ECL? was ..Re: Runtime determination of share directories?

Paul Zimmermann wrote:
>    > Setting the rounding mode to 53 bits isn't actually enough.  I saw some
>    > Java numerics slide that explained this.  
>
>    I do recall some paper discussing the problem, but I do not have time to 
>    look into this. If you find the slides feel tree to send them to me off 
>    list.
>
> moreover it is not enough to set it at startup time, since any other component
> or even the user may revert the rounding mode. Thus this change has to be made
> at every call of a quaddouble function.
>
> Paul Zimmermann
>   
Looking back at the qd stuff I used, where I inserted a fix_fpu_start in 
each interface routine, I forget if I had any guarantee I had that some 
operating system interruption would not come in and change the rounding 
anyway.
(I think the lightweight threads in  the model used by Allegro CL on x86 
is not going to interrupt).

  Anyway, setting the rounding at startup time is, as you say, not 
enough, and if that is what Sage does, that might produce random behaviors.