From: Simon Strandgaard <0bz63fz3m1qt3001@...>
Date: 2003-06-27T18:04:41+09:00
Subject: Re: collect info about ruby-api

On Fri, 27 Jun 2003 18:21:28 +0900, ts wrote:

>>>>>> "S" == Simon Strandgaard <0bz63fz3m1qt3001@sneakemail.com> writes:
> 
> S> I don't know what plruby_call_handler() does.. but it definitly uses
> S> Init_stack()  
> 
>  plruby_call_handler() is the entry point for plruby : each time that
>  postgres need to call a procedural function written in ruby it will call
>  plruby_call_handler() 
>
> S> int pl_call_level;   hmmm, don't know what this is either.
> 
>  a procedural function written in ruby can call another procedural function
>  written in ruby : pl_call_level will say if it's the first call or if it's
>  a "recursive" call.

If its a nested request, then Init_stack will only be invoked once (first
request). If its not a nested request, then Init_stack is executed. OK?

Browsing mod_ruby, I see that Init_stack is invoked within their
library-bootstrap function (ruby_startup). It is not invoked per request.

Why this difference ?

BTW: In VIM/if_ruby.c Init_stack is not used at all!

 
> S> Any hints ? :-)
> 
>  Normally ruby_init() (which call Init_stack()) must be called at the
>  beginning of the program (postgres in this case) to make sure that ruby
>  has the right value for the beginning of the stack.
> 
>  It's not possible (and I don't want) to modify postgres to be sure that
>  ruby_init() will be called at the right place (in main()). The
>  consequence is that plruby will call ruby_init() (to initialize ruby) the
>  first time that a procedural function written in ruby is called.
> 
>  Now where is the problem : imagine this case (the values for the stack are
>  just examples) 
> 
>  * The first time that postgres call plruby_call_handler() the value of the
>  stack is 0xe0000000
> 
>   - ruby_init() is called, and ruby will remember that the beginning of
>     the stack is 0xe0000000
> 
>   - ruby will store VALUE from 0xe0000000 to 0xc0000000 (this is just an
>     example) 
> 
>  * Another procedural function is called latter (and this is not a
>    "recursive" call) but this time the value for the stack is 0xf0000000
>    (i.e. postgres use less stack when it call plruby_call_handler())
> 
>   - ruby will store VALUE from 0xf0000000 to 0xd0000000
> 
>   - if ruby call the GC, the GC will mark the objects from 0xe0000000
>     (beginning of the stack when ruby_init() was called) to 0xd0000000.
>     When ruby run the sweep phase, it will remove all objects which are
>     stored in 0xf0000000-0xe0000000
> 
>   - now after the GC, if plruby try to access objects stored in
>     0xf0000000-0xe0000000, it just crash.

Yes I see, the objects withing the range 0xf000-0xe000 is un-marked and
will thus be sweept. I agree that trying to access destroyed instances can
cause segfaults.

 
>  * To correct this : each time that plruby_call_handler() is called (and
>    this is not a "recursive" call) it call Init_stack() to be sure that
>    ruby has the right value for the beginning of the stack when it run the
>    GC 

Solution to the second request, am I right ???
Init_stack(0xf0000000);  


--
Simon Strandgaard