Replace reproducible_repr() with usage of displayhook

[user202729]

• It is not very well-known that doctest has magic features to ignore the ordering of dict and set (see for example Fix compilation and doctests with flintlib 3.2 #39413 (comment) ), I try to make it more discoverable.
• Since the displayhook is in place (since 68b10e1), the reproducible_repr (added in e68d6eb) is mostly redundant.
• Implement _repr_pretty_ for KeyConvertingDict to make doctest deterministic. (based on the framework of Make Sequence pretty-printed in IPython #39027)

There's a slight difference between reproducible_repr ordering (which always order by string representation) and this function (which tries first to order by element values, then by string representation), and the ordering of elements is reversed for multivariate polynomial ring (in R.<x,y,z> = QQ[], z < y < x). However, it should still be deterministic.

📝 Checklist

• The title is concise and informative.
• The description explains in detail what this PR is about.
• I have linked a relevant issue or discussion.
• I have created tests covering the changes.
• I have updated the documentation and checked the documentation preview.

⌛ Dependencies

[github-actions]

Documentation preview for this PR (built with commit d5a22f1; changes) is ready! 🎉
This preview will update shortly after each push to this PR.

[tobiasdiez]

and the ordering of elements is reversed for multivariate polynomial ring (in R.<x,y,z> = QQ[], z < y < x)

Would it be desired/easy to change this to the more natural ordering x < y < z?

[tobiasdiez]

Should we then mark it as deprecated?

[user202729]

Not sure if it's necessary. On the other hand it was never used except for one doctesting case, and letting it stay there might serve as a signpost to the correct handles of reproducible dict/set repr. Thoughts?

[tobiasdiez]

If it's no longer used/needed in the sage codebase, then I think it should be deprecated.

[user202729]

Actually there's a little caveat — if the repr forms a total order but the values themselves do not form a total order then it might be desirable to use this function (e.g. CIF in #39413 ). But I think it's rare enough that you can add an explicit sorted(…, key=[some reliable key]) anyway.

[user202729]

Also there's still one place where the function is used below it (in trace_method which is also used in doctest):

        print("exit {} -> {}".format(meth, reproducible_repr(res)))

I think this usage is slightly more reasonable (it's not much longer than the alternative); on the other hand it can also be replaced with from IPython.lib.pretty import pretty → pretty(res).

Thoughts?

[tobiasdiez]

In general, if we can remove sage-specific things in the basic infrastructure then this is a big win. There is a considerably higher chance that people know about pretty from IPython than that they go hunting for a reproducible_repr method in sage. So if you think all the existing usage can be covered, then do please do so and deprecate the method.

[tobiasdiez]

Why is this here z < y < x while in the other examples a bit down it is still x < y < z?

[user202729]

Ugh, must be because I don't have the optional packages.

By the way, how can optional packages be installed in meson again? (sage -i doesn't work if I recalled correctly?) Find the corresponding conda package, but how do you know which conda package corresponds to which sage package?

[tobiasdiez]

Find the corresponding conda package, but how do you know which conda package corresponds to which sage package?

There are conda.txt files in build/pkgs that tell you the mapping from sage pkg to conda pkg. (But they are not actively maintained and thus may be incorrect.)

[user202729]

End up most of the packages needed seems not available on conda anyway (fes ginv macaulay2 magma msolve pycryptosat) so I just try to fix it by best effort.

[user202729]

Would it be desired/easy to change this to the more natural ordering x < y < z?

Basically it is determined by Singular's monomial comparison (p_Cmp in singular_polynomial_cmp). If we just flip it, we get the unfortunate result of x^2 < x.

On the other hand we can flip the order we pass variables to Singular, which is technically possible, but it seems rather unnatural and the last time Singular polynomial comparison was changed a bunch of tests need to be fixed.

You can also argue that in R.<x,y>, then x has exponent vector (1, 0), y has exponent vector (0, 1), the latter is smaller. (which is probably how Singular implements the comparison internally.)

MPolynomialRing_polydict has the same behavior anyway.

If anything we should fix this upstream as well.