Re: [PATCH] Provide posix_spawn implementation

[Petr Hosek <phosek at chromium dot org>]

Thanks a lot for the comments! Attached is the updated patch; let me
know if you spot any other issues.

Petr

On 27 September 2013 21:53, Jeff Johnston <jjohnstn@redhat.com> wrote:
> On 09/16/2013 04:07 PM, Petr Hosek wrote:
>>
>> I have ported over posix_spawn implementation from FreeBSD to newlib.
>> It implements posix_spawn interface in terms of existing POSIX
>> interfaces, in particular vfork and exec.
>>
>> This should make it easier to port existing code which uses
>> posix_spawn, which is already supported by glibc as well as all BSD
>> variants. In the longer term, posix_spawn could be also interesting to
>> Cygwin as the interface is more akin to CreateProcess provided by
>> Windows so the applications using posix_spawn instead of fork/vfork
>> would get performance improvement compared to the current fork
>> emulation.
>>
>
> Some comments.
>
> 1. posix_spawn should be listed as an EL/IX 4 source since it is not
>    specified in the EL/IX standard (as opposed to a general source in
>    Makefile.am)
>
> 2. some of the posix routines are protected by #ifndef _NO_xxxxx
>    This is needed for this code (#ifndef _NO_POSIX_SPAWN)
>
> 3. execvpe is a gnu extension starting in glibc
>    2.11.  It appears to be only supported by
>    Cygwin in newlib.  I can't speak for RTEMS and powerpcle which are
>    users of the posix dir, but you will have to set _NO_POSIX_SPAWN for
>    x86-linux which don't supply this unless you can replace
>    the code or shield it with a Cygwin check.  The changes go in
>    configure.host.  You should turn it off for RTEMS and powerpcle
>    until they confirm one way or another.
>
> 4. Other posix code there such as popen and opendir adds a check for
>    HAVE_FCNTL before using it.
>
> 5. Please add documentation similar to what is found in popen.c
>
> -- Jeff J.

Attachment: posix_spawn.patch
Description: Binary data