Next Session — Stack-Ranked Remaining Work

Baseline: 29628f7e (post 15b cascade-stop fix, trunk) Session shape: Attack items 1-4 in order. Each is independent — commit after each one. Tree state: Clean, guard stamp valid, smoke green.

Pre-flight (2 min)

cd /Users/mathisto/projects/quartz
git log --oneline -3     # 29628f7e at top
git status               # clean
./self-hosted/bin/quake guard:check   # stamp valid

Item 1: SEND-RECV-SHADOW (1-2h)

The bug

extern "C" def send(sockfd, buf, len, flags) and extern "C" def recv(sockfd, buf, len, flags) in std/ffi/socket.qz:168,174 shadow the 2-arg channel builtins send(ch, val) and recv(ch). The externs win because they’re registered as concrete functions in the resolver’s all_funcs vector, which is checked after builtins in typecheck_expr_handlers.qz:2126-2132.

Any spec importing the prelude chain that uses bare send(ch, val) or recv(ch) gets QZ0200: requires at least 4 arguments, got 2.

Blocked specs

At minimum: backpressure_spec, semaphore_spec, graceful_shutdown_spec. Likely also: concurrency_spec, channel_handoff_spec, unbounded_channel_spec, async_channel_spec, send_sync_spec, async_spill_regression_spec.

Fix

Option A (recommended, simplest): Rename the POSIX externs in std/ffi/socket.qz:

send → posix_send (or socket_send — wrappers at line 395-401 already exist as socket_send/socket_recv)
recv → posix_recv
Also rename sendto → posix_sendto, recvfrom → posix_recvfrom (preventive)

Zero external users of the socket API. The existing socket_send/socket_recv wrapper functions at lines 395-401 are the intended public API anyway.

Option B: Add arity-aware resolution so send(ch, val) with 2 args resolves to the builtin. More complex, touches the resolver.

Verification

After renaming:

quake guard (fixpoint — the compiler itself doesn’t use these externs)
Run each unblocked spec individually (subprocess specs may hit the separate %push IR bug — see note below)
Verify std/ffi/socket.qz’s own socket_send/socket_recv wrappers still work (they call the renamed externs)

Key files

File	What
`std/ffi/socket.qz:168,174`	The extern declarations to rename
`std/ffi/socket.qz:395-401`	Existing wrapper functions that call the externs
`self-hosted/middle/typecheck_builtins.qz:820-821`	`send`/`recv` builtin registration (don’t change)

Item 2: Generic UFCS Dispatch (1-2d)

The bug

spec/qspec/generic_ufcs_dispatch_spec.qz — 8 tests, all fail. Bounded generic functions like def sum_iter<T: Iterator>(iter: T): Int with for x in iter in the body emit malformed code.

Root cause (from investigation)

The issue is in how for-in loops dispatch on Iterator-bounded type params. The dispatch chain:

typecheck_generics.qz:608-613 — marks Iterator-bounded type params
mir_lower.qz:3180-3193 — during specialization, flags params via ibparams
mir_lower_stmt_handlers.qz:1232-1242 — for NODE_FOR, checks if iterator is bounded → calls lower_closure_iteration()
mir_lower_iter.qz:615-722 — lower_closure_iteration() generates closure-call blocks

The ROADMAP says the enddef token in the output suggests “iterator closure synthesis breaks the block structure.” Start by compiling a minimal bounded-generic-with-for-in program and examining the IR output. The bug is likely in lower_closure_iteration() around lines 696-722 — block terminator or scope management.

Approach

Write a minimal reproducer (3-line bounded generic with for x in iter)
Compile with --dump-mir and --dump-ir to see where the block structure breaks
Fix in mir_lower_iter.qz
Run all 8 tests in generic_ufcs_dispatch_spec.qz

Key files

File	Lines	What
`spec/qspec/generic_ufcs_dispatch_spec.qz`	all	8 test cases
`self-hosted/backend/mir_lower_iter.qz`	615-722	`lower_closure_iteration()` — likely fix site
`self-hosted/backend/mir_lower_stmt_handlers.qz`	1232-1242	NODE_FOR bounded-param dispatch
`self-hosted/middle/typecheck_generics.qz`	608-613	Iterator-bound tracking

Note: This spec uses import * from qspec/subprocess. Subprocess-based specs currently hit a separate %push IR bug in the qspec infrastructure (see note at bottom). You may need to convert tests to direct (non-subprocess) form to verify, or fix the %push issue first.

Item 3: Move Semantics Enforcement (2-3d)

The bug

6 safety holes remain open in the move checker. The move/borrow infrastructure exists (tc_move_value, tc_move_on_call in typecheck_registry.qz) but isn’t wired into all AST node handlers.

