Issues

• ARM Thumb-2: a nested br_if / if / br block returns 0 on its i32.div_s path

  pulseengine/synth#1712 · Filter by author avrabe opened on Nov 13, 2026

• ARM Thumb-2: i64 select on a fused i32 compare returns wrong halves

  pulseengine/synth#1711 · Filter by author avrabe opened on Nov 13, 2026

• ARM Thumb-2: an explicit return of a computed i64 leaves the argument's high word in r1

  pulseengine/synth#1710 · Filter by author avrabe opened on Nov 13, 2026

• synth disasm decodes RV32 and AArch64 ELFs as ARM A32

  pulseengine/synth#1704 · Filter by author avrabe opened on Nov 11, 2026

• RV64 (RV64IMAC/RV64GC) backend: design and sequencing after RV32 reach

  pulseengine/synth#1703 · Filter by author avrabe opened on Nov 11, 2026

• Meld dispatch fallback passes no wasm arguments to imported functions (only the index in r0)

  pulseengine/synth#1699 · Filter by author avrabe opened on Nov 11, 2026

• ARM object: an import whose field name equals an own export binds the call to that export (no undefined symbol)

  pulseengine/synth#1698 · Filter by author avrabe opened on Nov 11, 2026

• ARM optimized path: an import call clobbers caller-saved registers, so a value held in r0 across it is silently wrong

  pulseengine/synth#1696 · Filter by author avrabe opened on Nov 10, 2026

• Cortex-M7 object of a wasm-opt -O2 module computes different floats than the module (synth 0.90.0)

  pulseengine/synth#1691 · Filter by author cpetig opened on Nov 9, 2026
  1

• SILENT MISCOMPILE: under --native-pointer-abi an i64.const naming a memory-0 static is never relocated, so i64 pointer comparisons miss

  pulseengine/synth#1684 · Filter by author avrabe opened on Nov 8, 2026

• synth-backend mpu.rs encodes every PMSAv7 region at twice its size (RASR SIZE field off by one)

  pulseengine/synth#1677 · Filter by author avrabe opened on Nov 8, 2026

• SILENT MISCOMPILE: an imported global shifts every defined global's slot (and the stack-pointer adjustment) on ARM

  pulseengine/synth#1676 · Filter by author avrabe opened on Nov 8, 2026

• multi-memory: f32/f64 load and store on a non-default memory decline (the deferred float sibling of #1622)

  pulseengine/synth#1673 · Filter by author avrabe opened on Nov 7, 2026

• SILENT MISCOMPILE: under --native-pointer-abi, a memory-0 address stored in a data segment is dereferenced as a raw absolute address

  pulseengine/synth#1672 · Filter by author avrabe opened on Nov 7, 2026

• WRONG BEHAVIOUR: --native-pointer-abi with --safety-bounds software guards memory 0 against a native pointer, so in-bounds fill/copy trap

  pulseengine/synth#1661 · Filter by author avrabe opened on Nov 6, 2026

• The #1069 VFP-pressure oracle runs only on a zeroed stack and zeroed registers, so uninitialised-read defects (#1594) are invisible to it

  pulseengine/synth#1613 · Filter by author avrabe opened on Oct 28, 2026

• Spec-suite census is blind to float codegen on ARM: --cortex-m is cortex-m3 (no FPU), 233 float functions never compiled

  pulseengine/synth#1609 · Filter by author avrabe opened on Oct 28, 2026

• SILENT MISCOMPILE: --native-pointer-abi access straddling the zero reservation into a data segment reads the reservation

  pulseengine/synth#1604 · Filter by author avrabe opened on Oct 28, 2026

• SILENT MISCOMPILE: --native-pointer-abi bakes a raw linear-memory offset for constant addresses — 81 of 162 measured shapes return wrong values

  bug
  pulseengine/synth#1568 · Filter by author avrabe opened on Oct 21, 2026
  1

• ubuntu-26.04 rollout: two oracle jobs break on the new image's apt toolchain — pinned to 24.04, port them

  pulseengine/synth#1567 · Filter by author avrabe opened on Oct 21, 2026

• authorised_close_set is keyed by issue number and filled with setdefault, so two artifacts on one issue collapse silently

  pulseengine/synth#1517 · Filter by author avrabe opened on Oct 9, 2026

• A figure pinned to a pre-squash lane head is uncheckable: the commit is not an ancestor of main

  pulseengine/synth#1516 · Filter by author avrabe opened on Oct 9, 2026

• verdict_prose_check --self-test advertises potency against real history and stays green with the capability it tests removed

  pulseengine/synth#1515 · Filter by author avrabe opened on Oct 9, 2026

• verdict_prose_check is an allow-list: four verdict words escape the population silently, NOT_A_VERDICT is inert, and the disposition branch is dead on the live tree

  pulseengine/synth#1476 · Filter by author avrabe opened on Oct 6, 2026

• @pulseengine/synth on npm is TEN releases behind: Release NPM has failed its auth preflight on every run since v0.69, and no release check asserts npm publication

  pulseengine/synth#1460 · Filter by author avrabe opened on Oct 1, 2026
  1