Devlog 23 Interpreter Pt3

December 24, 2022

1. Log 23
2. Terminal input buffer
3. Indexes pt2
4. Closing thoughts

Log 23

The plan today was to move the interpreter forward, but I discovered a design flaw in the terminal input buffer.

Terminal input buffer

In the last session, we were already processing the backspace character to “erase” characters from the TIB, but we haven’t even added characters to the TIB yet.

The idea was to add characters to the TIB as they arrive through the UART, to check them individually, and process them until a newline is seen, which would indicate the end of a command (unless in compile mode). That’s typically how Forth works:

1 1 + .<enter>2  ok

The approach used for reading a token (a word from the buffer) is to read backwards from the TOIN variable until a space character is found, and keep track of the length of the word. Then you move the address in TOIN backwards to where that word starts.

This works, but combined with the reference interpreter implementation (derzforth which copies sectorforth which assumes keyboard entry), we have a few problems:

1. UART characters sometimes get lost
2. The buffer won’t hold multiline definitions
3. The data stack is not even used!

I’ll explain each issue in the order listed above:

1. UART characters missing

You see, as our reference design suggests, we’re forced to process each character as it arrives over the UART. If there’s any kind of delay in processing, or if characters arrive too quickly, they will be lost. This happens when “pasting” or “uploading” code over a slow UART.

2. No multiline

For a multiline definition, you want to ignore the newline character and just keep going until you get the semicolon. The reference design checks each character as it arrives, and then uses the newline as a separator thus making it impossible to hold a multiline definition.

3. No data stack!

A typical Forth stores words on the data stack, and executes words based on what’s stored in there. The reference interpreter design does not even use the stack - at all.

I plan to fix these issues in a future session, once the v1 of this implementation is complete.

Indexes pt2

Upon reviewing my Indexes idea from devlog 22, I decided to also put this on hold until I’ve completed the v1 implementation of this Forth. I realize I might need to undo a lot of code in order to go back and implement indexing and alternative buffering, but for the moment I just want to get a working Forth haha.

Closing thoughts

This session was brief, again with no code, but I’ll get to that in the next session, where I plan to finish the interpreter and the lookup function.