Aw: Re: changes to graphics.cc and subplot.m

[Stefan Mahr]

>>>>>>>>> oops!  The test only fails for me the 1st time it is run, and only 
>>>>>>>>> because the axes handles are in the wrong order.
>>>>>>>>> 
>>>>>>>>> Ben
>>>>>>>> 
>>>>>>>> It's weird. With your test script I get the same results as you.
>>>>>>>> However, the resulting output still has different y-axis size for both
>>>>>>>> plots.
>>>>>>>> 
>>>>>>>> If I copy-paste your example block-wise in interactive mode, I get more
>>>>>>>> errors. Could it be some kind of timing problem?
>>>>>>> 
>>>>>>> The extra errors are because the test aborts after the first error.
>>>>>>> 
>>>>>>>> -----------
>>>>>>>> 
>>>>>>>> graphics_toolkit fltk
>>>>>>>> close all
>>>>>>>> hf = figure (gcf (), "__graphics_toolkit__", "fltk", "visible", "on");
>>>>>>>> 
>>>>>>>> 
>>>>>>>> subplot(2,1,1); plot(rand(10,1)); subplot(2,1,2); plot(rand(10,1))
>>>>>>>> hax = findall (gcf (), "type", "axes");
>>>>>>>> positions = cell2mat (get (hax, "position"));
>>>>>>>> outerpositions = cell2mat (get (hax, "outerposition"));
>>>>>>>> looseinsets = cell2mat (get (hax, "looseinset"));
>>>>>>>> tightinsets = cell2mat (get (hax, "tightinset"));
>>>>>>>> close all
>>>>>>>> 
>>>>>>>> 
>>>>>>>> subplot(2,1,1); plot(rand(10,1)); subplot(2,1,2); plot(rand(10,1))
>>>>>>>> hax = findall (gcf (), "type", "axes");
>>>>>>>> assert (cell2mat (get (hax, "position")), positions, 0.001)
>>>>>>>> assert (cell2mat (get (hax, "outerposition")), outerpositions, 0.001)
>>>>>>>> assert (cell2mat (get (hax, "looseinset")), looseinsets, 0.001)
>>>>>>>> assert (cell2mat (get (hax, "tightinset")), tightinsets, 0.001)
>>>>>>>> 
>>>>>>>> ------------------------
>>>>>>>> 
>>>>>>>>> assert (cell2mat (get (hax, "position")), positions, 0.001)
>>>>>>>> error: assert (cell2mat (get (hax, "position")),positions,0.001) 
>>>>>>>> expected
>>>>>>>> 0.13000   0.58384   0.77500   0.32733
>>>>>>>> 0.13000   0.12384   0.77500   0.32733
>>>>>>>> but got
>>>>>>>> 0.13000   0.60122   0.77500   0.30994
>>>>>>>> 0.13000   0.18964   0.77500   0.26152
>>>>>>>> maximum absolute error 0.0658047 exceeds tolerance 0.001
>>>>>>>> error: called from:
>>>>>>>> error:   /usr/share/octave/3.7.5/m/testfun/assert.m at line 235, 
>>>>>>>> column 5
>>>>>>>>> assert (cell2mat (get (hax, "outerposition")), outerpositions, 0.001)
>>>>>>>>> assert (cell2mat (get (hax, "looseinset")), looseinsets, 0.001)
>>>>>>>>> assert (cell2mat (get (hax, "tightinset")), tightinsets, 0.001)
>>>>>>>> error: assert (cell2mat (get (hax, "tightinset")),tightinsets,0.001)
>>>>>>>> expected
>>>>>>>> 0.042857   0.013122   0.000000   0.026316
>>>>>>>> 0.042857   0.013199   0.000000   0.026316
>>>>>>>> but got
>>>>>>>> 0.04286   0.09122   0.00000   0.01696
>>>>>>>> 0.04286   0.13964   0.00000   0.00000
>>>>>>>> maximum absolute error 0.126443 exceeds tolerance 0.001
>>>>>>>> error: called from:
>>>>>>> error:   /usr/share/octave/3.7.5/m/testfun/assert.m at line 235, column 
>>>>>>> 5
>>>>>>> 
>>>>>>> The second error is the one I hoped to catch.  I'll work in fixing the 
>>>>>>> test.
>>>>>>> 
>>>>>>> Ben
>>>>>> 
>>>>>> Does the attached version isolate the second error?
>>>>>> 
>>>>>> Ben
>>>>> 
>>>>> It's not related to the test itself, but line
>>>>> hf = figure (gcf (), "__graphics_toolkit__", "fltk", "visible", "on");
>>>>> fails for me. I need to add 'graphics_toolkit fltk' in the first line to 
>>>>> make it work.
>>>>> 
>>>>> 
>>>>> If the test in run in script mode, it always pass. While testing Rik's 
>>>>> plot issue I realized, that subplot is not neccessary for catching the 
>>>>> y-axis bug. The optical difference is not as big as with subplot, so I 
>>>>> overlooked it in the past. With attached patch the bug is reproducable in 
>>>>> script mode. Please note the 'sleep(2)' after the plot command. Without 
>>>>> this sleep, the test pass without error.
>>>>> 
>>>>> 
>>>>> Stefan
>>>> 
>>>> Stefan / Dimitri / others,
>>>> 
>>>> I'm still hoping to add some tests that demonstrate the problem.
>>>> 
>>>> Does running this script produce one expected failure and 1 passing test?
>>>> 
>>>> Ben
>>>> 
>>> 
>>> No, since it's inverted. The first test passes, the second fails.
>>> 
>>> However, it's still not 100% sure:
>>> 
>>> octave-cli:44> test stefan; test stefan
>>> [...]
>>> PASSES 2 out of 2 tests (1 expected failure)
>>> PASSES 2 out of 2 tests
>>> 
>>> When calling 'test stefan' twice in one command line, the second run passes 
>>> both tests. It also happens when you copy&paste
>>> 
>>> test stefan
>>> test stefan
>>> 
>>> to command line:
>>> 
>>> octave-cli:68> test stefan
>>> [...]
>>> PASSES 2 out of 2 tests (1 expected failure)
>>> octave-cli:69> test stefan
>>> PASSES 2 out of 2 tests
>>> octave-cli:70> test stefan
>>> PASSES 2 out of 2 tests
>>> 
>>> If you wait until the plot window is closed before starting new test, you 
>>> get the expected failure.
>>> 
>>> Stefan
>>
>> Stefan, can I impose on you to modify the tests so that they give the result 
>> we desire (i.e. one failure and then a passing test)?
>> 
>> Ben
>> 
>> 
> 
> I wish I could.
> 
>  octave-cli:19> for i=1:10, stefan; end
>  PASSES 2 out of 2 tests (1 expected failure)
>  PASSES 2 out of 2 tests
>  PASSES 2 out of 2 tests
>  PASSES 2 out of 2 tests
>  PASSES 2 out of 2 tests
>  PASSES 2 out of 2 tests
>  PASSES 2 out of 2 tests (1 expected failure)
>  PASSES 2 out of 2 tests (1 expected failure)
>  PASSES 2 out of 2 tests
>  PASSES 2 out of 2 tests
> 
> 
> Stefan
> 

Ben,

the problem above is fixed now (build with lastest changeset 
http://hg.savannah.gnu.org/hgweb/octave/rev/ea19ea629a09 )

Stefan