Re: Large amounts of text seem to cause PicForth to generate bad code - updated

Alex Holden wrote:
> I've made some progress in narrowing this bug down. The problem seems to 
> be that the yield word in multitasker.fs generates bad code once the 
> amount of program memory in use gets above 2K. I think it's probably 
> just missing out the bank selects but I really don't understand how 
> multitasker.fs works so I'm not sure how to fix it.

Seems that Samuel has made some significant changes to the PicForth 
compiler kernel without checking impacts on multitasker.fs and 
priotasker.fs.

You probably saw an earlier message, where I flagged the fact that 
programs using priotasker.fs won't even compile with latest PicForth.

Something you might like to try is compiling your prog with earlier 
versions of PicForth, to see if 'yield' generates different (and 
possibly more correct) code.

That would help Samuel (if/when he next shows up on this list) to narrow 
down the problem.

I'm more of a forth power-user rather than hacker. Once forth code gets 
really esoteric, I get badly lost.

One thing that has sometimes worked around compiler bugs is to try 
putting in a 'forth> no-cbank' before any words that don't compile 
correctly.

For instance, with your code:

task : alarm
   begin
     [char] a emit
     forth> no-cbank
     yield
   again
;

This causes the forth compiler to forget its codebank status, and often 
forces it to explicitly generate correct PCLATH set instructions.

-- 
Cheers
David