cygwin: numerical noise in sage/rings/integer.pyx

[williamstein]

sage -t  "devel/sage/sage/rings/integer.pyx"                
**********************************************************************
File "/home/wstein/sage-4.4.3/devel/sage/sage/rings/integer.pyx", line 1681:
    sage: 2^float(1.5)       # python float
Expected:
    2.8284271247461903
Got:
    2.8284271247461898
**********************************************************************
1 items had failures:
   1 of  26 in __main__.example_42
***Test Failed*** 1 failures.
For whitespace errors, see the file /home/wstein/.sage//tmp/.doctest_integer.py
 [15.2 s]

CC: @jpflori

Component: porting: Cygwin

Reviewer: Jean-Pierre Flori

Issue created by migration from https://trac.sagemath.org/ticket/9172

[sagetrac-drkirkby]

comment:1

The correct answer is 2*sqrt(2), which is 2.8284271247461900976033774484193961...

So the expected value is 2.0239x10^-16 high, and result on Cygwin is 2.9760x10^-16 low. So the errors on Linux/OSX/Solaris is not much lower than on Cygwin. We can't really expect any more from a floating point number.

We could change the Expected value to 2.8284271247461... What I don't like about that, is then much larger errors can exist and them not be detected. But this is far from the only such case, so I suggest just changing the test to 2.8284271247461..., which will solve this.

Dave

[kcrisman]

comment:2

This file passed doctests on a build of mine on XP.

[kcrisman]

comment:3

But when doing the test by hand, the same thing happens.

[kcrisman]

comment:4

This now fails with

2.82842712474619

which I suppose is an improvement. Maybe we can use abs tol?

[jpflori]

comment:5

And the test passes for me (64bits W7 + 5.6.rc0).

[kcrisman]

comment:6

And the test passes for me (64bits W7 + 5.6.rc0).

In which case it might just be a 32-bit versus 64-bit issue. Did you try it by hand as well?

[dimpase]

comment:7

Replying to @kcrisman:

And the test passes for me (64bits W7 + 5.6.rc0).

In which case it might just be a 32-bit versus 64-bit issue. Did you try it by hand as well?

works for me, both ways. I think we can close this one.

[kcrisman]

comment:8

Hmm, I'm reluctant to not just change this a bit for 32-bit...

[jpflori]

comment:9

Yeah, I think we should make sure this passes on 32 bits.
I'll double check when I have the time to build on such a computer.

[jpflori]

comment:11

I have no problems on my 32 bits Windows 7 install, so Cygwin must have gotten better.
I really doubt Cygwin on XP would have a different behavior for this one, so let's close it.

[jdemeyer]

Reviewer: Jean-Pierre Flori