I have been tossing around the idea of a tool to
aid in testing a VRML browser. Historically
we have tested 2D interfaces (e.g. Motif apps)
by synthesizing user input and taking pictures.
Then later the synthesized input is replayed
and the captured pictures are compared with the
current app state on the screen. This forms
a regression test.
This has lsad to a gesat deal of hair loss in
the R&D lab (:-) as the new pictures can be
different in any of a number of amusing ways.
Here are some examples:
-> the regression test was replayed on a
different machine, but the first capture
included the name of the host
-> a date or time appears on the screen, and
(surprise) the new date is different
-> a font glyph changed
-> etc. etc.
Generally a difference means that somebody
has to spend hours analysing minor differences,
or watch the screen while a new run is made to
recapture a new set of images.
So,
I am looking around for some sort of "fuzzy" comparison
tool for checking visual scenes that a browser might
produce. In my imaginings, the "fuzzy" tool could
be asked to look for a projection of a cube, cylinder
or cone in a picture, and maybe a color, and it would say "yes,
there is a red cube there". And it wouldn't matter
if the cube was displayed using in a width and
height of 20, 21, or 22 pixels.
Am I desaming to imagine such a tool, or has anybody
actually heard or even seen such a thing?
Thanks,
Jim Andreas
Hewlett-Packard Corvallis Oregon USofA
andreas@cv.hp.com