can build and run the tests with its default heap sizes.

[rtoy]

One additional comment.

Oct uses a submodule to get named-readtables which is a fork of melisgl/named-readtables.

If quicklisp updates melisgl/named-readtables to the latest version that recently added support for cmucl, the submodule could be removed in favor of the quicklisp version of named-readtables.

[quicklisp]

I use that latest version of named-readtables - I can't build oct until the submodule is removed.

[rtoy]

I've removed the submodule and updated oct.asd not to try to load its own version. But it seems a heap of 2 GiB isn't enough. I tried 4 GiB which seems like enough. Don't understand why sbcl needs so much memory to compile up the tables.

[xach]

The bundled copy of "rt.asd" also causes problems, is it necessary?

[rtoy]

Ah, I can't right now because I modified the version from quicklisp to add some things I needed to be able to mark tests as expected failures. This is basically the same as the rt version from ansi-test. I don't know where the upstream version of rt is for quicklisp. I'll need your advice on how to proceed here. It doesn't have to be called rt. I can rename it. But perhaps adding these changes to rt would be useful to someone else besides me.

[phoe]

I don't know where the upstream version of rt is for quicklisp.

https://github.com/quicklisp/quicklisp-projects/blob/master/projects/rt/source.txt states that it's kmr-git.

[rtoy]

Where is kmr-git? I can't find it.

[phoe]

http://git.kpe.io/?p=rt.git;a=summary

[rtoy]

Thanks. Not really sure how I can propose a change there though. But at least I can now create a fork of the version used by quicklisp.

[rtoy]

I'm just too stupid because I can't figure out how to clone the repo.

[xach]

The clone URL is http://git.kpe.io/rt.git

…

On Sun, Jan 17, 2021 at 6:49 PM rtoy ***@***.***> wrote: I'm just too stupid because I can't figure out how to clone the repo. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1958 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAEB7YTMDVABMK7EN5VWJLS2NZRVANCNFSM4VIMLKOQ> .

[rtoy]

Thanks. I see that this version has changed quite a bit from the version used by ansi-tests.

But I wonder, is the problem because I have an rt.asd file? Would it work if I just got rid of it and loaded up rt as part of the test-system?

[xach]

Yes, the main problem is rt.asd. Incorporating rt.lisp into the test files would avoid the problem.

[rtoy]

Ok. The latest version has removed rt.asd. Let me know if there's more you'd like me to do.

[quicklisp]

There is still a file named rt/rt.asd and it's causing the same problem.