z80 survey
The previous sprint marked the end of epic 4 (“let’s add an assembler!”), so I held a retro with the LLM dev team to get a few things off my chest before we move on:
- the
LD,register ordering issue was annoying and could have been avoided by reading the spec - the tag encoding issue was the result of horrible organic design, and could have been avoided
by considering the bigger picture in upfront architecture design. The “what happens if you
forget the#” footgun was unacceptable. - the huge tranch of absent z80 instructions from the extended set was a worry. this should have been noticed sooner, and now makes me wonder what our coverage is.
- very late in the day I discovered that we don’t support comments! So now I’m also wondering what our ANS Forth Core compliance looks like…
Happily, the next epic is a short one. It held just a single story, to implement the standard
word MARKER (a sort of “undo” for the dictionary).
I took the opportunity to shoe-horn in some extra stories:
- do a complete z80 opcode survey (just a report, no code) so I can understand the gaps
- implement comment words
- implement
MARKER - do a complete ANS Forth Core survey, to see how compliant we are (again, report only)
So, not much development and a lot of donkey-work: perfect for LLMs.
The team took their licks from the retro, and the scrum master wrote up the revised epic and created the first story 5.0 immediately. There’s no real development, so it doesn’t make sense for the dev agent to do the work, so I just asked the scrum master (who was already loaded anyway) to do it with “please complete the survey task in 5.0”.
The results from the survey are disappointing: 27 instructions not covered at all,
including a lot of fan favourites like NOP, DI, EI, LD A, (BC). RST. DJNZ !!

How am I meant to live without DJNZ ?! Fortunately scrum master Bob senses my mood
and offers to add a new story for immediate intervention. I instruct him to proceed with
haste.
You can view the survey on github