Re: Documenting test script arguments and revising runtest_file_p

[Jacob Bachmeyer]

Ben Elliston wrote:
> On Fri, Jan 11, 2019 at 10:38:05PM -0600, Jacob Bachmeyer wrote:
>   
> > Is the contents of the "runtests" variable part of the DejaGnu API
> > currently?  The variable is not documented and runtest_file_p exists
> > to abstract its meaning.
> >     
>
> No. It should only be procs. We should be providing accessor procs for
> any variables, ideally.
>   

The "testsuite file" command provides this for the *dir variables, but 
the "testsuite" multiplex call is intended for commands relating to the 
testsuite as a whole (another one on my TODO list is "testsuite depend" 
or "testsuite depends" for tracking dependencies not directly visible to 
the framework).

I have been trying to make a distinction between "testsuite options" and 
"test run options".  Essentially, a "testsuite option" is something 
related to the code in a testsuite and simply setting it probably will 
not work or be useful in general.  The "MULTIPASS" variable is an 
example of a testsuite option, as would be a future "API level" 
declaration.  A "test run option" is something that users can 
more-or-less freely set, like --verbose or the name of a compiler binary 
to use.

In this model, the "testsuite" command should probably stick to 
"testsuite options", although "testsuite file" indirectly accesses the 
*dir variables, which are "test run options", so the question then 
becomes what to name the new API call.  I will start with offering 
"testsuite script args" defined to return what is currently the second 
element in $runtests.  If we also want the first element of $runtests 
(and that is not entirely redundant to [info script] or we want it 
anyway), I suggest "testsuite script name" for that purpose.

Comments?


-- Jacob