[Rd] Package checking: verifying plots

[Prof Brian Ripley]

On 22/01/2014 14:35, Therneau, Terry M., Ph.D. wrote:
> I pride myself on having an extensive and comprehensive test suite for
> the survival package.
>
> However, I don't have a good procedure for testing the plotting
> routines.  (Forget "good", I don't even have a third-rate procedure for
> this.)  This was brought home with the latest release; within 24 hours I
> got a note about a problem.  They y-axis range for a plot was calculated
> incorrectly, giving no warning or error, just a graph with 2/3 of the
> data outside the plot window.  It was a side effect of a change for
> another reason, and of course I had checked out that consequence
> thoroughly.
>
> Any good ideas for this issue?
>
> Terry Therneau
>

The R tests are comparisons of uncompressed PDF files (which are almost 
text files).  See the 'tests' directory.

Paul Murrell gave a talk about another method at an early DSC: 
https://www.stat.auckland.ac.nz/~paul/Talks/qa.pdf and now has a 
graphicsQC package.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595