[PATCH] undef MPROTECT_VDB for darwin?

[Boehm, Hans]

The collector does not currently scan data segments associated with dynamic libraries on Darwin.  (DYNAMIC_LOADING is not defined, and the support code in dyn_load.c is missing, since it's apparently not an ELF platform.)  Thus I doubt that any dynamically linked or multithreaded gcj-compiled code could run reliably on Darwin.  I suspect the test cases that do run are short enough not to need the GC, or they get very lucky.

Single-threaded, statically-linked code may be reliable.  For anything else, you're currently better off turning off the garbage collector (either at compile time, or with the environment variable).

The dynamic library issue fundamentally requires functionality similar to dl_iterate_phdr() in glibc.  Patches are welcome.

Hans

> -----Original Message-----
> From: Andreas Tobler [mailto:toa@pop.agri.ch]
> Sent: Friday, February 07, 2003 9:23 AM
> To: Boehm, Hans
> Cc: java-patches
> Subject: Re: [PATCH] undef MPROTECT_VDB for darwin?
> 
> 
> Boehm, Hans wrote:
> > That's fine.  Andrew and Tom had concluded that it would 
> make sense to reintegrate my latest tree into the gcc tree.  
> But it wasn't completely clear that someone with the right 
> CVS expertise had the time to do so.  This would also get 
> fixed as part of such a merge.
> > 
> > There are of course still problems with Darwin threads and 
> dynamic libraries ...
> 
> 
> Hm, do they really plan also on 3.3? If not it would be a little step 
> towards a cleaning up.
> 
> I know there are things with darwin, at least I tried the 
> latest tar and 
> the dylib was built :)
> 
> On some libgcj testcases I get spurious failures which end in 
> boehm-gc. 
> FileHandleGcTest.
> 
> I could help testing if you need some results.
> 
> So I wait until this merge will happen.
> 
> Thanks,
> Andreas
> 
> 
>