libgcj and the NPTL posix threads implementation

Mon Mar 1 21:28:00 GMT 2004

Jeff Sturm writes:
 > On Mon, 1 Mar 2004, Andrew Haley wrote:
 > > Eventually the LD_ASSUME_KERNEL kludge will go away, but until then we
 > > have to deal with the possibility that people will use gcj with
 > > libraries that need to use LD_ASSUME_KERNEL=2.3. We don't want to
 > > break libgcj in such situations.
 > 
 > If "we" == "Linux distribution vendors", then yes definitely. But
 > I'd expect nearly everyone building and using their own gcj would
 > want Anthony's patch.
I don't think so. I have worked fairly recently on systems that have
a working NPTL but gdb doesn't grok NPTL threads. In this case, the
only way to find important bugs is to use LD_ASSUME_KERNEL=2.3 in
order to do debugging. Without this ability I would have been up a
creek without a paddle.
It should be simple to use a function pointer to switch _Jv_ThreadSelf
to different personalities, and the runtime cost would not be huge.
Andrew.