R devel - Feb 2007 - proposal to allow just transcript files (output only) in the 'tests' directory

Currently, as far as I can see, both input and output fles must be 
supplied in the 'tests' directory (the input in a '.R' file, and
the
output in a corresponding '.Rout.save' file.)  Often, the '.R'
file is
redundant, as it could be reconstructed by simply stripping the commands 
out of the '.Rout.save' file.  E.g., the command
$ sed -n -e 's/^[>+] //p' ./src/library/stats/tests/nls.Rout.save
produces the same R commands as in ./src/library/stats/tests/nls.R

If I've missed something, and this is already supported, my apologies, &
please let me know.

If indeed this facility is not present, would anyone from the R core 
group consider an addition to allow '.Rt' (or '.Rt.save') files
in the
'tests' directory?  These would by just like the '.Rout.save'
files,
except that no corresponding '.R' file would be required -- the R 
commands would be extracted from them by simple line processing.  Of 
course, '.R' and '.Rout.save' files could still be used where
necessary
or preferred.

The primary advantage of this change would be to make it easier to add 
and maintain tests -- only one file would need to be created or changed, 
and no redundant information would be required.  My view is that tests 
should be as easy as possible to create and maintain, and the current 
system adds a small amount of unnecessary difficulty.

If such an addition would be considered, I would be happy to supply 
patches to implement it (and test them in a build under Windows & 
Linux).  I have tested a prototype, and it seems pretty simple to add 
this: as far as I can see, small changes are needed in three files:

./share/make/tests.mk
./src/gnuwin32/fixed/share/tests.mk
./bin/check

Please respond with 
comments/suggestions/corrections/pointing-out-of-gotchas/etc!

thanks,

Tony Plate

On Tue, 20 Feb 2007, Tony Plate wrote:
> Currently, as far as I can see, both input and output fles must be
> supplied in the 'tests' directory (the input in a '.R'
file, and the
> output in a corresponding '.Rout.save' file.)  Often, the
'.R' file is
> redundant, as it could be reconstructed by simply stripping the commands
> out of the '.Rout.save' file.  E.g., the command
> $ sed -n -e 's/^[>+] //p'
./src/library/stats/tests/nls.Rout.save
> produces the same R commands as in ./src/library/stats/tests/nls.R
Some output lines start with "+ ".  E.g., in
    gamlss/tests/manual.Rout.save
I see
    > mod3<-stepAIC(mod1,
scope=list(lower=~1,upper=~(x1+x2+x3+x4+x5+x6)^2))
    Distribution parameter:  mu
    Start:  AIC= 319.16
     y ~ x1 + x2 + x3 + x4 + x5 + x6

            Df    AIC
    + x4:x5  1 307.07
    + x1:x5  1 316.94
    - x6     1 317.16
so R would try to parse "x4:x5  1 307.07".

You could use awk so the "^+ " lines had to immediately
follow "^> " (or a properly preceded "^+ "), but I
suspect
there will still be problems.

I have experimented with similar things to compare Splus
and R output for the same commands.  Having the original
input lines helps there.  The hard part is parsing the output
so it can be recognized as a matrix, a vector with names,
an expression, lm() output, etc.  Then I could see that
the following outputs are equivalent:

   R    :
   R    : Call:
   R    : lm(formula = log(x) ~ x, data = list(x = 1:10))
   R    :
   R    : Coefficients:
   R    : (Intercept)            x
   R    :      0.2432       0.2304
   R    :

   Splus: Call:
   Splus: lm(formula = log(x) ~ x, data = list(x = 1:10))
   Splus:
   Splus: Coefficients:
   Splus:  (Intercept)         x
   Splus:    0.2432038 0.2304068
   Splus:
   Splus: Degrees of freedom: 10 total; 8 residual
   Splus: Residual standard error: 0.2388027

This would also be handy for comparing results in different versions of
R, where low level printing routines may be changed a bit.

----------------------------------------------------------------------------
Bill Dunlap
Insightful Corporation
bill at insightful dot com
360-428-8146

 "All statements in this message represent the opinions of the author and
do
 not necessarily reflect Insightful Corporation policy or position."