The MBASIC 5.21 sources - where they are, and when to reach for them

Created: 2026-08-05 Location: /home/wohl/src/mbasic2025 - https://github.com/avwohl/mbasic2025

The original 5.21 sources were lost. That repository reconstructs them: the nearest surviving 5.2 sources, adjusted until they assemble byte-for-byte identical to the mbasic.com we test against. Changes made to get there are marked ;5.21 in the code.

mbasic_521/mbasic_src/   the reconstructed 5.21 sources (.mac, 8080 assembler)
mbasic_521/com/          the reference mbasic.com
mbasic_521/disasm/       a disassembly of it
mbasic_52/               the original 5.2 sources it started from
mbasicz/                 a later Z80-optimised version
4k8k/                    4K and 8K BASIC, with the Altair manual

When to use them

Whenever the question is "what does MBASIC actually do here". Reading the routine is faster and more certain than disassembling the binary or inferring behaviour from outputs - and a comment in the source often says why, which no amount of measurement recovers.

Where the answers already used here came from:

bintrp.mac    PEEK and POKE, the statement dispatch, expression evaluation
f4.mac        the arithmetic and the maths library
bimisc.mac    RND and RANDOMIZE
bistrs.mac    string handling, VARPTR
bio.mac       terminal I/O
dcpm.mac      the CP/M interface

Grep for the keyword in lower case - the sources are written in lower case with labels like peek: and poke:.

What it does not replace

The binary under cpmemu is still the arbiter for anything observable, because it is the thing users actually ran and because a reconstruction can only be as right as its verification. The source says what the code does; the binary says what happens. Use the source to find the answer and the binary to confirm it - that is how RND_ALGORITHM.md was settled, and the two agreed.

See tests/HOW_TO_RUN_REAL_MBASIC.md for running the binary.