Subj : Re: RCU+SMR
To   : comp.programming.threads
From : Joe Seigh
Date : Fri Jul 01 2005 07:53 pm

Randy Howard wrote:
> On Thu, 30 Jun 2005 20:06:52 -0500, Joe Seigh wrote
> (in article <3cqdnbE9cbI2C1nfRVn-ig@comcast.com>):
> 
>>And as you can see by this topic practically being a monologue,
>>there's no huge demand for this kind of stuff, so I'm taking
>>advantage of the more leisurely schedule that allows.
> 
> 
> I wonder if lack of portability is partly to blame?
> 
> What would be the minimal number of extensions to C (or 
> whatever) that would allow this to be implemented portably by a 
> compiler supporting, for example 'C99+atomic'?

That topic pops up from time to time.  It doesn't go anywhere
due to lack of interest.   Very few people use that kind of stuff.
There are some api's out there like atomic_ops here
http://www.hpl.hp.com/research/linux/qprof/
but it's fairly limited.  If you have to add your own
you might as well do the whole api on your own as it's not
that much more work.  If I used atomic_ops, I'd still have
to provide my own atomic ops package.

> 
> If you can quantify that, you might actually be able to make a 
> case for language extensions so that lock-free algorithms can be 
> used in cross-platform production projects.

It's mostly a chicken and egg question.  Present usage doesn't
justify putting in the features.

> 
> I would be very interested in anything along those lines, but as 
> long as it is either OS-centric, CPU-centric, or both, I don't 
> see much value in it for the type of software I am involved 
> with.
> 

You have to start somewhere and I can't exactly port to hardware
I don't have.  So probably ia32-64 on Linux for the time being.
Apple is getting off of ppc and I'm not going to buy propietary
hardware just for the privelege of porting to it.  Ditto with
Sun.  Also, their list of Solaris qualified hardware is unusable.
You'd likely end up getting hardware you really didn't want,
only to find it doesn't work with Solaris.  I can't afford
to do that kind of experimentation.

You get what you pay for.  Free software gets little or no
support.

The other thing here is api's are a really hard sell.  I did
a low level api once, a program that translated boolean
queries into decision trees.  Almost no interest apart
from my own use.  Tacked it onto a program to scan files,
a sort of flat file database and it became hugely popular.
More than 10 years after I stopped supporting it and after I
left IBM, someone tracked me down to get it enhanced (couldn't,
I didn't have the source anymore.)  So that's why I'm looking
for a moderate sized project to do instead.


-- 
Joe Seigh

When you get lemons, you make lemonade.
When you get hardware, you make software.

.