Re: [rkward-devel] test updates: what about plots?

[Thomas F. <tho...@ru...>]

Hi,

On Wednesday 29 July 2009, meik michalke wrote:
> i've just contributed my first svn commits :-) fixed a wrong variable name
> in a plugin (so, the tests begin to have effects...), and provided tests
> for all irt dialogs.

great! Thanks. The parameter estimation tests still seem to have 7 (instead of 
5) digits of default precision. Hence, they all fail, here. If you develop 
your tests inside a (saved) workspace, note that you will need to manually 
source / Shift+F9 test_framework.R on changes. (But in fact for this purpose, 
it's best not to use a saved workspace at all).

I'm still struggling a bit with some details regarding configurable keyboard 
shortcuts (there seems to be a crash associated with that: 
https://sourceforge.net/tracker/?func=detail&atid=459007&aid=2828002&group_id=50231), 
so you still have a chance to update the test standards before 0.5.1-rc1...

> that includes even the plot functions, i'm not sure about that, because the
> "results" are all saved to graph.png, hence overwritten, and graphs aren't
> checked in the tests (yet), are they? at least these tests show if
> something is seriously broken.

True. Of course, we *could* compare the graph.png-files (or a hash of these 
files), but I think that would cause a lot of false alarms on irrelevant 
changes of the plotting internals. We still have the checks for equality of 
gernerated R code, equality of warning messages from that R code, and of 
course information on whether the plugin could be started at all. So for 
plots, the tests are slightly less sensitive, but they should still detect a 
wide range of problems.

Regards
Thomas