On Fri, 24 Feb 2017 13:12:12 GMT, alex...@gmx.net wrote:
> This is a bug report for perl from alex...@gmx.net,> generated with the help of perlbug 1.40 running under perl 5.25.11.> > > -----------------------------------------------------------------> [Please describe your issue here]> > When analyzing system calls, I found that Perl socket always does> an additional fcntl(F_SETFD,FD_CLOEXEC). Modern socket(2) syscalls> support the SOCK_CLOEXEC flag to do this atomically. In general> this avoids a threading race, I did it to save the additional> syscall. I had to add logic for socket descriptor 0, 1, 2, but> this case should not happen in practice.> > I know that my patch is only a micro optimization, but it should> not do any harm.> > I have startet to implement this for socket(2). If you agree that> this is useful, I will also do it for similar operations. I have> started a new test file t/io/cloexec.t so I can add other cases> later.> > To get all tests pass, I had to adjust my email in AUTHORS>
Given code freeze, I'm marking this ticket and its patch for consideration post-5.26.0 release.
Thank you very much.
--
James E Keenan (jkee...@cpan.org)