aowlhl

The shared high-level IR layer for aowl's idiomatic backends. It reads a sem'd typed NIF (<module>.s.nif, pre-hexer) and exposes it as a small, target-neutral surface every HL backend consumes: the module-import graph, the dependency-first initialization order, the routine classification (foreign / closure), and the grammar walk skeleton (shape decoders for locals, params, procs, if and case).

The point is that walking and classifying the HL-IR is done once. A backend imports aowlhl and supplies only the action — the aowli interpreter/VM evaluates each node; the aowljs emitter prints the equivalent JavaScript — over the same shapes.

Status: landing. Three modules are live and consumed by two real backends:

• aowli (interpreter / VM) — uses the loader (hlload) for module-init order, the classification (hlclassify) for foreign/closure dispatch, and the walk skeleton (hlwalk) for if/case.
• aowljs (JavaScript emitter) — uses hlwalk for local/param/proc/if/case; output is byte-identical to the hand-navigated emitter it replaced.

📖 Full docs → aoughwl.github.io/docs/aowlhl

Modules

Module	What it shares
aowlhl/hlload	the user-module import graph + moduleInitOrder (dependency-first init)
aowlhl/hlclassify	routine-pragma classification — hasImportc / importcName / hasClosure (+ generic hasPragma / pragmaStrArg)
aowlhl/hlwalk	grammar shape decoders — decodeLocal / decodeParam(s) / decodeProc / decodeIf / decodeCase

Each decode* takes a Cursor at a node's (, advances it past the whole node, and returns a shape of captured sub-cursors. Cursors index an immutable token buffer, so those sub-cursors stay walkable after the node is consumed — the backend drives evaluation/emission over them in whatever order it needs.

Consuming it

aowlhl is a source library, consumed the way the rest of the stack consumes sibling checkouts — add -p:/path/to/aowlhl/src at build and import aowlhl/…. It reuses aowl's sem'd-NIF reader and IR libraries via --path; it vendors no compiler source. See docs/design.md.

The wider aim

Because the semantic model — not surface syntax — is what the HL backends share, aowlhl is also the home for the deeper lowering the idiomatic TypeScript and Python emitters need (structured control flow, iterator inlining, value-copy placement, Option/distinct decisions). That design is captured in docs/design.md; the modules above are its foundation.