On Sun, Nov 08, 2015 at 11:25:00AM -0200, Antonio Terceiro wrote:
> On Sat, Nov 07, 2015 at 08:46:10PM -0500, James McCoy wrote:
> > Except there's another test that still fails when ruby 2.2 is actually
> > used (unlike in the build log you provided or any of the testing I had
> > done recently), due to code trying to modify nil which is now frozen,
> > along with true/false.
>
> do you have the log? nil is actually always frozen, so this is likely
> something that used to return a String, but now returns nil.
No, this part of the test suite hasn't changed in years. It was passing
nil into a function, eventually causing nil to be passed in as "self" to
rb_set_pool:
VALUE
rb_set_pool(VALUE self, VALUE pool)
{
if (NIL_P(pool)) {
VALUE old_pool = rb_ivar_get(self, id___pool__);
rb_hash_aset(rb_pools(self), rb_obj_id(old_pool), old_pool);
rb_ivar_set(self, id___pool__, Qnil);
} else {
if (NIL_P(rb_ivar_get(self, id___pool__))) {
rb_ivar_set(self, id___pool__, pool);
} else {
rb_hash_aset(rb_pools(self), rb_obj_id(pool), pool);
}
}
return Qnil;
}
I sent a patch[0] which appears to be the fix for it, but I'd like to
wait a day for feedback.
http://mid.gmane.org/20151109042607.ga13...@freya.jamessan.com
> > I'm sending a patch upstream for the general test-unit problem, but I
> > still don't have a solution for the frozen nil problem.
>
> subversion is the one pending package with enough reverse dependencies
> to stop the ruby transition from going on. How important is that one
> test, and the Ruby bindings in general? Can we have a temporary
> workaround to allow us to go on with the transition?
If I don't get any push back from upstream, I should have an upload
ready for subversion in the next day or so. Starting a test build with
the test-unit and frozen-nil patches now.
Cheers,
--
James
GPG Key: 4096R/331BA3DB 2011-12-05 James McCoy <james...@debian.org>
