Re: Maintainers, porters, and burden of porting

[Kurt Roeckx <kurt@roeckx.be>]

On Wed, Aug 31, 2011 at 02:42:41PM +0200, Lucas Nussbaum wrote:
> On 31/08/11 at 12:58 +0100, Ben Hutchings wrote:
> > On Wed, 2011-08-31 at 11:57 +0200, Lucas Nussbaum wrote:
> > [...]
> > > But a different thread library that has clear POSIX compliance bugs[*]
> > > is the kind of things that make me fear that many more packages than we
> > > see currently are broken on kfreebsd. And I'm not sure that it's where
> > > we want to spend our manpower.
> > > 
> > > [*] due to linuxthreads: #639658 [kfreebsd] waitpid from a thread does
> > > not work for child processes created by other threads.
> > > there's also some signals+thread fun.
> > 
> > Of course, Linux had those bugs for many years, so most multithreaded
> > programs that run on Linux are probably tolerant of them.
> 
> But Linux hasn't had them since many years too (NPTL in Debian: 2003),
> so multithreaded programs that were written more recently might not be
> so tolerant.

If my memory is any good, amd64 was the first arch that was released
with NPTL support.  We decided to switch to NPTL directly and not
support a 2.4 kernel, and didn't have LinuxThread support at all.
But at that time we didn't release officially with Debian.  But I
think glibc build both the LinuxThreads and NPTL version, but you
got the headers of LinuxThreads on all arches except amd64, and
the NPTL headers on amd64.  And Sarge was released like that.

During etch we decided to completly drop the 2.4 kernel and some
more arches switched to only use NPTL, and this started somewhere
around 2006. 

Because amd64 switched to NPTL before any other arch we also saw
alot of problems because some functions weren't available anymore
or behaved differently.


Kurt