hp-ux vs linux - interim report

[David Relson]

Allyn,

I have bogofilter up and running at testdrive.hp-ux.com on the hp-ux risc 
machine.  I've run the test framework and the set of 10 all passed, while 
the singleton test failed, with a "set: Syntax error" message.  As the 
hp-ux machine is running csh, that's not too surprising.  I'll worry about 
it later.

I then ran the test.1127.sh script I sent you - the one that runs the 8 
tests for which I didn't received detailed output from you.  I transferred 
the outputs to my linux box and ran diff.  Lo and behold, there were 
differences - see below.  The Graham method uses a 15 element array for 
storing the most interesting tokens, i.e. those whose spamicities are 
furthest from 0.5.  On my linux box, I get 3 at 0.01 and 12 at 0.99 (all of 
which are 0.49 from 0.50).  The hp-ux box gives 15 tokens at 0.99.  This 
looks like a rounding problem, on which I shall check.

linux:	msg.2b.g.vvv
X-Bogosity: Yes, tests=bogofilter, spamicity=1.000000, version=0.9.0.3
	0.010000  0.010000  call
	0.010000  0.000102  working
	0.010000  0.000001  yet
	0.990000  0.000102  abuse
	0.990000  0.010000  appear
	0.990000  0.500001  bordercolor
	0.990000  0.990000  car
	0.990000  0.999898  cars
	0.990000  0.999999  colspan
	0.990000  1.000000  copy
	0.990000  1.000000  ff0000
	0.990000  1.000000  million
	0.990000  1.000000  secrets
	0.990000  1.000000  table
	0.990000  1.000000  valign

hp-ux:	msg.2b.g.vvv
X-Bogosity: Yes, tests=bogofilter, spamicity=1.000000, version=0.9.0.3
	0.990000  0.990000  --06986e0e1e196312e032afbc
	0.990000  0.999898  abuse
	0.990000  0.999999  bordercolor
	0.990000  1.000000  car
	0.990000  1.000000  cars
	0.990000  1.000000  click.asp
	0.990000  1.000000  colspan
	0.990000  1.000000  lnk
	0.990000  1.000000  mailings
	0.990000  1.000000  sat
	0.990000  1.000000  secrets
	0.990000  1.000000  table
	0.990000  1.000000  unsubscribe.asp
	0.990000  1.000000  valign
	0.990000  1.000000  yours