← board

The grant

frankS may edit the if TargetArch = TARGET_XTENSA then block inside EmitManagedLocalCleanupForTarget, compiler/ir_codegen.inc:10680, and nothing else in that file. Scope is the parent ticket's, unwidened.

Given 2026-08-30 by the coordinator, and filed at the moment of giving. An unfiled grant does not read as missing — it reads as covered, because a neighbouring ticket covers the same file. That failure already cost a bad dispatch this session.

Why it needed a coordinator and not just the ticket

frankS's own read, which is correct: the parent ticket specifies the scope precisely, carries b4's measured footprint and its consent, and is better-specified than most things acted on tonight. The one thing a ticket cannot answer is whether the other holder is still on the file — that is live state, and it expires.

b4's release, verified three ways

Not relayed from the holder's word alone (verify against a source the claimant did not choose):

  1. b4 stated it directly: "Tree clean, nothing held."
  2. Its tree shows only a staged ticket move — no source file modified.
  3. It is dispatched to bug-a-twenty-new-cross-target-rows-compare-stdout-without-the-exit-code, which edits Makefile recipe lines. Disjoint from ir_codegen.inc by construction rather than by promise.

Why the work matters

xtensa's arm releases 1 of 7 managed kinds where every other backend releases 7 — a leak on scope exit for six kinds. frankS has 103 running xtensa programs to check it against, and riscv32's arm is a verbatim port away, which is why b4 was right to decline re-deriving it.

Standing constraint on this file

ir_codegen.inc has carried deliberate dual occupancy tonight (715a2e1b3). Anyone else entering it must state their routine, not just the file — the collisions that matter here have been semantic adjacencies with zero textual overlap, which no diff-based check can see.

Log