Subject: Re: profile results for new UT_* implementations?
From: Martin Sevior (msevior@mccubbin.ph.unimelb.edu.au)
Date: Tue Jun 19 2001 - 23:00:38 CDT
On Tue, 19 Jun 2001, Paul Rohr wrote:
> Hi folks,
>
> In wading through the sources today looking at all those leaks, I realized
> how substantially the codebase had changed due to the pre-0.9.0
> (re-)implementations of:
>
> strings
These make writing code for Abi MUCH easier. No more static string arrays.
Can do:
props += "margin-left:";
props += "1.0in; ";
> hash tables (now renamed, right?)
We needed these so we could delete styles.
> ref-counted objects
>
> It's obviously not time to revisit those design decisions, but I was
> wondering if anyone could point me at the relevant design threads to help me
> get up to speed on the rationale behind the new implementations of such very
> fundamental data structures.
>
> In particular, I'd love to see the profile results which showed that we've
> gotten better time/space tradeoffs than we had with the older Jeff-style
> code I was familiar with. (Or are Martin's view-level rewrites responsible
> for all the speedups?)
The view level changes helped enormously but I also got substantial speed
wins from caching doc positions in the Frags and keeping Vector of
all Frags for binary searches. Abi is now usable for large
documents. On my 400 Mhz Pentium I can scroll at over 10 lines a second on
a 100 pages document. Cuts/pastes/texts insertions and deletions all
happen at basically emacs speed on this document too.
Cheers
Martin
This archive was generated by hypermail 2b25 : Tue Jun 19 2001 - 23:01:09 CDT