refactor(pi): extract slash/explicit command parsing (Unit 2) - #6
Merged
Conversation
Closes out the W3 vocabulary: every JS dispatch (onOpen, onMessage, onClose) transitions the connection through live -> parked, and a new scanIdle(threshold_ms, now_ms) sweep transitions long-parked connections to dormant while returning the crossed-over ids. live connections are never swept (a handler blocked mid-dispatch is a handler bug, not a hibernation case); already-dormant connections aren't re-reported. What this closes: - the ConnectionState enum is now actually used (previously set to .parked on every frame, never distinguished from .live) - the scanIdle hook gives an evented-loop successor a clean point to deregister dormant fds' read side without hunting the pool itself What this doesn't close: - per-connection thread cost. ws_frame_loop is still thread-per- connection, so "dormant" today is an observational label rather than a resource-savings win. The real thread release needs a shared evented-I/O replacement for ws_frame_loop.run plus a timer that drives scanIdle periodically. That's a separate, larger refactor — flagged as a future optimization once measured pressure justifies it. What was already true before this commit: - runtime is released between messages (handler pool lease only held during dispatch), so "dormant holds no runtime" is already the case for every parked connection - setAutoResponse (W3-a) short-circuits matching frames without waking JS - ping/pong are answered by the codec, never dispatched to JS Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds two module-level WebSocket consistency rules surfaced through
the existing rule-registry + expert tooling pipeline:
ZTS320 websocket_import_without_events (warning): handler imports
zigttp:websocket but doesn't export onMessage. The module surface
is useless without at least one event handler.
ZTS321 websocket_events_without_import (error): handler exports
onOpen/onMessage/onClose/onError but doesn't import
zigttp:websocket. The exports would be dispatched but
send/close/etc. would be unreachable.
The checker (packages/zigts/src/ws_consistency.zig) is a pure
function over a small Inputs struct, trivially testable without
standing up the full contract-builder pipeline. Seven unit tests
cover every combination of imports_websocket_module x
exports_on_(open|message|close|error).
ContractBuilder.build() calls the check after its existing scans
and logs findings to stderr during every contract build, so
zig build -Dhandler=x.ts and `zigttp compile` both surface them.
zigts describe-rule / zigts search / zigts meta all see the new
entries.
Golden fixtures (meta, verify_paths_clean, verify_paths_missing)
and policy-hash.txt refreshed to match the two added rules
(policy hash 9ede01bd... -> 422ab281...; rule count 25 -> 27).
Deferred for a follow-up slice:
- Durable-fetch misuse (ZTSF00x): needs call-site AST walking of
fetch(...) invocations to inspect the durable option object.
The rule-registry entries and runtime-side validation (which
already rejects malformed durable options at call time) are in
place; compile-time diagnostics are the remaining piece.
- Separate ZTSW/ZTSF prefix categories: current rules live in the
ZTS3xx verifier range to reuse existing infrastructure. A split
requires extending RuleCategory and the policy hasher.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…e D)
Documents zigttp:websocket and zigttp:fetch as first-class virtual
modules with full API references, runtime flags, and not-yet
sections that enumerate the deliberately deferred pieces
(thread-cost hibernation, durable-fetch compile diagnostics, etc).
New examples:
- examples/websocket/chat.ts: named-peer broadcast handler using
serializeAttachment for display names and getWebSockets(room) for
fan-out. Smoke-tested end-to-end.
- examples/fetch/webhook.ts: idempotency-keyed POST dispatcher with
retry + exponential backoff, annotated with the SIGKILL-restart
replay protocol.
Both examples compile clean against the current ruleset (no ZTS3xx
diagnostics, policy hash unchanged).
Extends the AWS deploy manifest's Metadata.zigttp block with two
optional sub-keys:
- `websocket`: { onOpen, onMessage, onClose, onError } emitted when
the contract has any WS event export. Deploy consumers that need
a different binding (Cloudflare Durable Objects, Azure WebPubSub)
branch on this instead of re-parsing the contract.
- `fetchHosts`: the egress host list scoped to handlers that import
zigttp:fetch, so targets can configure a narrower allow-list
than the full egress set.
Both blocks are omitted entirely when the underlying facts are
empty, keeping existing-handler manifests byte-identical. Three
new renderAws tests cover present/absent for each block.
