Faster Virtual Method Dispatch

[James Dunne]

James Dunne wrote:
> Walter Bright wrote:
> 
>> James Dunne wrote:
>>
>>> Can you provide me a few legitimate examples of when you won't know 
>>> what to free and when to free it?  I'm having a tough time coming up 
>>> with examples on my own...
>>
>>
>>
>> One example that has plagued me in the past is having two distinct 
>> tree data structures, and it becomes necessary for a node in one tree 
>> to refer to data in the other.
>>
>> Or when you have a third data structure, and you need to point to data 
>> in one or the other or both of the first two trees.
>>
>> A typical example of that is symbol tables inside the compiler. In the 
>> DMD front end (which is garbage collected) the data has all kinds of 
>> cross-connections in it.
>>
>> I.e. the problem crops up when the data in memory just doesn't fit 
>> into a strictly heirarcical format.
> 
> 
> Your example is having two trees _point_ to data within each other.  The 
> original data is still _owned_ by the original tree which the data 
> resides in.  You wouldn't be swapping ownership of the data like you're 
> talking about.  In this system, when memory is allocated it is assigned 
> an owner and that ownership does not ever change.  So the system 
> resembles a tree (one parent only), not a directed graph.
> 
> Funny you mention compiler, because that is the example application I 
> was referring to. :)
> 

I should also mention that my compiler implementation makes use of 
context structures, so every memory block that the compiler will 
allocate is somehow parented to that context structure.  So when the 
compile job is complete, free the context structure and all the memory 
goes away.

-- 
-----BEGIN GEEK CODE BLOCK-----
Version: 3.1
GCS/MU/S d-pu s:+ a-->? C++++$ UL+++ P--- L+++ !E W-- N++ o? K? w--- O 
M--@ V? PS PE Y+ PGP- t+ 5 X+ !R tv-->!tv b- DI++(+) D++ G e++>e 
h>--->++ r+++ y+++
------END GEEK CODE BLOCK------

James Dunne