Open holes

Hole	AST Node	What’s Missing	Fix
#4	NODE_WHILE	Move states not restored after loop body error	Wire `tc_move_on_call` into the while-loop body handler
#5	List comprehension	Move tracking inside comp bodies	Wire `tc_move_on_call` into comprehension body lowering
#8	NODE_RETURN	Return expression position doesn’t mark value as moved	Add `tc_move_on_call` to return handler for Drop types
#10	NODE_MATCH	Match subject not consumed	Add `tc_move_on_call` for Drop-typed match subjects
#11 partial	NLL + move	Double-move after NLL borrow release should error	Verify NLL borrow release properly re-enables move tracking
#12	Return borrow escape	Multi-level borrow escape (fn A returns &local, fn B calls A)	Extend QZ1210 checks to transitive borrow escape

Already fixed (don’t re-break)

Holes #3, #6, #7, #9, #13 are all passing. These are in s25_safety_holes_spec.qz.

Approach

The move checker is in self-hosted/middle/typecheck_walk.qz and typecheck_expr_handlers.qz. The pattern for each hole is:

Find the handler for the AST node (e.g., NODE_WHILE handler in typecheck_walk.qz)
Add tc_move_on_call(tc, var_name, context_str, ...) at the appropriate point
Run the spec to verify the error fires

Error codes: QZ1212 (use-after-move), QZ1210 (borrow escape).

Key files

File	What
`spec/qspec/s25_low_holes_spec.qz`	Tests for open holes
`spec/qspec/s25_safety_holes_spec.qz`	Tests for fixed holes (regression guard)
`self-hosted/middle/typecheck_registry.qz`	`tc_move_value()`, `tc_move_on_call()`
`self-hosted/middle/typecheck_walk.qz`	AST statement handlers
`self-hosted/middle/typecheck_expr_handlers.qz`	AST expression handlers

Note: Both s25 specs use import * from qspec/subprocess and currently hit the %push IR bug. You’ll need to either fix the %push issue first, or convert these tests to non-subprocess form to verify your changes.

Item 4: Module Path Resolution (Small)

The bugs

modules_spec: Tests file resolution using spec/qspec/fixtures/math_helper.qz. Needs the fixtures directory passed via -I. The subprocess test helper at std/qspec/subprocess.qz:508-514 may not pass the right paths.
accept_worker_spec: Uses import std/net/http_server (hierarchical path). File exists at std/net/http_server.qz but resolution fails.

Root cause

In self-hosted/resolver.qz:278-316: hierarchical path resolution (resolve_is_hierarchical() detects / in path) only tries base_path/module_name.qz and base_path/module_name/mod.qz, then searches -I include_paths. For std/net/http_server, it needs either -I . (project root) or smarter prefix resolution.

Approach

Read resolver.qz:278-316 (hierarchical path handling)
For accept_worker_spec: the fix is likely to search include_path + "/" + hierarchical_path + ".qz" by splitting the hierarchical path and walking the directory tree from each include dir
For modules_spec: verify the subprocess test helper passes -I spec/qspec/fixtures

Key files

File	Lines	What
`self-hosted/resolver.qz`	278-316	Hierarchical path resolution
`spec/qspec/modules_spec.qz`	~151	File resolution test
`spec/qspec/accept_worker_spec.qz`	1	Multi-level import
`std/qspec/subprocess.qz`	508-514	Include path setup for subprocess tests

Known blocker: `%push` IR bug in qspec subprocess infrastructure

Several subprocess-based specs (s25_low_holes, s25_safety_holes, generic_ufcs_dispatch, and others) produce llc: use of undefined value '%push'. This is a UFCS codegen bug where result.push(...) in std/qspec/subprocess.qz:1096 emits an indirect call through a %push local variable that was never declared.

This is pre-existing (the affected specs never compiled successfully) and appears to be a name-resolution collision in the qspec subprocess module’s import chain. It is NOT caused by the 15b fix.

If this blocks your work on Items 2-3: Either fix it first (grep for push in the import chain to find the collision) or convert the affected tests from subprocess-based assert_run_exits to direct assert_eq-based tests that don’t need subprocess compilation.

After Items 1-4

The next tier (from the prioritized list) is:

Collection stubs (reversed, sorted, unique, flatten, etc.) — collection_stubs_spec, 21 tests, stdlib table stakes
VS Code extension — .vsix build, syntax highlighting
Stdlib narrative guide — “how do I…” documentation
Pattern matrix exhaustiveness — non-exhaustive match detection
String ergonomics — String + Int auto-coerce

Prime directives check

D1: Items 1-4 are highest-impact for language usability. SEND-RECV-SHADOW is trivial but unblocks 6+ specs.
D2: Research done for all 4 items. Root causes identified.
D5: The %push IR bug is real. Don’t claim specs pass without running them.
D6: Every hole gets filled or filed. Update ROADMAP after each fix.
D8: quake guard before every commit.

Next Session — Stack-Ranked Remaining Work

Pre-flight (2 min)

Item 1: SEND-RECV-SHADOW (1-2h)

The bug

Blocked specs

Fix

Verification

Key files

Item 2: Generic UFCS Dispatch (1-2d)

The bug

Root cause (from investigation)

Approach

Key files

Item 3: Move Semantics Enforcement (2-3d)

The bug

Open holes

Already fixed (don’t re-break)

Approach

Key files

Item 4: Module Path Resolution (Small)

The bugs

Root cause

Approach

Key files

Known blocker: %push IR bug in qspec subprocess infrastructure

After Items 1-4

Prime directives check

Known blocker: `%push` IR bug in qspec subprocess infrastructure