[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: new DejaGnu team member

From: James Dein
Subject: Re: new DejaGnu team member
Date: Thu, 24 Jul 2003 22:33:14 -0700

On Thursday, July 24, 2003, at 9:50 PM, Dan Kegel wrote:

Ben Elliston wrote:

Dan Kegel <address@hidden> writes:

While we're on the subject of Dejagnu improvements, anyone
interested in souping it up to be able to take advantage of multiple
identical target systems so the tests can be run in parallel?

I am.

Done properly, a nice side effect of running tests on multiple targets
is that native testing or testing on simulators can take advantage of
multiprocessor systems.

Would you care to sketch out a plan?

The idea is to execute dg-style tests (where dg is a simple
high-level driver used by many of the gcc tests) in parallel,
and leave other styles of test alone.  (I think gcc is moving
more of the tests towards dg in order to make it easier to support
that other regression testing framework.)

This lets you easily pipeline compilation and execution,
since dg-style tests do nothing but run simple standalone programs
in batch mode on the target.
- Dan

It's not quite so simple as that, because when programs are downloaded to target boards, the harness has to wait for them to finish so that it can glean the exit status, which sometimes means grepping the status info from the program's output. For native testing, and for simulators where you can take advantage of the system's handling of concurrency (i.e., fork/exec/wait), it's not such a big deal.

I should probably mention at this point that I am currently working on enhancements to `dg' to handle multiple source files per test. We will be writing tests for this new feature and testing them -- and the DG enhancements -- regularly, on native. I will, of course, perform some non-native runs of the tests to make sure I haven't broken the `remote' system. Anyway, let's make sure our improvements dovetail rather than clash. The main trick (with my work) is to make sure the source files get downloaded to the "remote host" before the compiler gets invoked. Executing executables does not become harder (or easier :-). Handling dg-warning and friends requires sorting messages by source file. There's no rocket science involved, I think.

As for Python... I would rather use Elmo my new scripting language; Elmo is my answer to Tcl :-).

Jim Dein

DejaGnu mailing list

reply via email to

[Prev in Thread] Current Thread [Next in Thread]