[boost] Re: New linux thread impl needed for NPTL?

From: Alexander Terekhov (terekhov_at_[hidden])
Date: 2003-05-24 08:54:53

graydon hoare wrote:
> 
> Alexander Terekhov <terekhov_at_[hidden]> writes:
> 
> > Yes. The wrong thing is that NPTL *DOESN'T* "translate" thread
> > cancelation (and thread exit) into a C++ exception... ala
> 
> aha. well, I'm no master of understanding these matters, but having
> now read your exhaustive thread on the matter, I can see that it might
> be nice to translate the cancellation event into a named exception, as
> you write. on the other hand, it also seems to be an implementation
> defined choice how to map into the C++ runtime. so, it may not be
> good, but I think it's at least roughly conformant.

Yeah, that's the problem. Well, keep an eye (it's coming soon) on 
the "Upcoming ISO/IEC <thread>... and <pthread.h> -> <cthread> 
transition" thread on comp.programming.threads... I'll 
pthread_create() it in an hour or two from now.

> 
> what I'm not certain of is whether you think things used to be ok with
> linuxthreads 

http://google.com/groups?selm=c29b5e33.0202151543.19a850a7%40posting.google.com

>              and got *worse* with NPTL, 

To some extent, NPTL is certainly MUCH better than rather archaic 
linuxthreads. However, don't run a bank (or tank) on it, yet. E.g.:

http://listman.redhat.com/archives/phil-list/2003-March/msg00123.html

The latest versions seem to have NOT changed in this respect. Read-
write locking was pretty much broken too (the last time I've looked 
at the sources). Well, but "performance" shines, that's true. ;-)

>                                         or whether NPTL is simply
> maintaining what you consider to be not-so-good semantics. you're
> entitled to your opinion, but a regression (against standard-mandated
> behavior) is probably worth filing a bug over.

I don't file bugs, sorry. Well, I've tried to "explain" (pointing 
out) one violation of standard-mandated behavior to Ulrich:

http://listman.redhat.com/archives/phil-list/2003-February/msg00114.html

He doesn't seem to get it (same problem as with Bill, I guess ;-) ).

> 
> finally: is it your belief that boost::thread will break under NPTL,
> in some not-currently-tested arrangement? that is the point of neal's
> question anyways, and the only thing I really meant to discuss :)

Only testing and, more importantly, NPTL and Boost.Threads *CODE 
REVIEW* can answer such questions "for sure". To put it simple, 
try it out but don't bet your last cent.

".2 EURO"

regards,
alexander.