A full Cloudflare DO target remains a separate plan — the metadata
hook gives downstream tooling what it needs to emit one without
forcing the zigttp core to carry CF-specific rendering.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Fresh-eyes cleanup on the recently added WebSocket runtime: removed unused FrameEvent union and max_room_peers constant, inlined a one-line unixMillisNow wrapper, renamed Pool.unlockLock to unlock, and replaced hand-rolled asciiEqlCaseInsensitive and toUpper/toLower with their std.ascii counterparts used elsewhere in the codebase. Pure behavior-preserving; zig build test reports 997 + 146 passing, same as baseline. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Restructure the 22 virtual modules into six domain subfolders under packages/zigts/src/modules/, packages/zigts/module-specs/, and docs/virtual-modules/: security, data, net, http, workflow, platform. Infrastructure files (resolver, compiler, file_resolver, module_graph, util, types, root) stay at the modules/ top level. Also adds dedicated doc pages for the 18 previously-undocumented modules plus a virtual-modules README that serves as the landing page and effect-classification reference. builtin_modules.zig is the single source of truth: all 22 imports and governance entries updated with the new paths; a build-time assertion already guards specifier/entry drift. No handler-facing API change - specifiers are unchanged, only internal paths moved. README's virtual module table now links each row to its page; the duplicated effect classification paragraph is replaced by a link to the docs/README. zig build test reports 1503/1507 passing (4 skipped), identical to the pre-reorg baseline. Doc link audit: 23 pages, zero broken relative links. git mv preserves blame history for every moved file. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Align the code layout with docs/virtual-modules/: its top-level has only the six group folders plus a README index, while the code top level was cluttered with six infra files (resolver, compiler, util, types, file_resolver, module_graph) sitting next to the group folders. Moving those into packages/zigts/src/modules/internal/ leaves only root.zig and the six group subfolders at the code top level, matching the docs shape. No behavior change: ~38 import paths updated (31 infra refs inside 22 virtual-module files, 6 in modules/root.zig, 10 external callers in module_binding/trace/http/file_io/module_audit), plus the six moved files themselves bump their ../X.zig refs to ../../X.zig. Git mv preserves blame on every moved file. zig build test --summary all: 1503/1507 passing, 4 skipped, identical to the pre-move baseline. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds websocket capability, expanded ContractCategory variants (scope_name, cookie_name, cors_origin, rate_limit_key, service_call, fetch_host), sets_scope_used flag, Law/LawKind/AbsorbingPattern, and ModuleBinding.self_managed_io to zigttp-sdk so the SDK now covers every field zigts module_binding.zig carries. Accepts both zigttp: and zigttp-ext: specifier prefixes in validateBindings; adapter still enforces zigttp-ext: for extensions. Phase 0-1 (step 1) of the peer-package move in docs/virtual-modules-peer-package.md. Plan and SDK inventory doc checked in alongside.
Adds extractString/extractInt/extractFloat, createString/createObject, objectSet/objectGet, throwError, resultOk/resultErr/resultErrValue/ resultErrs to zigttp-sdk. Each wraps an extern fn whose zigts-side impl lives next to the existing capability exports in module_binding.zig. JSValue crosses the ABI as u64 since both packages verify the packed-struct layout is bit-identical. Module state (getModuleState/setModuleState) is deferred to a later commit because the deinit callback's std.mem.Allocator parameter is not C-ABI stable. Stateless modules (11 of 22) can port without it. Phase 1 step 2 of the peer-package move.
Creates the empty zigttp-modules package that will host the 22 built-in virtual modules after Phase 3-7 ports. Depends only on zigttp-sdk; builds as a standalone module with its own test step. Ports internal/util.zig to use SDK handle ops: extractString, createString, createObject, objectSet/Get, result builders, throwError/TypeError/CapabilityPolicyError. Modules written against this util can compile against zigttp-sdk alone, no zigts internals. Registered as dependency in root build.zig.zon but not yet imported into any binary; zigts continues to source modules from packages/zigts/src/modules/ until Phase 8 wire-up. Phase 2 of the peer-package move.
