From: Paul Brannan <pbrannan@...>
Date: 2008-10-22T01:58:23+09:00
Subject: [ruby-core:19418] Re: rb_errinfo() vs rb_rubylevel_errinfo()

On Wed, Oct 22, 2008 at 12:34:19AM +0900, SASADA Koichi wrote:
> def m
>   # (3)
> end
> 
> begin
>   raise # (2)
> rescue => e
>   # (1)
>   m()
> end
> 
> On (1), we can access $! with variable "e".  $! and "e" is same
> place on VM stack (not cfp, but VM stack witch is accessed by dfp).
>  This is means that $! is ruby level "local variable".
> rb_rubylevel_errinfo() returns that contents.

Is this just an optimization that e and $! are the same place on the
stack?  Assigning to e does not change $!.

> On (2), there is no local variable to store error information
> throwed by "raise" method.  To propagate errinfo,
> rb_thread_t#errinfo is used.  "raise" (rb_raise) stores errinfo into
> rb_thread_t structure of current method, and bring it to VM
> exception procedure.  VM receives errinfo and store it into local
> variable which I described above if corresponding "rescue" clause
> (or ensure) is there.
> 
> On (3), $! is also returns contents of "e" variable in (1).  VM
> backwinds cfp and looking for exception handling frame.  If no
> exception handling are there, only "nil" was returned.
> 
> Is this answer your question?

Yes, it does, thank you.

Two more:

1. Why do you choose to use a local variable to store errinfo in YARV?
   It was thread-local (global variable that is swapped when the thread
   is swapped) in 1.8.

2. On 1.8 it was possible to assign to $!, but this has been disabled on
   YARV.  Is there a particular reason?

Paul