Week 1 Studio: Getting started

Toolchain setup, and a skeleton hand-rolled lexer that recognises integer literals — the smallest thing the course can call a running compiler

22 February 2027

Before the session#

Arrive with an OCaml toolchain installed: an opam switch and dune for building. The setup page has the exact switch-creation steps and the versions this course’s own reference implementation was built against — follow it and confirm dune build runs before you arrive. If you have never set one up before, budget real time for this — a broken toolchain on day one costs you every week after it, and it is much cheaper to fix before the course needs it for anything.

If you’ve never written OCaml before, the OCaml crash course is a short, scoped-down read — let, pattern matching, and variant types, the three things this week’s lexer actually uses, with real code from it.

Skim the lecture’s naming-dispute aside again before you arrive. You will be asked, briefly, which account you’d trust and why.

In the session#

• Toolchain check (15 minutes). Confirm dune build runs a trivial project end to end. If it doesn’t, this is what we fix first — everything else in the semester assumes it works.
• Skeleton lexer (main work). Starting from a bare dune project, write a small hand-rolled lexer — a function that scans a string character by character with an index and produces one rule: match one or more digits and produce an integer-literal token. Hand-rolled rather than ocamllex-generated from the very first week, on purpose: SNOBOL4’s arithmetic operators are blank-sensitive (the same character can be a binary operator, a unary operator, or a concatenation boundary, depending on what’s adjacent to it — you’ll meet this properly in week 2), which needs an explicit character-by-character scan tracking “was the previous character a blank” rather than a generated automaton’s regex-driven rules. Starting hand-rolled avoids swapping lexer technology out from under a working pipeline partway through Movement I. This is deliberately the smallest possible lexer — no whitespace handling, no operators, no strings yet. The point is to see the whole pipeline (source string → a token you can print) working before it has to do anything interesting.
• Wrap-up. Everyone should leave with a lexer that can turn "42" into a token and print it back out. That is week 1’s entire deliverable, and it is intentionally small: the shape of the pipeline matters more this week than its coverage.

Sample transcript#

Genuine captured output from this course’s own week-1 reference lexer, run against a three-line file of bare integer literals:

$ dune exec bin/main.exe -- examples/digits.sno
INT 42
EOF
INT 7
EOF
INT 100
INT 200
EOF

The third line, 100 200, is the one worth tracing by hand: it holds two literals separated (and preceded) by blanks, and the only thing this week’s lexer has to get right is that a run of digits becomes one INT token while a blank is just a separator — the trace should show a single EOF closing that line, not one per literal.

This week’s real starter — the same lexer skeleton with the digit-recognition rule itself stubbed out — is downloadable at week-01-starter.zip.

Afterwards#

What you build this week is the foundation the next three weeks extend directly — the arithmetic lexer in week 2, the parser in week 2–3, and the evaluator in week 4 all sit on top of this file rather than replacing it. Keep it; you’ll be growing it, not rewriting it.