From: Daniel Berger <djberg96@...>
Date: 2006-06-13T23:19:51+09:00
Subject: Re: Getting a list of aliases

Ryan Davis wrote:
> 
> On Jun 11, 2006, at 7:17 AM, Daniel Berger wrote:
> 
>> Is there a way to get a list of instance method aliases somehow?
>>
>> # Hypothetical syntax
>> Array.instance_method_aliases # ['size', 'map', ...]
>>
>> If not, I think it would be a nice little bit of extra inspection I 
>> could use in certain situations, like when I'm in irb and feeling too 
>> lazy to look at the documentation. ;)
>>
>> Actually, this idea struck me when I tried to figure out just the 
>> "core" methods for a given class and wanted to exclude the aliases.  
>> At the time, I had to view the source code directly and count them.  
>> This proved more difficult than I thought, because sometimes methods 
>> would redefine functions with different names rather than use 
>> rb_define_alias() in the source code.
>>
>> To give you an example of what I mean, in array.c the Array#map 
>> function is not declared using rb_define_alias(), but is simply mapped 
>> (no pun intended) to a different function:
>>
>> rb_define_method(rb_cArray, "collect", rb_ary_collect, 0);
>> rb_define_method(rb_cArray, "map", rb_ary_collect, 0);
> 
> WARNING: this gets gross:
> 
> It is a little less relevant with the core classes (seeing how they 
> don't change very much), but easier to figure out what is aliased. You 
> can use ParseTree to get the sexps for the class Array, and then it 
> becomes pretty "obvious":
> 
> ruby -I../RubyInline/dev:lib -rparse_tree -e 'class Array; alias :blah 
> :map; end; p ParseTree.new.parse_tree(Array)'
> 
> [[:class, :Array, :Object, ...
>   [:defn, :blah, [:fbody, [:cfunc, 504080, 0]]], ...
>   [:defn, :collect, [:cfunc, 504080, 0]],
>   [:defn, :collect!, [:cfunc, 504206, 0]], ...
>   [:defn, :map, [:cfunc, 504080, 0]],
>   [:defn, :map!, [:cfunc, 504206, 0]], ...
> ]]
> 
> So, with c-implemented functions, your :defn is pretty much just a 
> pointer to the C-function. You can compare the addresses and see what's 
> been aliased. Likewise, with a real alias to a :cfunc you can dig a bit 
> deeper and see that it points to it properly.
> 
> In the case of aliases on ruby methods, it gets harder:
> 
> % ruby -I../RubyInline/dev:lib -rparse_tree -e 'class X; def y; end; 
> alias :x :y; end; p ParseTree.new.parse_tree(X)'
> 
> [[:class, :X, :Object,
>   [:defn, :x, [:fbody, [:scope, [:block, [:args], [:nil]]]]],
>   [:defn, :y,          [:scope, [:block, [:args], [:nil]]]]]]
> 
> So, while you can conclude that :fbody is a marker for an alias to 
> SOMETHING, you can't tell what because it is already expanded. You 
> should be able to either compare sexps or use some C to dig in and see 
> that the implementations (nd_body pointer) have the same address. The 
> code in question is:
> 
>>     st_insert(RCLASS(klass)->m_tbl, name,
>>       (st_data_t)NEW_METHOD(NEW_FBODY(body, def, origin), 
>> orig->nd_noex));
> 
> so the addresses should be the same. (yes, either option is rather icky).
> 
> 

Very interesting Ryan, thank you.

What do you think about adding an Object.aliases method, though?  Or is 
it just not worth the memory/speed hit?

Dan