discussion related to source for porting headers and methods to RTEMS and newlib

[Eshan Dhawan]

On Wed, Mar 25, 2020 at 4:18 AM Joel Sherrill <joel at rtems.org> wrote:

>
>
> On Tue, Mar 24, 2020 at 5:33 PM Gedare Bloom <gedare at rtems.org> wrote:
>
>> email subject can be shortened. almost everything is a discussion :)
>>
>> On Tue, Mar 24, 2020 at 3:49 PM Eshan Dhawan <eshandhawan51 at gmail.com>
>> wrote:
>> >
>> > Hello everyone,
>> > I have identified sources to port headers and methods to RTEMS and
>> Newlib. I have given priority to FreeBSD for the choice of source. But
>> method not present in FreeBSD can be ported from alternative sources like
>> NetBSD and musl.
>> >
>> Considering the simplicity, you may want to compare musl vs freebsd as
>> well.
>>
>> > -> Missing methods of math.h: I have compiled a list of methods and
>> their sources. Some need to be implemented from scratch.
>> > list: Math.h missing functions
>> >
>> > -> sys/ipc.h: This header and its function can be ported from FreeBSD.
>> > ipc.h : https://github.com/freebsd/freebsd/blob/master/sys/sys/ipc.h
>> > ftok.c:
>> https://github.com/freebsd/freebsd/blob/master/lib/libc/gen/ftok.c
>
>
> sys/ipc.h has a lot in it. If you head down the path of adding sys/ipc.h,
> you need to evaluate what else is in the file.
>
> Adding support for the System V style SHM, Semaphores, or Message Queues
> is possible (I think) but is definitely harder than the likely value to the
> community.
>
I would like to know missing headers and methods that would be more
valuable to the community

>
>> >
>> > ->fmtmsg.h: its implementation is also present in FreeBSD.
>> > https://github.com/freebsd/freebsd/blob/master/include/fmtmsg.h
>> > https://github.com/freebsd/freebsd/blob/master/lib/libc/gen/fmtmsg.c
>
>
> By my recollection, this would be a discrete thing to add to Newlib.
>
>>
>> >
>> > ->spawn.h: Its implementation is in FreeBSD.
>> >
>> https://github.com/freebsd/freebsd/blob/master/lib/libc/gen/posix_spawn.c
>> >
>> I'm worried about the large number of includes there. It may easily
>> lead you to trouble chasing down transitive include headers, like what
>> happened with Vaibhav before with search.h support.
>>
>
> posix_spawn() cannot be supported by RTEMS. It is a new safer way to do
> fork/exec. Don't spent any time on this. It will never work on RTEMS.
>
> It is also a HUGE amount of methods and constants. :(
>
Ok I will remove it from the proposal

>
>> > -> pselect() from <sys/select.h> : its implementation is also from
>> FreeBSD
>> > https://github.com/freebsd/freebsd/blob/master/lib/libc/sys/pselect.c
>
>
> This would be added to rtems-libbsd. There may be specific technical
> challenges
> why it is not present. Or it could just be an oversight.
>
I haven't dig that deep.

>
>
>>
>> >
>> > -> confstr() from <unistd.h>: Its implementation is from FreeBSD
>> > https://github.com/udp/freebsd-libc/blob/master/gen/confstr.c.
>>
>
> The implementation of this would go in RTEMS.
>
> It is possible that the FreeBSD code is a guide but this cannot be a direct
> copy to port. Each value defined by POSIX as fetchable must be properly
> defined for RTEMS. You would need to make a table of all the "names"
> that can be looked up in your proposal.
>
what would be better writing it from scratch of just doing minor tweaks in
the FreeBSD code
We also have an option to take code from musl
https://git.musl-libc.org/cgit/musl/tree/src/conf/confstr.c

>
> FWIW I think the FreeBSD version is more right than wrong for RTEMS
> but it can't be 100% right and needs analysis.
>
> >
>> > I would like everyone to review it and provide your suggestions.
>> >
>> include details in your proposal also.
>>
>
> +1
> I have already added these detatils to the proposal
> but will change spawn.h
>
>>
>> > Thanks
>> > -Eshan
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20200325/cab37496/attachment.html>