[comp.sys.amiga] Lattice C 3.10 bug

pej@cuuxb.UUCP (01/28/87)

I have been having problems with Lattice 3.10 on "large" programs. Last night
there was a note on the Lattice BBS saying that the compiler will generate 
bad code if functions declare more than 3 register pointer variables. Aparently
there are only 3 address register variables available for assignment and
the compiler screws up if more are declared.

MicroEmacs 3.7 is *real* flakey compiled under 3.10 (worked fine under 3.03).
About 14 functions declared more than 3 address pointers. After fixing
the functions it's working fine.

Hope this helps someone else out also.....

					Paul Jatkowski
					{ihnp4!}cuuxb!pej
			

papa@bacall.UUCP (01/29/87)

> I have been having problems with Lattice 3.10 on "large" programs. Last night
> there was a note on the Lattice BBS saying that the compiler will generate 
> bad code if functions declare more than 3 register pointer variables. Aparently
> there are only 3 address register variables available for assignment and
> the compiler screws up if more are declared.
> 
> MicroEmacs 3.7 is *real* flakey compiled under 3.10 (worked fine under 3.03).
> About 14 functions declared more than 3 address pointers. After fixing
> the functions it's working fine.

Yes, it is true of all versions 3.02, 3.03 and also the new 3.10.  I just
stopped using register variables with Lattice C.  Now you know that not
only MANX can generated BAD code.  Also there is no word about BLINK fixes
for the overlays.  The software distillery people seams to be deaf to inquiries
about the BLINK status.  Has anybody downloaded the latest BLINK from their
BBS and tried it with overlays?

-- Marco Papa
   Felsina Software