These are chat archives for dropbox/pyston

30th
Sep 2015
An Long
@aisk
Sep 30 2015 03:56
@undingen @Daetalus looks strange, #942 passed full check on my local 14.04 vm ...
Dong-hee Na
@corona10
Sep 30 2015 04:22
@kmod can you review #941 ??
Dong-hee Na
@corona10
Sep 30 2015 05:03
@undingen i wrote on src/analysis/type_analysis.cpp by
void* visit_ellipsis(AST_Ellipsis* node) override { return Ellipsis; }
Marius Wachtler
@undingen
Sep 30 2015 09:38
I think you have to return a ConreteCompilerType* (the void* return type doesn't make this clear :-() you get this by calling typeFromClass() in your case it's typeFromClass(Ellipsis->cls)
or typeFromClass(ellipsis_cls)
Marius Wachtler
@undingen
Sep 30 2015 09:44
and type analysis and the irgen have always to return the same type that's why I mentioned yesterday that you also have to change type_analysis (but I didn't read the change carefully enough to notice that it returned a typeFromClass(type_cls) instead of typeFromClass(ellipsis_cls)
Marius Wachtler
@undingen
Sep 30 2015 10:00
the compiler types are normally this all uppercase vars like SET, STR etc... in case you are wondering
Dong-hee Na
@corona10
Sep 30 2015 11:21
is there any tip why boxed_cls is null??
ConcreteCompilerType* typeFromClass(BoxedClass* c) {
    assert(c);
    return NormalObjectType::fromClass(c);
}
It occured after i add ELLIPSIS = typeFromClass(ellipsis_cls); on src/runtime/types.cpp
Marius Wachtler
@undingen
Sep 30 2015 11:25
this has probably something todo with the init sequence ellipsis_cls will probably created later on
But because the ELLIPSIS is uncommen I think it's fine to not even create a ELLIPSIS variable but instead to always call typeFromClass(ellipsis_cls) inside type_analysis (like we currently do even for the unicode_cls...)
if you can't access the ellipsis_cls, Ellipsis->cls should work
Dong-hee Na
@corona10
Sep 30 2015 12:10
thanks that was problem. Ellipsis->cls is might be better than make
#define ellipsis_cls for making define hell
Sun
@Daetalus
Sep 30 2015 15:59
This message was deleted

oh I just saw that you merged my change so now one only needs to remove the skip-if line inside multiprocessing_ctypes_test.py test

Hi @undingen , If my understanding of #937 was correct, the multiprocessing_ctypes_test.py test will still fail in CI. We can't remove the skip-if line for now. Would you mind to confirm it?

Marius Wachtler
@undingen
Sep 30 2015 16:04
yes exactly this test fails on travis-ci (and only there :-() same thing for the ctypes_test.py inside your PR
I just meant so say with the quoted message that in order to trigger it on travis ci one has to remove the skip-if line - sorry for the confusion
Sun
@Daetalus
Sep 30 2015 16:06
I see. Thanks!