[tlbuild] axodraw2 test failure on 32-bit Cygwin

[John Collins]

Hi,

Ken: You've actually uncovered a small bug.  The visual appearance of 
one object is incorrectly sensitive to rounding error in a normal use, 
including in axodraw2's manual :-)  I'll correct that in the next release.

Akira: Yes, you are of course perfectly correct; it's dangerous to use a 
test with real values, at least when non-trivial computations have been 
done.

It would be OK not to do the test for axodraw2; it's in very standard C, 
so I wouldn't expect problems.  On the other hand Karl suggested a test, 
and that did seem a good idea.

To keep the test, I suggest replacing the files test.ax1 and 
test.ax2-std by the attached ones.  They only exercise an elementary 
part of the program, but should be immune to rounding error issues as 
far as I can see.

The test would still need diff to invoked with an option set to ignore 
the different line endings.  (There's an unfortunate incompatibility 
between POSIX and linux diff here!)  I think "diff -b" would do the job.

I can go either way.  Let me know what it best.

For the CTAN distribution, I can still provide the full test, since it 
would be useful to someone modifying the program. But the automake 
configuration would either invoke no test or the simple test.

Thanks for bringing up these issues.
John


>> Building from r46610, I'm getting an axodraw2 test failure on 32-bit 
>> Cygwin, but not on 64-bit Cygwin.  The test log is attached.  Please 
>> let me know if there is anything I can do to help track this down.



> In the case of Visual Studio 2015, both of 32bit and
> 64bit versions generate the same test.ax2, and the
> test fails. I attach test.diff (diff test.ax2-std test.ax2 >test.diff).
> 
> A test of files with real values is difficult.
> I propose not to test axodraw2.
> 
> BTW, axohelp.exe contained in the package generates the same result
> as my one, except for the EOL of \r\n. My one has EOL \n,
> since I changed "w" ---> "wb".
> Thus the test fails even if I use the binary given in
> the package.
> 
> Thanks,
> Akira

-------------- next part --------------
[1] AxoLine 0 3 100 3 0 0  0 0 0 0 0 0 0 0  0 1;
-------------- next part --------------
\axo at setObject{1}%
{AxoLine 0 3 100 3 0 0  0 0 0 0 0 0 0 0  0 1;}%
{ 1 w 0 3 m 100 3 l S}