fix trig function having a multiple of pi/3, p/4, pi/6, pi/12; fixes …

[rwst]

…#1035

[rwst]

The problem (apart from the incompleteness) was that the original author returned the wrong expression for unhandled cases in eval()s rarg. Returning the right thing however was impossible unless the canonical check was made less stringent, else an infinite loop or a memleak occurred (because is_canonical bailed out where it should not).

[certik]

Thanks @rwst for fixing it! I need to go over your code more carefully.

[isuruf]

@rwst, I have some improvements to this PR. Can you enable Allow edits from maintainers for this PR? (There should be a tick box in the right side bar)

[rwst]

Please go ahead.

[isuruf]

@certik, @rwst, can you review?

[isuruf]

Change I've made is to reduce sin(r*pi + x) to the first quadrant for any r rational number.

[certik]

I think that looks good to me.

How are benchmarks with trigonometric functions affected?

[isuruf]

We don't have any that uses the code paths changed in this PR. Only simple arguments like x are used for trig functions

[certik]

I see, thanks. +1 to merge then.