Re: Read/write integers?

On Sun, Apr 27, 2008 at 10:59 PM, David King <[email protected]> wrote:
[..]
>  At the moment, I've limited each entry to one line of text, delimited by
> newlines ("window <bychunk> /\n/ /\n/"). That makes it relatively easy to
> communicate with crm114 just by "print line into pipe, read line". (And it
> works very well!) However, given the line-length limit imposed by crm114 (on
> my 32-bit machine, crm114 will crash if I feed it lines longer than 4k, and
> on my 64-bit machine, it crashes on lines longer than 1k; probably
> libreadline or somesuch's fault), that means that for longer entries, I can
> only classify based on the first ~1k of the message, which isn't ideal, and
> keeps me from adding some of the more interesting tokens from our database,
> as they could be quite long.

Let me digest the rest of what you're saying there, but one thing is important:

crm114 should *NOT* crash. Never. _Ever_.

So what you're running into is a bug; besides, the minimum buffer for
window <bychunk> should be somewhere around 8 MEGAbyte, unless you
changed the 'data windows size' using a commandline switch. ('-w'
switch)

The next border that you may hit somewhere is MAX_PATTERN, which
currently stands at 16KByte.


>  Since it's not possible for crm114 to parse or generate the length-header

On output, length of a string can be produced using the :#:var: 'eval'
length operation, but this is just off the top of my head and without
checking what you _really_ need yet.


>  Let me know. I'd use it :) If one could pull the ordinal value of a
> character, it would be easy to multiply to combine several and get a
> representation of the integer value (taking endianness into account as
> appropriate)

I have something almost working already - at least for the latest
private GerH build - and the concept is really simple. I'll post the
(ahem) proof of concept later tonight. No endianess 'handling' in
there though. :-|




-- 
Met vriendelijke groeten / Best regards,

Ger Hobbelt

--------------------------------------------------
web: http://www.hobbelt.com/
 http://www.hebbut.net/
mail: [email protected]
mobile: +31-6-11 120 978
--------------------------------------------------

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone