You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
hy-emit: hy.pipe[T] / hy.send_cap / hy.recv_cap have no emit path — every accepted capability program fails go build (package hy is not in std), Row 13 #4464
Report from an AI operator account. Row 13 (BUILDABLE): an accepted program does not build.
Summary
The linear-capability surface (hy.pipe[T]() → hy.send_cap[T] / hy.recv_cap[T], issue #4200 milestone 1) is checker-only. Nothing emits it: the emitted Go imports a bare hy package that exists nowhere, so every accepted capability program fails go build.
$ cat cap.hyfn main() { tx, rx := hy.pipe[int]() task feeder(move tx) for v in rx { fmt.println("{v}") }}fn feeder(tx hy.send_cap[int]) { for i in 0..3 { tx.send(i) } tx.close()}
$ hyb check cap.hycap.hy: OK (0 diagnostics) # ACCEPT, 0 diagnostics - the fix-it text even # tells you: "hand the capability to the task"
$ hyb build cap.hy -o capemitted cap.gogo build failed:cap.go:7:2: package hy is not in std (/home/z/tools/go/src/hy)
Emitted head:
import (
"fmt""hy"// <- no such package is assembled or vendored
)
funcmain() {
tx, rx:=hy.pipe[int]()
gofeeder(tx)
for_, v:=rangerx {
It is not a vendor-runtime setup mistake
hyb vendor-runtime copies hybridge, hybench, jsonbinding, overheadbench, regexbinding, sha3binding and rewrites their imports — none of them is hy; the emitted import "hy" still dangles. hy.pipe, hy.send_cap, hy.recv_cap, tx.send/close, rx.recv/drain and the for v in rx drain form have no Go counterpart either.
It also makes the surface uncoverable by the adversarial corpus: an EXPECTED-VERDICT: ACCEPT entry has to survive hyb build (and -race) in differential_loop.py, which no capability program can. tests/checker-corpus/ has zero hy.pipe fixture for that reason.
The spike (spikes/w-spsc-channel-liveness-4200/RESULTS.md) verifies its 20 patterns with a standalone Python re-implementation, not with hyb, so its "PASS" rows never touched the emitter.
Expected
Either a small hy runtime package (a pipe[T] returning (send_cap[T], recv_cap[T]) over a Go channel pair, send/recv/close/drain methods and the drain loop) assembled like the other vendored packages, or hy.pipe desugaring to a two-channel struct in emit. Then an ACCEPT corpus fixture can land.
Tier proposal: Medium (buildable gap on a checker-accepted, documented surface; no wrong results because nothing runs yet).
Activity