Re: dejagnu(1) multi-launcher and dejagnu-report-card

[Jacob Bachmeyer]

Ben Elliston wrote:
> On Tue, Nov 06, 2018 at 06:59:34PM -0600, Jacob Bachmeyer wrote:
>   
> > Second, the first subcommand to implement would be "dejagnu
> > report-card", for which I already have shell and Tcl
> > implementations.  This would then be usable as part of the DejaGnu
> > testsuite to provide a nice summary at the end of a testsuite run.
> >     
>
> There is a pretty high maintenance overhead of maintaining multiple
> implementations.

DejaGnu "MULTIPASS" can help with this -- run the same tests against all 
implementations, ensuring that any regressions will be caught.  
Maintenance for the report-card feature overall should be tiny -- the 
output of DejaGnu is unlikely to change significantly, since the kind of 
testing DejaGnu supports is mature and well-understood by now.

> Why not just decide, conservatively, what the right
> language is for the job, and we'll just make it a prerequisite to be
> checked by 'configure'? If the tool (eg, AWK) is not available, we can
> just abort the configure script, or not install the subcommand script.
>   

Part of the problem is that "right language for the job" is somewhat 
dependent on the size of the workload.  For small testsuites, even the 
shell version performs well, but the advantages AWK holds over Tcl seem 
to increase with the number of tests in the suite.  It is also possible 
that I am missing significant optimizations in the Tcl version.  For 
example, does Tcl byte-compile loop bodies in open code or only in 
procedures?  The Tcl version is currently a simple linear script and 
seems to spend most of its time reading input.


-- Jacob