[boost] Re: Thread-safe containers

From: Tony Juricic (tonygeek_at_[hidden])
Date: 2004-09-05 12:19:20

"Holger Grund" <yahoo_at_[hidden]> wrote

> It's probably much harder for certain other architectures.

You are right. The rationale for providing API for singly-linked list is, I
believe, based on typical library/reusability/avoid-boilerplate code mantra.

There is a body of work for implementing Intel-like CAS (conditional store
and swap) via LL/SC (load linked/store conditional) instructions on other
architectures (notably Motorola).

IMO, the issue boils down to usefulness of adding such functionality to the
part of the library that deals with concurrency. Personally, I find it very
useful, even if I would never claim that all concurrent data-access issues
boil down to thread-safe access to standard containers. In my view, it is
library functions like thread.join(0 that are trivial enough so that anybody
can keep reimplementing them in his platform-specific programs, while
thread-safe containers are nontrivial and error-prone enough to deserve
resuable library implementation.

Tony