Moves the canonical implementation of zigttp:crypto into packages/modules/src/security/crypto.zig, depending only on zigttp-sdk. The zigts-side modules/security/crypto.zig becomes a two-line adapter that wraps the SDK binding via adapter.adaptModuleBinding. Phase 8 will delete the stub and wire builtin_modules.zig directly against zigttp-modules. SDK additions needed by this port: - getAllocator(handle) — borrow the runtime allocator - sha256(handle, data, out) + hmacSha256(handle, data, key, out) — capability-gated crypto wrappers so modules can avoid direct std.crypto calls forbidden by module-audit ZVM001 Also relaxes adapter.adaptModuleBinding to accept zigttp: builtin specifiers (previously only zigttp-ext:). The zigts dependency graph gains zigttp-modules; root build.zig wires it into the zigts-tests module too. governance update: builtin_governance_entries crypto module_path points to packages/modules/src/security/crypto.zig so verify-modules reads the canonical source. Phase 3 wave 1 of the peer-package move (crypto only; auth, validate, decode still to port).
Applied /simplify findings from three parallel reviews: - Delete 12 no-value pub const re-exports in packages/modules/src/internal/util.zig; crypto.zig calls sdk.* directly. Dropped the dead no-op test. - Cross JSValue directly across the extern ABI instead of u64. Both packages verify layout equivalence, so the .raw bookkeeping is ceremony; simpler signatures, same machine code. - Drop unused handle parameter from zigttpSdkSha256/HmacSha256. The capability check lives inside sha256ForActiveModule, which reads the active-module threadlocal, not the handle. - Drop the now-redundant requireCapability hop in sdk.sha256 and sdk.hmacSha256 wrappers. The zigts-side *ForActiveModule already enforces the check; double enforcement was one extern call per crypto op. - getAllocator returns *const std.mem.Allocator directly instead of *anyopaque + @ptrCast/@constcast dance. - Collapse six near-identical ordinal-check if-blocks in module_binding_adapter.zig into a single assertOrdinal helper. - Strip commit-referencing rot ("Phase 8 will delete…") and narrating block comments. The design doc is the source of truth. Tests stay at 1503/1507.
Moves zigttp:auth into packages/modules/src/security/auth.zig. Zigts-side stub delegates via adapter.adaptModuleBinding. SDK gained one new helper: parseJson(handle, json) which wraps builtins/json.zig's parseJsonValue. Everything else auth needs (hmacSha256, getAllocator, extractString/Float, createString, objectGet for exp/nbf claim lookup, nowMs, resultOk/Err) is already in the SDK from prior commits. Coverage gap: 9 unit tests from the original auth.zig (base64url roundtrip, pure byte-compare tests) don't migrate yet — the modules package cannot link zigts exports standalone, so its test root fails to resolve zigttpSdk* symbols. Proper fix requires the SDK test-harness planned for Phase 1 follow-up. Tests exercised std behavior and are covered indirectly by integration. Phase 3 wave 1 (auth). validate + decode still to go.
Completes Phase 3 of the peer-package move. Both modules now live
in packages/modules/src/security/; zigts-side files reduced to
two-line adapter stubs.
SDK additions needed:
- isString/isObject/isArray/arrayLength/arrayGet/arraySet/
createArray/stringify as extern-backed helpers
- isNull/isUndefined/isTrue/isNullish/isNumber as inline JSValue
methods (bit-pattern checks, no extern)
- getModuleState/setModuleState with SdkStateEnvelope wrapper on
the zigts side. SDK callback uses `callconv(.c)` with just the
state pointer; zigts stores the SDK deinit in an envelope and
registers a static internal deinit_fn that invokes it. Works
around std.mem.Allocator not being C-ABI stable.
- parseJson(handle, json) wrapping builtins/json.zig parseJsonValue
SchemaRegistry is now allocated from sdk.getAllocator(handle) and
stored via sdk.setModuleState; its deinit is a `callconv(.c)`
function that frees the registry through its stored allocator.
decode.decodeJson/decodeObject become module-to-module API in the
package, called directly from decode.zig via `@import("validate.zig")`.
Coverage: 21 unit tests from validate + decode don't migrate since
the modules package cannot link zigts exports standalone (test
discovery boundary across the Zig module graph). Same follow-up as
the auth port — the SDK test-harness needs to come before Phase 8.
Tests: 1473/1477 (was 1494 before this commit, 1503 at session
baseline; net -30 tests, all from module-package test discovery).
