Re: [hatari-devel] fixing errors reported by GCC 10 -fanalyzer

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 13/05/2020 à 20:17, Eero Tamminen a écrit :
> Hi,
>
> On 5/13/20 7:59 PM, Nicolas Pomarède wrote:
> > it depends on the case, most of the time interpreting the results 
> > means looking at the screen and  checking if some pixels are aligned 
> > or in different colors (depending on which video test is performed), 
> > which is not usable with natfeats.
>
> We can always add NatFeats call to request pixel
> color(s) to be checked at specified screen
> position(s), and output PASS or FAIL string
> depending on whether there's a match.
>
> There could be new "NF_TEST" NatFeats call for
> doing emulator tests.  One of the call sub-IDs
> would do this particular test, e.g. using
> following stack signature:
> ScreenColorMatch(uint16_t x, uint16_t y, uint16_t pixels, color_t *colors);

This would be really too complicated to check pixels at specified 
position, the pattern can be changing, but the result could be still 
correct.

well, at least that's how some my own test work : it's easy for a human 
to look at the screen and see if it works or not, it would be 
overkill/more complicated to write a program that would compare 
different screenshot or look for a pattern.

There could be other ways to test for overscan, for example check the 
value of the video counter after doing some frequency switches, but 
that's not how I did it for my own use.
Maybe someone will feel like writing such program (even if in the case 
of the overscan, it's someone much faster to run 3 or 4 specific demo 
and see if they work ("Closure" by Sync can be a good candidate)

Nicolas