Re: INDIRECT and real-time question

[Pavel Shevchenko <pavel telemetry-investments com>]

Jody Goldberg writes:
> Ok, it looks like I may finally have cracked this nut.  The patch is
> in CVS to support dynamic dependencies for things like INDIRECT,
> OFFSET, INDEX and friends.  It seems to work and should have decent
> time/space performance.  Bug 5223 is at long last closed (the oldest
> non-enhancement bug).  I'll need to have a look to see what the next
> target is.
Thanks. I will try it soon.
> I'd be interested to hear your thoughts on a real time interface.
> Given the plethora of data feeds and the ugly parsing required for
> some of the page based services I'm quite interested in what you're
> aiming for.  Sending an overview to the list or coming by #gnumeric
> on irc.gnome.org would be a good start.
I think my case is not very interesting. I am writing real-time interface to 
our own system connected to the Reuter feed. Last three weeks I spent on
comparing gnumeric and OpenOffice spreadsheets and here is my random thoughts:

1. High performance real-time interfaces should be area-based, so real time call
should looks like RTS(<symbol_range>, <field_range>, "<out_cell_name>")
for example: RTS(A1:A101, B1:F1, "A2") to insert 100x5 = 500 real-time values
into cell range A2:F101. 

2. OpenOffice is much slower than gnumeric, by factor of 10. I measured SNP500
index (its 500 stocks and approximately 25 updates per second) and on 
PIII 1 GHz gnumeric takes 0.3-0.5 % CPU load, OpenOfficve takes 3-5%.
Main problem with OpenOffice is that provided way to write asynchonous 
function is cell-based, and there is no way to control spreadsheet from
inside your function (in OpenOffice your function is an object which 
can inform spreadsheet object what data are changed), so spreadsheet recalculates 
after updates in every single cell. If somebody is interested in simple
real-time interface to OpenOffice I can send my code, but I think it is
not useful for high performance spreadsheets.

Pavel.