The integer-arena approach (parallel global arrays) never freed nodes,
causing 6GB+ memory usage when running the self-hosted kernel. PHP
arrays don't shrink and we have no GC for them.
- Replace int-arena with plain Node objects so PHP's refcounting GC
reclaims unreachable subtrees automatically.
- Hash-cons Stem/Fork nodes to collapse duplicate immutable subtrees.
- Cache ofNumber(0..255) to avoid re-creating byte trees.
- Eliminate indirection (tag 4) nodes entirely; projection rules now
eagerly reduce and copy tag/a/b into the original App node.
- Remove all deref() loops — every node is exactly what it says.
Memory: 6GB+ → 24MB for `id "hello"` with runArboricxToString kernel
Replace JSON-based bundle manifest with a CBOR-encoded format. The manifest
is now a canonical CBOR map with order-strict key decoding, raw 32-byte hash
payloads (instead of hex-encoded JSON), and compact binary representation.
freeVars did not descend into TStem, TFork, or SList, so dependency analysis
under structural nodes and lists was silently missed.
toSKIDB's _other = _K `SApp` TLeaf fallback returned a constant leaf when the
binder occurred under a structural node, losing the abstraction entirely.
Replace with explicit lowering: BStem/BFork/BList are converted to application
form before SKI abstraction, and any other unsupported DB term errors explicitly
Reorder recursive byte-stream consumers so the consumed input is inspected
before loop-control arguments can drive evaluation. Previously, partially
applying `readBytes` to a known count, such as `readBytes 2`, allowed the
evaluator to specialize the recursive worker using known counter values
while the byte stream was still abstract. This caused symbolic recursion
over unknown input and produced an enormous normal form.
The recursive worker now takes the byte stream first and immediately
case-analyzes it. As a result, partial application blocks at the input
boundary instead of unrolling the counter loop.
This preserves the fully-applied behavior of `readBytes`, while making partial
application such as `readBytes 2` normalize safely.
Add multi-root bundle support across the toolchain:
- `compile`: Accept multiple definition names via `-x NAME` (repeatable or
comma-separated). Exports all requested definitions as named roots in a
single bundle. Defaults to "main" when no names are given.
- `export`: Accept comma-separated hashes in the positional argument and
multiple `-n`/`names` flags. Exports all resolved roots in one bundle.
- Server: Add `GET /bundle/roots?n=...&h=...` endpoint that resolves
multiple stored-term names and/or raw Merkle hashes, returning a single
bundle containing all of them as roots.
- Wire: Export `defaultExportNames` helper for generating default export
names when none are supplied.
- Drop `cereal` dependency from `tricu.cabal` (no longer used).
Introduces a read-only HTTP server (WAI/Warp) backed by the content
store, exposing three bundle-export endpoints:
- GET /bundle/name/:name — export by stored term name
- GET /bundle/hash/:hash — export by full Merkle hash
- GET /terms — plain-text listing (debug)
Also adds `tricu server` (aka `--serve`) CLI mode, move `resolveExportTarget` /
`namesForHash` / `looksLikeHash` out of `Main.hs` into `ContentStore.hs`,
and cleans up unused exports and imports across `FileEval.hs` and `Wire.hs`.
