[openbeosnetteam] Re: libbind.so status

> > libbind.so compiles fine and there are not linker errors when I run 
> > Mozilla.
> > But it does not resolve the hostname. :(
> 
> A bad point when resolving hostname is what libbind.so is expected to 
> do at most :-)

:)

> > All functions are called, but select() always times out.
> 
> Maybe an issue with our select() or BONE select() implementation?

It is fixed now. But our select() emulation crashes when used with Vision (uses 
libnet). Vision starts three threads which call select() simultaneously. Then, 
one dies.

> > Shall I commit the source so you can have a look at it, too? It would 
> > be a 1.2MB commit, 
> 
> Damn, too late, I just saw the CVS notifications, I should have react 
> earlier.
> My reaction would have been: "What? 1.2MB!?!"

Yes, I thought this could get it smaller, but the resolv stuff seems to depend 
on many things. I will look if I can somehow drop many things out again, but 
last time it did not work...

> You've port the whole BIND library I guess, but we really only need the 
> client part.
> Why not only port/update the client part we do have ported already in 
> src/kits/network/libnet/* ?

The README file in BIND says: the BIND resolver library can be found in 
lib/bind, so I compiled everything from there.
Also, BONE's libbind.so is about 300k which is the same as ours (our old BIND 
is a tenth of this size).

> I'm worried you put ISC headers in public/posix tree instead of private
> /net/isc or even under src/kits/network/libbind/headers where it have 
> make more sense. Most if not all ISC headers are in no way POSIX ones 
> but, well, just BIND headers. Sound kind of private to libbind to me...
> 
> For POSIX standard headers list, check here:
> http://www.opengroup.org/onlinepubs/007904975/toc.htm

They say netdb.h should only contain gethostbyname(), not its gethostbyname_r() 
version, but Mozilla wants that one to be defined, too. Actually, you may be 
toally right about the other headers that I moved into our public folder.
Nobody will use them anyway. I only looked at what BONE exports and saw all 
these headers there, so I thought for full BONE compatibility we need all that, 
too.

> Oh, BTW, you're right, we need pthread support for better POSIX 
> compliance, even if I always thought being 100% POSIX 1003.1 compliant 
> for R1 is not our goal. (I never know pthread.h was made an official 
> POSIX API!)

Yes, and we should really support it to get other apps ported (MONO). I think 
on the MONO mailinglist someone said that BeOS not having full pthread support 
is one reason for not supporting it, but I do not remember when and where it 
was...

> I may not being best placed to debate about code as I recently dropped 
> out of OBOS top 10 contributors list ;-), but, well, here my point:

This is why I did it. I wanted to become #1! :))

> To put shortly, I concerned about the UNIXification of OBOS code.
> BeOS was not an UNIX, and I always like the cleaness feeling you get 
> browsing develop/headers/*.
> I think it's important for the project to keep control on our public 
> headers to keep this BeOS *feature* that way.
> 
> The net kit is more exposed to such issues due to de-facto UNIX origin 
> of most INET network stuffs...
> 
> So, let's toss blindly all these headers under private/* subtree!
> Just kidding :-p
> But I think we should decide some policy on that topic.
> 
> What's your opinion, guys?

My main aim was to get something finished. You are right that we are slowly 
becoming a mess. Although, I always had the feeling our netstack is just a 
mess, too, and that we should just get things done and when it is finished we 
can clean everything up.

Bye,
Waldemar

Other related posts: