Re: another subtle signals issue

[Roland McGrath]

> You just have to read it in a way that says "partial results are
> permissible, and are part of the normal behaviour". And then the fact that
> when ^Z happens you get partial results from pipes is not "different
> behaviour" from a qualitative standpoint - even though in fact we'd get a
> full result if the ^Z didn't happen.

I'm not talking about reading from pipes, that was your example.  I was
talking about calls with timeouts, like semop, whose interface do not
permit partial results.  Anyway, I find your reading insupportable even in
reference to read or write.  read and write are explicitly specified to
return partial results when interrupted by a signal, and are not permitted
to do so otherwise.  1003.1-2001 2.4.4 defines "interrupted" in reference
only to signals that are caught.