🚀 go-pugleaf

joe@modcomp.uucp (Joe Korty) writes:

>In <1993May3.171615.25882@spectrum.xerox.com> leisner@wrc.xerox.com (Marty Leisner x71348 ) writes:

>>This is necessary if each thread gets its own copy of errno...

>There are many ways to implement threads without making compiler
>changes.  For example, the errno problem can be handled by making errno
>a macro which expands to the code necessary to fetch and/or modify the
>thread-specific errno value.  The POSIX 1003.4 thread draft (or
>standard?, I haven't been following it lately) would be a good place to
>start looking for ideas on this subject.

This in itself is true. It is what SGI does in IRIX. But my question is:
is it flexible enough? The errno value is stored in some thread-local
storage. But suppose I wanted to write my own library with my own errno
variable? I could write the functions to get the thread specific values,
but how do I create thread local storage for each and every thread?

Special casing errno is one step, but I'd like to hear from someone
who has a solution that:
	- has user extensible thread local storage
	- doesn't require compiler and/or linker support

One reason for extensible thread local storage is MT safe
libraries. Many unix library calls use some sort of local,
static, storage for their return value.  If we want to make
those calls MT safe, we must be able to allocate some
storage, per thread.

Casper