Re: [PATCH] SIGPIPE redux -- please review
Hrvoje Niksic writes:
SL Baur <steve(a)xemacs.org> writes:
> Kyle Jones <kyle_jones(a)wonderworks.com> writes in xemacs-beta(a)xemacs.org:
>
> > I think the patch introduces a descriptor leak.
>
> Not according to my tests.
I think you are right.
Kyle, a question: if write() is failing, doesn't it mean that the
descriptor is practically dead? Does it really need explicit
close()-ing?
Yes. Writing to a closed descriptor produces a return value of
-1 with errno set to EBADF on POSIX systems. SIGPIPE is signaled
only when writing to an open pipe/socket descriptor whose reader
has closed the other end of the pipe (perhaps by exiting).
[...]
Also, I *would* like close() getting called on that fd eventually, no
matter what.
Me too. It must be getting called somehow, if Steve isn't
seeing descriptor leakage. But I certainly don't set how that
is happening, given that the closer method isn't called in
Lstream_close when the LSTREAM_FL_IS_OPEN flag bit is 0.