Repository navigation
t27b dogfood: our own specs first #6457
Description
The owner's order (2026-10-05 ~15:30Z): t27b is developed by dogfooding -- our own .t27 + t27c, written on ourselves, fixing what that turns up. This issue tracks the dogfood view of the t27b lab; it is a lane-tracking issue, not a task card.
Measured at ~15:30Z (origin/master 11f5e6a, lab run 682564f)
| What | Measured |
|---|---|
| t27b itself | 19,752 lines of Rust (cli/t27b) |
| steward rules | specs/tri/t27b/steward.t27, 466 lines, 111 tests; reach Python through gen-c + ctypes; t27b passes them (lab: 111 tests, 1 invariant, 130 asserts) |
| reference pass, t27b pass with runtime checks | specs/queen/{dispatch,priority,task_analysis,views}.t27, specs/compiler/{c_array_repeat_wrapper,meta_compile,rust_constant_case}.t27 |
| reference pass, t27b blocked | specs/compiler/optimizer.t27 (StructDecl), specs/compiler/rust_value_parameters.t27 (StmtAssign), specs/compiler/zig_primitive_bindings.t27 (tuple, alias) |
| blocked in BOTH reference and t27b (t27c cannot carry the compiler's own specs) | specs/compiler/{lexer,parser,typechecker,pipeline,stdlib,diagnostics,linker,mod_structure}.t27, specs/compiler/core/t27core.t27 (self-host epic #5980), specs/queen/{lotus,brain_summaries}.t27 |
First reference errors from the lab JSON results[] (run 682564f):
| Spec | First reference error (zig on t27c's output) |
|---|---|
| compiler/core/t27core.t27 | spec.zig:1714:11: redeclaration of local variable 'n' |
| compiler/lexer.t27 | spec.zig:327:9: pointless discard of local constant |
| compiler/parser.t27 | spec.zig:104:4: expected '(', found 'error' |
| compiler/pipeline.t27 | spec.zig:420:4: expected '(', found 'error' |
| compiler/typechecker.t27 | spec.zig:80:53: expected ',' after initializer |
| compiler/stdlib.t27 | spec.zig:232:36: expected ',' after initializer |
| compiler/diagnostics.t27 | spec.zig:132:62: expected enum or tagged union, found 'comptime_int' |
| compiler/linker.t27 | spec.zig:48:36: expected field initializer |
| compiler/mod_structure.t27 | spec.zig:38:35: type '[]spec.ModuleEntry' does not support array initialization syntax |
| queen/lotus.t27 | spec.zig:427:5: use of undeclared identifier 'then' |
| queen/brain_summaries.t27 | spec.zig:82:24: use of undefined value here causes illegal behavior |
Over the wider "own specs" set (specs/{tri,queen,compiler,blog,automation,tools}/**) the same run has 300 files: reference pass + t27b pass 33, reference pass + t27b pass_vacuous 112, reference pass + t27b not passing 8, reference not passing 147 (blocked 140, fail 7).
Plan
tri t27b dogfood: reads the lab latest.json, prints for our own specs the reference verdict, the t27b verdict and the first blockers, grouped as t27b-only blocked (a t27b lane) and reference-blocked (a t27c issue). The path rule and the row classification live inspecs/tri/t27b/steward.t27, reaching Python through gen-c.tri t27b next: our own specs win ties in the lane order (steward.t27).- One t27c issue per root cause of the reference-blocked own specs, deduplicated against open issues.
Activity
gHashTag commentedon Oct 5, 2026
Dogfood table, lab run 682564f (master), classified by
gen/c/tri/t27b/dogfood.c(PR #6461). The reader that printed it is a one-off and is not committed (owner rule 2026-10-05 15:45Z: no hand-written non-t27 code in a PR), sotri t27b dogfooddoes not exist as a command yet.output (300 own specs)
gHashTag commentedon Oct 5, 2026
Reference-blocked own specs, one row per root cause. Source: lab run 682564f (master, 2026-10-05T15:01:42Z). The first Zig error was checked against the generated code on the t27c lab where noted.
union/erroris not escaped (fn union(,self.error()value ignored&[_]T{}gives'_' used as an identifierT?, keyword field.error =, array repeat[v; N]errorfield); tools/schema (str?) (5)use std;duplicates the automatic std importnot yet implementedstubsvar n ...; const n = ...in t27core testminimal_module_compiles, seen in the generated code)_ = start_pos; // dead after const-inliningfor a const still in use.{ value } ** COUNT32*2-1"\x1b[0m"emitted as"\\x1b[0m"(test fails at runtime)@intFromEnum!= undefinedstubs, Zig captures, undeclared names)Total: 35 + 112 = 147. No t27c PR was opened, per the owner's 2026-10-05 rule ("t27c fixes in Rust: file issues only").