Phase 4 of the peer-package move. All five modules now live in packages/modules/src/platform/; zigts-side files reduced to two-line adapter stubs. SDK additions needed: - readEnv(handle, name) — capability-gated env var read (wraps zigts readEnvChecked) - allowsEnv(handle, name) — policy-check gate, used by env before reading - objectKeys(handle, obj) — enumerate own keys as a JS array; log module iterates context-object properties for NDJSON output - isFalse on JSValue (inline, bit-pattern) Updated the "governance entries stay aligned" test in builtin_modules.zig to assert the new env module_path. 30 pure tests in the ported files don't run yet (same modules-package test discovery gap as Phase 3). Tests: 1455/1459 green (33/35 build steps, all OK).
Phase 5 of the peer-package move. All three http modules now live in packages/modules/src/http/; zigts-side files reduced to two-line adapter stubs. No new SDK surface was required - existing handle ops (objectGet, objectKeys, arrayLength, arrayGet, extractString/Float, createObject, objectSet, createString, isObject/isTrue/isFalse) covered every zigts-internal call the three modules made. The router's object-key iteration uses sdk.objectKeys → JS array, trading two extern-fn hops per iteration for ABI-stability. Cost is a single-digit-ns per route-table entry; negligible for the typical ≤ dozen routes in a handler. Tests: 1441/1445. 14 pure unit tests in the ported files don't run under test-zigts yet (known modules-package test discovery gap).
The pi loop tests called std.testing.tmpDir(.{}) (which creates
.zig-cache/tmp/<sub_path>/) but then passed only tmp.sub_path (the 16-char
random component) as workspace_root. std.fs.path.resolve joined that
against CWD, so every test write landed in <repo>/<sub_path>/ instead of
the tmp dir. tmp.cleanup() only deleted the .zig-cache/tmp side, leaving
one stray repo-root folder per test run.
Compose the full relative path via a tmpWorkspacePath helper so writes
land inside the real tmp dir. Also update the .gitignore glob comment to
reference the fix; the glob stays as defense-in-depth.
Phase 6 (data group), first two modules. Both are stateful with per- runtime stores owned through setModuleState/sdkDeinit. ratelimit uses the clock capability; cache additionally uses policy_check to gate the namespace through the capability policy. Added to the SDK: - numberFromF64 helper that folds safe integers into the int32 tagged form, matching the runtime's allocFloat hot path. - allowsCacheNamespace wired through a new zigttpSdkAllowsCacheNamespace extern; the zigts-side implementation delegates to allowsCacheNamespaceChecked. sql stays in packages/zigts/src/modules/data/ for now; it needs a larger SDK surface for prepare/step/bind/columns and will land in a follow-up commit.
Completes Phase 6 (data group). The SQL module now lives in packages/modules/src/data/sql.zig and talks to SQLite through a new SDK surface. Thirteen extern fns cover what sql.zig needs: open/close/errmsg/changes/lastInsertRowId for the db, and prepare/finalize/step/readonly/stmt_errmsg/param_count/param_name/ bind_null/bind_int64/bind_double/bind_text/column_count/column_name/ column_type/column_int64/column_double/column_text for statements. Opaque SqliteDb and SqliteStmt handles are the raw *c.sqlite3 and *c.sqlite3_stmt pointers cast to opaque; no extra heap allocation. arrayPush and allowsSqlQuery rounded out the SDK additions. installStore keeps living on the zigts side of the split: it runs during runtime bootstrap, outside any module invocation, so it can't use the SDK's handle-gated setModuleState. The shim allocates the module's SqlStore via ctx.allocator and writes it directly into the slot the module reads from. Tests green at 1441/1445. The two sql integration tests from the previous file relied on full-context setup that only makes sense inside zigts itself; flagging those for the Phase 8 SDK test harness alongside the other cross-boundary unit tests.
Phase 7 partial. Four of seven net/workflow modules now live in packages/modules/. Each zigts-side file becomes a thin stub that adapts the modules-package binding via adapter.adaptModuleBinding and keeps an installState shim for runtime bootstrap to wire up callbacks. - compose is comptime-only, no runtime coupling. - fetch, service, websocket dispatch through runtime-installed callbacks. The state struct lives in the modules package with SDK-typed function pointers; the zigts-side InstalledState wraps the runtime's Context-typed callbacks with ModuleHandle-typed thunks that cast through handleToContext. SDK additions: isCallable, readFile + freeFileBuffer. Deferred to a later pass: durable, io, scope. durable has an un-diagnosed function-pointer mismatch between install-time and read-time addresses when going through the wrapper; io pulls fetchSync state via a threadlocal that's read by zigts HTTP code; scope manipulates GC roots directly. Each requires SDK surface or runtime-flow changes that don't fit in this phase. Tests green.
