cvs commit: src/lib/libpthread Makefile src/lib/libpthread/test
thr_create.c thr_find_thread.c thr_init.c thr_kern.c thr_nanosleep.c
thr_private.h thr_sig.c thr_sigmask.c ...
eischen at vigrid.com
Thu Jul 17 16:30:20 PDT 2003
On Thu, 17 Jul 2003, Julian Elischer wrote:
> This makes it possible to make the same code run in the same mode as
> we have not compared them for performance yet.
> it is also not fully decided how this will be done in production..
> will we have 2 libraries and selct one? Or will it be possible to
> use a run-time selector?
> At the moment it is a compile time selection as to which you make and
> a run-time (via libmap) decision wich you link to..
There is not much code removed when compiling for 1:1,
and David actually had an earlier version of it that
was based on just a variable being set to true or false.
If the library knew (at init time) if it was called
libpthread or libthread, it could automatically do the
> We'll probably make a libpthread-1:1.so and a libpthread-m:n.so.
> As they are the same code the interface is pretty much guaranteed to be
> the same.. Note that like libthr, the 1:1 version is incapable of doing
> process-scope threads, but otherwise they are compatible.
I'd like to make libpthread and libthread (ala Solaris).
Note that even in 1:1 mode, POSIX semantics for priority
ceiling and inheritence mutexes, as well as wakeup ordering
for condition variables, still work.
More information about the cvs-all