An update on my problem:  as Tom Cwikla pointed out on comp.arch.embedded,
it can't be an interrupt as when I query INTCON in my ISR it shows GIE is
set and when an interrupt is thrown GIE is cleared by the hardware.  So
there must be something else at work here.  I've already posted the bulk of
my source to the list but I missed this bit, and it appears to be the
culprit, even though I can't see anything wrong with it.  When I call this
routine I end up in the ISR.  It is a simple compound loop which is supposed
to take W as it's only input and delay for approximately W x 1 ms.

delayWms
 movwf delaycounter_ms
 movlw 0xa6
 movwf delaycounter2
 movlw 0x0a
 movwf delaycounter3
delaymsloop
 decfsz delaycounter2, F
 goto delaymsloop
 movlw 0xa6
 movwf delaycounter2
 decfsz delaycounter3, F
 goto delaymsloop
 movlw 0x0a
 movwf delaycounter3
 movlw 0xa6
 movwf delaycounter2
 decfsz delaycounter_ms, F
 goto delaymsloop
 return

I hope someone can point to a problem with this.... my code obviously fits
in 1 bank, though it may extend across multiple pages.  I've never had to
cross a bank boundary yet with my code (will have to soon)  but I've nearly
filled a bank with no ill effects, so if anyone has information about page
and bank boundaries I'd certainly be interested (gotchas that aren't obvious
from the literature, don't want to waste list members' time)

Thank you all,
        -Carl

--
http://www.piclist.com hint: The list server can filter out subtopics
(like ads or off topics) for you. See http://www.piclist.com/#topics