Phase 8. builtin_modules.zig now imports ported bindings directly
from the zigttp-modules peer package and adapts them in-place via
adapter.adaptModuleBinding. The 15 pure-adapter stub files under
packages/zigts/src/modules/{security,platform,http,data,workflow}/
are gone; the directory only holds the modules that still need
zigts-internal access:
- sql, fetch, service, websocket — installState/installStore shims
called from runtime bootstrap, outside any module invocation.
- io, scope, durable — genuinely coupled to zigts internals
(fetchSync threadlocal, GC roots, open function-pointer issue).
Ported bindings live under a local `ported` namespace so their
names don't shadow the `<name>_binding` locals in the governance
assertion tests at the bottom of the file.
modules/root.zig trimmed to only re-export the seven still-in-tree
modules plus the module resolver/compiler internals.
Tests green.
Specs move: git mv packages/zigts/module-specs/ packages/modules/module-specs/. The 22 JSON spec files live alongside their implementations now. Tooling: - builtin_modules.zig: spec_path strings in all governance entries point at the new packages/modules/module-specs/ root. - module_audit.zig: looksLikeBuiltinAdjacentPath accepts source under packages/modules/src/ as well as the legacy packages/zigts/src/modules/ (still holds io, scope, durable, and the install-shim files). - check-capability-helpers.sh: glob covers both module-source roots so the audit runs over the peer-package modules too. Docs: - README virtual-modules section notes the peer-package layout and which modules still live in zigts for runtime coupling. - docs/architecture.md: tree now shows packages/zigttp-sdk/ and packages/modules/; enforcement-points list points at the new paths; module-binding key-files list adds the SDK entry.
- Extract contextFromHandle, sdkValue, internalValue, internalArgs helpers in module_binding_adapter.zig; fetch/service/websocket shims use them instead of hand-rolled @ptrCast/@bitcast chains. - Trim WHAT-narration comments in builtin_modules.zig and the modules/root.zig preamble. Keep only the WHY lines. - Add a one-line rationale for the 20000 branch-quota bump in validateBindings (was 5000 default; O(n*m) name/specifier dup checks blow it out once the full roster hits). - Drop redundant freeFileBuffer from the SDK: readFile already allocates via the runtime allocator exposed by getAllocator, so callers free through that. Also drops the paired zigts-side bridge fn.
FetchState.sdkDeinit was never referenced — the zigts-side shim overrides deinit with its own stateDeinitAdapter. The allocator field only existed to feed that dead method, and its remaining use in stateDeinitAdapter is redundant with the allocator that setModuleState passes in as the second arg.
…ator field Same pattern as fetch: sdkDeinit was never referenced, and the allocator field only fed it. The zigts-side stateDeinitAdapter now uses the allocator passed in by setModuleState.
Never referenced. deinitSelf stays (called by the zigts-side shim's stateDeinitAdapter for the StringHashMap teardown).
