This repository was archived by the owner on Oct 14, 2026. It is now read-only.
Repository navigation
v0.24.0
Beagle v0.24.0
Beagle v0.24.0 makes macro expansion structurally hygienic, adds a typed
structural matcher for authoring macros, and turns the full compiler gate into
a bounded parallel release preflight. The Racket compiler and checked-in
self-hosted compiler carry the same macro and identity contracts.
Language and compiler
- Macro hygiene now follows lexical scope identity end to end. Binding and
reference edges are resolved before lowering, preserved through source facts
and hosted emitters, and serialized through the self-hosted compiler. Local
shadowing, nested expansion, and intentional capture therefore retain the
identity chosen at the macro boundary instead of depending on rewritten
spellings. - Stable lexical identities survive layout-only source changes. Scope and
reference identities derive from structural provenance rather than transient
names, so formatting or unrelated source movement does not silently remint
the branch-corpus identity oracle. - Macros can use
syntax-matchfor structural dispatch. Patterns can
distinguish identifiers and literals, match list, vector, and map shape,
capture a final tail, and select clauses in source order. Invalid categories
and malformed tail patterns report at the pattern span, while successful
expansions continue through normal type checking. - The self-hosted compiler implements the same matcher. The checked-in seed
mirrors structuralsyntax-match, typed constant expansion, reachable branch
checking, and malformed-pattern diagnostics, including a fixture that gates
oracle and self-host agreement.
Test infrastructure and performance
- The full release tier runs as supervised parallel shards. Each setup phase
and shard has an explicit deadline, completion receipt, retained failure log,
and subtree reaping. Worker results merge deterministically into the existing
active, demoted, and gated-tier verdict. - Cold work is balanced and shared instead of repeated. Shard assignment is
based on measured compiler-heavy costs; Wasm tests share canonical generated
fixtures; and Native Core checkpoints reuse verified frozen programs before
C17 materialization. Resolved Wasm toolchain work has its own measured
process-lane capacity instead of inheriting the former missing-tool skip
cost. The ordinary active loop remains separate from the full release
preflight. - The parallel C17 checkpoint suite is advisory for this seal. Three
independent release-preflight failures came from cold-cache prerequisites or
scheduling rather than candidate behavior; the bounded replay of the final
failure passed and verified the published checkpoint. The suite still runs
under its unchanged 90-second owner, but it must be requalified before a
later release can promote it back to blocking status.
Compatibility
This pre-1.0 minor release removes the legacy name-rewriting hygiene path.
Macros or tools that inferred binding identity from rewritten symbol spellings
must consume the compiler's structural binding and reference identities. The
new syntax-match form is additive.
Full Changelog: v0.23.0...v0.24.0