From: Brian Adkins <lojicdotcom@...>
Date: 2008-12-18T05:07:16+09:00
Subject: Re: why Hash corrupts 'key' object ?

Mike Gold <mike.gold.4433@gmail.com> writes:

> Brian Adkins wrote:
>> 
>> What the above demonstrates is that dup'ing and freezing a TestStr
>> does not nilify TestStr#dupstr, so the C code for Hash seems to be
>> doing more than just dup'ing and freezing. Does that make sense?
>>
>
> Yes, sorry, I misunderstood.  I did re-read your post several times but 
> was still a little puzzled.
>
> You are right that something appears to be amiss.  rb_str_new4() 
> duplicates the String portion of the TestStr object, but does not copy 
> over the TestStr-specific data.  A subclassed String is not duplicated; 
> the documentation is wrong.
>
> I see the reason for it: the raw string data is being shared between the 
> object and its "dup".  I guess sharing should only be done if it's a 
> real String and not a subclass thereof.  This fixes the problem reported 
> in this thread.
>
> --- a/ruby-1.8.7-p72/hash.c  2008-06-08 14:25:01.000000000 -0400
> +++ b/ruby-1.8.7-p72/hash.c  2008-12-13 13:47:33.000000000 -0500
> @@ -986,7 +986,8 @@
>      VALUE hash, key, val;
>  {
>      rb_hash_modify(hash);
> -    if (TYPE(key) != T_STRING || st_lookup(RHASH(hash)->tbl, key, 0)) {
> +    if (rb_obj_class(key) != rb_cString ||
> +        st_lookup(RHASH(hash)->tbl, key, 0)) {
>    st_insert(RHASH(hash)->tbl, key, val);
>      }
>      else {

Wow, complete with a patch! Have you passed this on to the appropriate
channel for possible inclusion ?

> -- 
> Posted via http://www.ruby-forum.com/.
>

-- 
Brian Adkins
http://www.lojic.com/
http://lojic.com/blog/