Re: felix regression on arm?
On Mon, 2007-10-01 at 14:12 +0200, Julien Cristau wrote:
> On Mon, Oct 1, 2007 at 11:15:48 +0200, Stefano Zacchiroli wrote:
>
> > John,
> > do you have any explanation for the felix test failure on arm? Here is
> > some logs you can look at:
> > http://buildd.debian.org/fetch.cgi?&pkg=felix&ver=1.1.3-2&arch=arm&stamp=1190987953&file=log
> >
> > I actually don't remember if ATM arm is a release architecture (Julien?)
> > but if it is the above failures would prevent felix to enter testing
> > and, unless we do something, the OCaml transition to happen.
> >
> Is felix built using ocamlopt on arm?
The Felix build system uses ocamlopt.opt if it detects it,
then tries ocamlopt, then ocamlc.opt, then ocamlc.
Mike Furr or someone else knowing the repository holding the
packaging would have to remove the build-depend on
ocamlopt and ocamlopt.opt for the ARM for Felix config to
fail the detection and revert to ocamlc.opt or ocamlc.
> There are bugs in the asm code
> generation on this architecture, which might explain this problem.
It might, however the first failure with a large file could
easily be an out of memory. The standard library list test,
however, is quite small, and its failure couldn't be
explained that way.
I would have expected a code generation bug to cause many more
of the tests to fail: the point being that if the Felix compiler
flxg is bugged, it would be surprising only 1-2 tests exercised
the bugged part of the generated code -- possible, since they're
unit tests, but still unlikely IMHO.
Having said that I regularly make bugs in the compiler which
are detected by 0-2 tests.. if I'm unlucky it is 0 :)
--
John Skaller <skaller at users dot sf dot net>
Felix, successor to C++: http://felix.sf.net
Reply to: