[Mono-list] mcs compiles on linux. Now what?

Paolo Molaro lupus@ximian.com
Fri, 8 Mar 2002 11:08:46 +0100

On 03/08/02 Piers Haken wrote:
> How long do you recon until we're dogfooding? ie ((mcs compiler built by
> csc) builds mcs) builds mcs), all running on linux?
> what do you need?

Well, the things that I outlined in my mail: optimizing corlib so that
compling a new test case takes less time, compiling programs and testing
The thing I'm most insterested now are programs that compile in Linux
but that don't work. I think I may have a bug in the Field token fixup
code, because in some case the field tokens get wrong and we end up
initializing them in the wrong way, leading to crashes.
So, this requires to build test cases, compiling and checking
the disassembled output. Or, uhm, it requires some brain on my part when
programming... After a small fix and test compile I can get
the compiled mcs to run to some extent (--help works :-).

Anyway, I need people to compile programs with mcs and check that the
resulting program works. Bonus points for samll test cases that fail
and for reports that track down the real problem and if it fails on
windows too. Extra points for reports with patches included! :-)

Another needed thing is to check if mcs3.exe works as well as mcs.exe
on linux (mcs.exe is compiled by csc, while mcs3.exe is compiled by mcs
itself and it should be more similar to the compiler we compile on
linux). I only used mcs3 to compile a few small programs and it seemed
to work, but it would be very useful to test it with the whole test
suite and with compiling itself again on linux. This is very time
consuming, but it's easily parallelizable with many testers:-)

If all goes well, we may have mcs compiling itself on linux next week.
The step after that is getting corlib compiled as well (on windows
first, here miguel may need your help as well).


lupus@debian.org                                     debian/rules
lupus@ximian.com                             Monkeys do it better