density_plot() is broken with functions involving symbolic expressions

[sagetrac-tmonteil]

As reported on this ask question, the following does not work:

sage: f1(a, b) = 1 - b / a
sage: f2(a, b) = 1 - a / b
sage: def f12(a, b):
....:         if a - b < 0:
....:                 return f1(a, b)
....:         else:
....:                 return f2(a, b)
sage: density_plot(f,(1,2),(1,2))
KeyError: 'text/plain'

While the following works:

sage: f1(a, b) = 1 - b / a
sage: f2(a, b) = 1 - a / b
sage: def f12(a, b):
....:     if a - b < 0:
....:         return RDF(f1(a, b))
....:     else:
....:         return RDF(f2(a, b))
sage: density_plot(f12,(1,2),(1,2))

CC: @kcrisman

Component: graphics

Author: Dave Morris

Branch: bb5a812

Reviewer: Karl-Dieter Crisman

Issue created by migration from https://trac.sagemath.org/ticket/17684

[DaveWitteMorris]

Branch: public/17684

[DaveWitteMorris]

New commits:

bb5a812

trac 17684 density plot of symbolic values

[DaveWitteMorris]

Commit: bb5a812

[DaveWitteMorris]

Author: Dave Morris

[kcrisman]

comment:3

Nice. Unfortunately I can't upgrade my Sage currently to actually check this (quite old computer with fragile installation, based on past experience, and I can't afford to bork it), but patchbot says okay ... would you be willing to post some of the plots from the documentation (Including the tests for bug fixes, including your new one) here to verify? Then I'd be happy to provide positive review, the code should certainly be okay.

---

That said, the real bug here is in fast_float, because it should still be taking this function in setup_for_eval_on_grid and doing the "right" thing with it. See this admittedly bizarre Sage cell interact. Maybe we should open/search for a ticket about this error - I'm sure it comes up elsewhere, there are definitely open tickets with fast_float or fast_callable and I know some have suggested completely rewriting that code.

[kcrisman]

comment:4

(And of course if providing all that is too much to ask, that's fine too; I'm sure someone else will be able to review this, it's an obvious hotfix.)

[DaveWitteMorris]

Attachment: density_plots.pdf.gz

density plot doctests

[DaveWitteMorris]

comment:5

I think the attachment (density_plots.pdf) has all of the plots that are in the doctests of this file. (Let me know if I missed any that you were wondering about.) However, I jpeg-compressed the file, so there are some artifacts, especially where the plot has white at an edge.

I did not find the issue of this ticket listed anywhere. (Related: ticket #5572 seems to suggest replacing fast_float with fast_callable.) I agree that the real problem is in fast_float, so probably a ticket should be opened (or a comment added to #5572). I think that fix will take more care, because RDF works on this ticket, but I suspect that fast_float will need to consider the precision of its inputs before choosing the field of floating-point numbers to use.