From: James Edward Gray II <james@...>
Date: 2006-02-04T06:47:06+09:00
Subject: Re: Question about massive API changes

On Jan 31, 2006, at 9:06 PM, Sean E. Russell wrote:

> On Tuesday 31 January 2006 14:44, mathew wrote:
>>> What do you mean by an API facade?
>>
>> Something that accepts calls via the old API, and translates them  
>> into
>> appropriate calls into the new API.
>
> I thought that's what you meant.  Unfortunately, that's not the API  
> I'm
> talking about.
>
> In REXML, currently, you can do this:
>
> 	el.attributes['someatt'] = "foo"
> 	el.attributes['someatt'] << "bar"
> 	el.attributes['someatt'] == "foobar"    # true
>
> I can do this, because the attributes are being stored as Attribute  
> objects,
> even though you get back String objects from attributes[].  To get  
> the space
> saving, I'm only generating Attribute objects when the user makes  
> an API call
> that returns an Attribute object; otherwise, I store the original  
> String
> objects, which are less than 1/3 the size of an Attribute object.

What about using lazy evaluation for the old style methods?  The  
documentation can warn that they are no longer the way, possibly even  
depreciated, and that memory usage increases dramatically with their  
usage.  Then *if* one of them is called, the needed structures spring  
into existence and the old API is functional.

> Depending on the XML document, this can save a significant amount  
> of memory.

Have you played around at all with storing just offsets into the  
document?  I have no idea if this is remotely practical.  It's just a  
random thought I had.  A tag could be four Fixnums perhaps:  an open  
tag start offset, an open tag close offset, a close tag start offset,  
and a close tag stop offset.

How much time does it take to parse one tag knowing all those  
offsets?  The worry is that usage could force you to parse the same  
tag repeatedly, I guess, but you could potentially get away with  
never parsing some tags.  You could cache the Strings after you  
parse, for later usage.  You might even be able to reasonably leave  
the document on disk this way (using file seeking).

Again, not trying to tell you how to write your library here.  Just  
thinking out loud...

James Edward Gray II