Move the REPL's slash-command and explicit-command parsing out of repl.zig into a dedicated commands.zig module with a single command table driving both forms. repl.zig now focuses on dispatch, tokenization, and help rendering. Unit 2 of the parallel refactor plan: /Users/srdjans/.claude/plans/study-the-plan-provided-clever-wadler.md Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
srdjan
added a commit
that referenced
this pull request
May 24, 2026
nextGuardId returned a bare counter starting at 1. The value is stored verbatim in JSObject.inline_slots[FUNC_GUARD_ID] and the GC walks every inline slot as a JSValue via markValue -> isPtr. A counter value whose top 16 bits landed in the NaN-box pointer prefix range (0xFFFC_xxxx_xxxx_xxxx) would be misclassified as a heap pointer and dereferenced. The counter wraps to that range after ~281 trillion calls, but a hash-seeded or PRNG-seeded variant could land there immediately. Step the counter through the 48-bit payload space and OR in value.JSValue.INT_PREFIX (0xFFFD_xxxx_xxxx_xxxx) before returning. The JIT loads the slot as raw u64 (movRegMem / OBJ_FUNC_GUARD_ID_OFF) and compares against callee.guard_id, also raw u64; both sides carry the prefix bits so the equality check still passes byte-for-byte without any tag-aware code. Skip 0 so the sentinel is reserved. Two tests: 1024-iteration property test confirms (id & PREFIX_MASK) == INT_PREFIX and uniqueness across the run; end-to-end ensureGuardId test asserts !slot.isPtr() and !slot.isExternPtr() on the stored value. Intent: fix-defect Scope: zigts/bytecode, zigts/value, zigts/gc, zigts/jit Decided-Against: a parallel non-JSValue slot for guard IDs (touches createBytecodeFunction, createClosure, and 15 JIT call sites) Decided-Against: marking the slot as untyped via a parallel bitmap (GC walks all inline slots uniformly today) Refs: 2026-05-23 overall-repo review, P0 #6 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
srdjan
added a commit
that referenced
this pull request
Jun 28, 2026
…adapter xhigh review #2/#6: the SMT solver adapter collapsed every failure into one verdict and its module doc described the opposite policy from the code. - parseVerdict only inspected the first stdout token, so a z3 startup banner or a `success` line under :print-success was misread as a solver error. Scan for the first sat/unsat/unknown token instead, skipping benign preamble; an `(error ...)` anywhere still wins (fatal). Add banner/print-success/error-wins tests. - A pre-exec failure to RUN z3 (spawn/IO error, or z3 gone since available()) now maps to .unknown (environmental, benign/unproven), not .solver_error. .solver_error is reserved for z3 RUNNING and emitting no usable verdict - the only case that signals the present solver cannot evaluate the model. This stops a transient spawn hiccup from failing the audit with a misleading "missing FP/String theory" message while keeping the fail-closed property (an incapable z3 still manifests as z3 output -> .solver_error -> ZTS758). - Rewrite the stale module doc (claimed spawn/IO -> .unknown and "inherits the parent environment" while the code uses an empty environ) to describe the actual two-class policy. spec-check --audit: 13/13 SMT proved, 4/4 refuted, PASS. test-zigts-cli 76/76. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
srdjan
added a commit
that referenced
this pull request
Jul 12, 2026
pi's PropertiesSnapshot now mirrors the engine's full HandlerProperties boolean surface: adds cost_bounded and post_only (both were silently missing). The proof card highlights a `cost_bounded` chip, guaranteeCounts folds the new dimensions into the /ledger proven-path ratio (16/16 -> 18/18), and the persona gains a "Cost bounds" section teaching the loop-bound discharge forms (literal arrays, range(n), SQL LIMIT, schema maxItems). The engine->pi mapper is now a compile-time field copy instead of a hand-written literal, removing that drift point, and a comptime drift gate asserts pi mirrors every engine boolean property - a new proof dimension added to the engine now fails the build until the expert surfaces it (verified by temporarily dropping a field: the gate fires with an actionable message). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
packages/pi/src/repl.ziginto a newpackages/pi/src/commands.zigmodule.command_tablenow drives both forms (/meta,/features,/modules,/rule,/search,/verify,/check,/build,/testandzigts meta|features|modules|search|describe-rule|verify-paths|verify-modules|check ...).zig build <step>/zig build test*stays as a short special case next tolookup.isQuit/isHelpmove with the parsing;repl.zigkeepsdispatchLine,shouldDispatchTool,DispatchOutcome,Registry,rununchanged sotui/app.zigandapp.zigcallers and tests need no edits.commands.zigcovering slash lookup, explicit lookup, unknown command,zig build test-zigtsrouting, trailing args, and quit/help aliases.This is Unit 2 of the parallel refactor plan at
/Users/srdjans/.claude/plans/study-the-plan-provided-clever-wadler.md.Test plan
zig build test-expert-apppasseszig buildsucceedsprintf "help\nquit\n" | ./zig-out/bin/zigts expertshows/verifyandRegistered tools:printf "/meta\nquit\n" | ./zig-out/bin/zigts expertreturns"compiler_version"JSONprintf "zigts meta\nquit\n" | ./zig-out/bin/zigts expertreturns"policy_version"JSONGenerated with Claude Code.