Backtracking and revealing default assumptions

[sylee957]

I see there are some problems of the new assumption system that it is not able to resolve some facts inferred from the default assumption associated with the symbol.
I think that the the old assumptions system could solve because it has some backtracking.

For example, if the new assumptions system backtracks the conditions of Q.invertible and reveals some assumptions then the problem like below should be solved.

>>> from sympy import *
>>> from sympy.matrices.expressions.fourier import DFT, IDFT

>>> ask(Q.unitary(DFT(n)))
True
>>> ask(Q.invertible(DFT(n)))

>>> ask(Q.invertible(DFT(n)), Q.unitary(DFT(n)))
True

[asmeurer]

This sort of problem is solved by satask.

The challenge with doing this with the old handler system is avoiding expensive evaluation. In the worst case you could end up evaluating every possible handler for an object.

[sylee957]

I see things like Identity = staticmethod(CommonHandler.AlwaysTrue) are duplicated all over the different handlers to achieve same thing.
In my opinion, I think that the handler classes should be able to distribute them automatically in creation.

[oscarbenjamin]

The challenge with doing this with the old handler system is avoiding expensive evaluation. In the worst case you could end up evaluating every possible handler for an object.

Where it goes wrong is that in some cases the handlers try to do too much. It's okay for them to just return None some of the time and more important for them to be fast.

[sylee957]

What I mean is that would it be worth bounding the facts derived from CommonHandler.AlwaysTrue to the classes? This would still be cheaper

[Arpan612]

@sylee957 Sir, I have tested the code snippet you shared and I agree with the problem you have raised. Could you tell me how can we proceed with solving this problem? And do you expect a PR for this issue? If yes, I would like to work on it.

[asmeurer]

@Arpan612 you can try fixing it. You should be careful about potential performance issues, as I mentioned. If you do fix it, a PR is useful, even if it's just a work in progress.

[Arpan612]

@asmeurer Understood sir. Could you please give me some insights on how to proceed with the problem?