Hi,

A couple of months ago I noticed strange behaviour from a warrior.
Moving the boot pointer caused it to drop 90 points vs the hill, but
there was no significant difference vs my benchmark. Both booted
versions scored worse than unbooted.

I think I've narrowed the issue down to "end label" not always setting
the correct start address. Here's an example:

;redcode-94nop
;name broken
;author John Metcalf
;assert 1

mov.i #1, 1
d dat 1, 1
end d

It compiles correctly on KOTH.org:

Program "broken" (length 2) by "John Metcalf"

ORG START
MOV.I # 1, $ 1
START DAT.F $ 1, $ 1

It should die on the first cycle. However, the score vs the hill is like
an imp - 4% wins, 45% ties and 51% losses. I tested pMARS 0.8.0, 0.8.5,
0.8.6, 0.9.2 and 0.9.4, and it scored as expected (like a dat).

Can anyone narrow this down any further? Is it a pMARS issue, or
something on KOTH?

Thanks,

John

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

This has been fixed now, thanks to Tuc. Koth had recently started to
lose the "end" statement under specific circumstances.

John

On Tue, 02 Aug 2022 18:23:54 +0000, John Metcalf wrote:

Hi,

A couple of months ago I noticed strange behaviour from a warrior.
Moving the boot pointer caused it to drop 90 points vs the hill, but
there was no significant difference vs my benchmark. Both booted
versions scored worse than unbooted.

I think I've narrowed the issue down to "end label" not always setting
the correct start address.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)