Re: [MIT-Scheme-devel] pclsr

[Joe Marshall]

On Mon, Mar 14, 2011 at 3:15 PM, Taylor R Campbell <address@hidden> wrote:
>
> Any ideas?

Get yourself a *real* OS?

Is this a process-level problem, or thread level.  In other words, could you
run the open on a separate thread and just have it block?  Or do signals
to *any* thread in the process cause all pending syscalls to restart?

How about a truly disgusting idea:  when calling open to a remote file,
start a local subprocess that makes the open call and open a pipe to it.
The local subprocess should not need to handle timer interrupts so it
should eventually be able to open the file.  This adds a lot of overhead
in a spinning up an extra process for each remote file, but high-overhead
working code is preferable to low-overhead broken code.

Shitcanning Unix is the obvious correct idea, though.


-- 
~jrm