Apology

[Reuben Thomas]

Richard Stallman just pointed out that my recent release announcement for
GNU Zile 2.3.1, in which I described DejaGnu as "DejaGnu proved to be
insufficiently portable, and too flaky", was rather rude. I agree. I'm sorry
if I caused any offense.

My problems with DejaGnu were however not caused by bugs or indeed by design
flaws, but by a combination of lack of portability (as I also said) and by
flakiness in the underlying expect/tcl combination. I'm sorry I didn't make
that clear.

--
http://rrt.sc3d.org/ | historian, n.  a broad-gauge gossip (Bierce)


_______________________________________________
Bug-dejagnu mailing list
Bug-dejagnu@...
http://lists.gnu.org/mailman/listinfo/bug-dejagnu

[Rob Savoye]

Reuben Thomas wrote:

> for GNU Zile 2.3.1, in which I described DejaGnu as "DejaGnu proved to
> be insufficiently portable, and too flaky", was rather rude. I agree.
> I'm sorry if I caused any offense.

  Dejagnu is insanely portable, and runs on dozens of systems, both
natively and cross. Flaky I don't know about, but obscure and overly
complex comes to mind. Good cross testing of toolchains is actually a
difficult problem, which made things somewhat complicated, plus DejaGnu
was developed while also using also it for toolchain testing for release
at Cygnus.

> design flaws, but by a combination of lack of portability (as I also
> said) and by flakiness in the underlying expect/tcl combination. I'm
> sorry I didn't make that clear.

  Tcl has become unmaintained, and expect not much better. Course this
is 18 years after DejaGnu was written. If I ever did it again, I'd
probably use Python, but that would be a big project, and there aren't
the resources (funding and bodies) to do that in a way that wouldn't
cause problems in GNU toolchain testing. I still use DejaGnu on many,
many systems with my current projects, and haven't really seen any
portability problems, so I'm curious what platform you were on, and what
your problems were.

        - rob -


_______________________________________________
Bug-dejagnu mailing list
Bug-dejagnu@...
http://lists.gnu.org/mailman/listinfo/bug-dejagnu

[Reuben Thomas]

On Tue, 24 Feb 2009, Rob Savoye wrote:

> Reuben Thomas wrote:
>
>> for GNU Zile 2.3.1, in which I described DejaGnu as "DejaGnu proved to
>> be insufficiently portable, and too flaky", was rather rude. I agree.
>> I'm sorry if I caused any offense.
>
>  Dejagnu is insanely portable, and runs on dozens of systems, both
> natively and cross. Flaky I don't know about, but obscure and overly
> complex comes to mind. Good cross testing of toolchains is actually a
> difficult problem, which made things somewhat complicated, plus DejaGnu
> was developed while also using also it for toolchain testing for release
> at Cygnus.

I had Nelson Beebe building Zile on about 30 different systems, and on
some, e.g. Mac OS X, DejaGnu wouldn't work.

>  Tcl has become unmaintained, and expect not much better.

Tcl seems to have received a lot of attention recently, so perhaps there's
hope there.

I'm sorry, I forget the exact details of my last problem, which was to do
with detecting processes finishing, or perhaps killing them, but when I read
up on it it seemed to be something I couldn't reliably fix in Tcl.

I also spent ages trying to get around timing problems with ncurses. I did
in the end solve this reliably, but the tests still ran slowly, as waiting
for the test processes to exit took time. In the end rewriting the entire
test suite not to use DejaGnu took about the same time as solving that
problem had, and ended up with one less build-time dependency, and a test
suite that ran much faster, and was simpler to write tests for (because I
could write most of them in Emacs Lisp rather than having to program them
interactively).

I don't doubt that if I had more programs which needed interactive testing
and which were perhaps not using ncurses or didn't have the natural
extensibility of an Emacs-like editor, then it would indeed be more natural
to use DejaGnu; I just had a particular set of circumstances in which that
didn't, in the end, seem to be the best way.

--
http://rrt.sc3d.org/ | Brevity is the soul


_______________________________________________
Bug-dejagnu mailing list
Bug-dejagnu@...
http://lists.gnu.org/mailman/listinfo/bug-dejagnu