Re: threads in 1.7.1

[Mikael Djurfeldt]

Marius Vollmer <address@hidden> writes:

> Neil Jerram <address@hidden> writes:
>
>>    (I'm not sure of the rationale for dropping coop threads;
>> personally I'm a fan of them.)
>
> I, too, am a fan of coop threads.  In Guile, however, there was little
> difference between coop threads and preemptive threads for a Scheme
> program: even the coop threads could get preempted at any time in a
> Scheme program.  The advantage of coop threads was on the C level,
> where they made thread-safeness easier.  I'm quite sure now that we
> can handle the C level for real concurrent threads, so the advantages
> of coop threads mostly disappear.
>
> The switch to POSIX threads was mostly motivated by compatability, and
> to avoid maintaining a thread implementation of our own.  Guile should
> be useable together with other code that used POSIX threads.  If you
> want to have coop threads, you would need to use a POSIX thread
> implementation that is coop.

I think it also has made Guile "future-proof", considering the route
hardware development likely will take now that we're approaching some
hard speed limits for the classic PC architecture.  I'm thinking of
developments such as multicore CPU chips, which, for example, Intel
now seems to target for the personal PC market already at the end of
next year.

Also, please look at workbook/threads/thread-interface.text where some
design choices are discussed, and where it has been suggested the
possibility to provide COOP threads as an application link time
option.

M