Re: Adding files for Ada

[Laurent Guerby <guerby at acm dot org>]

[About the timing, if the commit only add an ada directory, it is
unlikely to break anything, so could it be done now? We have a long
weekend in France :]

> Does this use dejagnu?  

No, this is a simple (ugly :) 200 lines sh script that downloads the
test suite from the web, massage it and run a selected part of it
producing a log and pass/failure statistics.

A version of my script was kindly put on the RTEMS FTP server by Joel
Sherrill:

<ftp://ftp.oarcorp.com/pub/rtems/snapshots/contrib/acats>

This is not completely up to date with what I have on my disk, and I
haven't checked the latest modification done to the official suite,
but you should get a feeling of what it is if you're interested.

> A dejagnu-based test harness is clearly to be preferred, for better
> integration with the existing testsuites and tools that use them,
> working for all the embedded targets dejagnu supports, etc., though
> a non-dejagnu one can always be committed and converted to dejagnu
> later.

I agree with this. On the DejaGNU front, what should I mimic?  (I
think from several discussion I monitored that there is some old stuff
and some new (= better :) stuff out there.)

Running the ACATS is done in four steps:

1. customize a few bits of source to the compiler and testing
environment used. My script does some of that with sed (eg setting
directory of IO test files).

2. compile a set of support file, including some non-Ada source in
order to test "Interface with other languages" (I did C but not
Fortran yet). All subsequent compilations should "-I" the support
directory

3. run 3 tests testing that various pass/fail reporting
facilities work (this can be skipped)

4. for all executable tests, compile/bind/link/run and check for
PASSED in the output log.

There are of course a whole bunch of details to be sorted out, but I
got the whole thing running, so it's just a matter of choosing better
solutions when appropriate now.

I know that Joel Sherrill is interested in getting testing to work for
cross-gcc, is DejaGNU easily usable there? A little customization of
the script did the work, I'm afraid it'll be much more complicated
with DejaGNU but I can be wrong.

-- 
Laurent Guerby <guerby@acm.org>