From: Charles Mills <cmills@...>
Date: 2005-05-19T07:30:13+09:00
Subject: Re: howto make something like Errno::##, for my C ext, or do I need to?


Sam Roberts wrote:
> Thanks for your help, Charlie.
>
> Charlie Mills wrote:
> > In short:
> > rb_exc_raise(my_err_new(eno));
>
> Took me a minute to find that... it's in intern.h, can I call those
> functions, I mean, should I?
>

It is OK to call those... I asked about this once:
http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-talk/90163
http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-talk/1807

> Looking through that code, it seems my only option is to create the
> object myself, then call this internal function.
>
> Looks like it will work fine, but is this really recommended?

It is fine.  You also need to create your error class:
class MyErr < StandardError;  end # or in C:
eMyError = rb_define_class("MyError", rb_eStandardError);

Then you can define an accessor in Ruby or in C:
attr_reader :eno
rb_define_attr(eMyError, "eno", 1, 0);

I wouldn't worry about writing an init function if your only going to
be raising these errors from C.

You can instantiate/raise the errors as I described in the earlier
post.

-Charlie

>
> Daniel wrote:
> > Doesn't rb_sys_fail(0) work for this scenario?  I thought if a C
> > function failed, you could call rb_sys_fail(0) and it would raise
the
> > appropriate Errno.  Or am I confused?
>
> I didn't know about this function, but no, its not for me.
>
> It only makes sense if the C function you are calling is returning a
> system error number, i.e. from <errno.h>.
>
> For example, when a C function in our library returns 9, it means
"index
> out of bounds", but when a system call returns 9 it means bad file
> descriptor, and when the Crypoki API (random example) returns 9 it
means
> CKR_NEED_TO_CREATE_THREADS.
>
> I wouldn't want to return Errno::EBADF from my wrapper when the error
is
> index out of bounds!
>
> Charlie wrote:
> > Yeah that works too, if your using the standard library functions.
I
> > thought from the OP that the error numbers he was working with were
not
> > related to those used by the std lib, but I may be wrong.
> 
> Nope, bang on.
> 
> Sam
> 
> -- 
> http://www.certicom.com