lilypond-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: new comparisons available


From: Richard Schoeller
Subject: Re: new comparisons available
Date: Tue, 19 Dec 2006 13:13:54 -0500

Han-Wen,

If you overlay the two images, all you need is to display the pixels
that are different in a different color.  We used that approach when
testing regressions in the Motif toolkit back in '93/'94.  It is very
valuable.  The tool actually let you display new, old and overlay.  The
overlay was particulary good at catching the 1 pixel differences.

The one thing that I seen with that is that the baseline test results
may not port well.  The fonts may be slightly different from one
distribution to the next, resulting in very small images differences.
An example of this is that the "fixed" font on Solaris was not the same
as on OSF/1 or HP-UX.  There were a few places whare an extra pixel was
found on the end of a line. Differences in versions of some fonts might
introduce "fixes" in font metrics.  &c.

In regards your observation about bounding box differences... I would
say that if the bounding boxes have changed then that is a significant
change.  The overlay approach may be seem extreme for identifying that,
in that the images will be completely skewed.  However, it will make the
test result hard to ignore  8^).  Then you can decide to re-baseline the
test or undo the change.

BTW, I think that finding regressions is a major issue for LilyPond
development.  My experience with upgrades has been that I have to go
around fixing the layouts of my scores after an upgrade.  Often these
are minor changes in the placement algorithms that turn into collisions.
Having this kind of test will make it easier to remember to include
feedback on those kinds of changes in the upgrade docs.  These are
things that may look like features but should still be included.

Dick

> > What do you think about providing a third image which directly
> > overlays the two versions, using different colours, e.g. red and
> > black?
> 
> I think it will make things more confusing.
> 
> Note that it is possible that images are the same but output-distance
> thinks that images differ, since output-distance looks at bounding boxes. 
> 
> If you're interested in experimenting, I suggest you play
> with ImageMagick and buildscripts/output-distance.py
> 
> 
-- 
Dick Schoeller
mailto:address@hidden
http://schoeller.hsd1.ma.comcast.net/
781.449.5476





reply via email to

[Prev in Thread] Current Thread [Next in Thread]