#35083 shipped only in 1.4.0, but affected users run embedded 1.3.14 — backport to 1.3.x, or document that compiled binaries must be rebuilt #40241

Summary

#35083 ("crash_handler(windows): let foreign first-chance AVs reach SEH via JSC unwind info") fixed the VEH-vs-SEH crash class on Windows — nProtect GameGuard (npggNT64.des), BeyondTrust PGHook.dll, AV/EDR hooks. It shipped in 1.4.0 and nothing earlier.

The problem: a large share of the users who hit this bug run Bun as an embedded runtime inside a single-file compiled executable, not as a standalone bun. For them bun upgrade is a no-op against the crash, and there is currently no version of Bun they can reach that contains the fix.

Evidence (verified today)

Fact	Value
bun-v1.3.14 released	2026-05-13
#35083 merged (79ee451)	2026-07-23
bun-v1.4.0 released	2026-08-20
79ee451 vs bun-v1.4.0	behind=0 — present in 1.4.0, absent from every earlier tag
A 1.3.15 backport	does not exist; 1.3.14 → 1.4.0 directly

Concrete downstream, checked on a Windows 11 machine:

• OpenCode 1.18.21 — the latest release, published 2026-08-21, i.e. one day after bun-v1.4.0 — ships Bun 1.3.14.
  • anomalyco/opencode main pins "packageManager": "[email protected]"
  • Byte-scanning the shipped 179 MB opencode.exe yields only Bun v1.3.14 / bun v1.3.14; no 1.4 string is present.

So a user crashing today has no self-service remedy: their standalone bun can be 1.4.0 while the runtime actually hosting the vectored exception handler is still 1.3.14, baked into a binary they don't build.

Ask

Either would resolve it:

1. A 1.3.15 carrying just the crash_handler(windows): let foreign first-chance AVs reach SEH via JSC unwind info #35083 handler change (plus the crash_handler(windows): walk the fault CONTEXT via RtlVirtualUnwind #35074 unwind work it depends on), so downstreams pinned to 1.3.x can pick it up without taking a minor bump; or

2. If backports are out of scope — an explicit note on the closed threads (Segfault at 0x3C — nProtect GameGuard (npggNT64.des) injection kills any Bun process on Windows 11 #34055, Segfault on Windows: nProtect GameGuard module (npggNT64.des) appears in crash stack — deterministic reproduction #35764, nprotect gameguard crashes bun on windows #21930, Bun segmentation fault when opening Helldivers 2 #27051, Segmentation fault at startup when EasyAntiCheat (Helldivers 2) driver is running #32444, Crash with a segmentation fault at address 0x3C #29497) and in the 1.4.0 release notes to the effect of:

   If Bun is embedded in a compiled binary (bun build --compile, or any tool that ships its own runtime), that tool must be rebuilt against Bun ≥ 1.4.0. Upgrading your standalone bun install does not fix this.

   Those closed issues are where affected users land from search, and none of them currently draw that distinction — so the reasonable conclusion a reader reaches ("it's fixed, just upgrade Bun") is wrong for exactly the population that reported it.

Secondary: pre-release "1.4.0" is making open reports ambiguous

#37304 is open, reports address 0x000000E8 — the precise CRYPTSP cmp [rcx+0E8h], 11111111h signature described in #35083's own writeup — and states Bun v1.4.0. But it was filed 2026-08-10, ten days before bun-v1.4.0 stable, so it is a pre-release build and it is unclear whether final 1.4.0 resolves it. #32586 has the same shape (filed 2026-06-22, also labelled "Bun v1.4.0").

It may be worth asking both reporters for bun --revision and a re-test on stable, since right now those threads can't distinguish "fix didn't work" from "fix wasn't in that build" — and #35083 does flag one known remaining gap (LLInt frames not covered by the static .pdata lookup, noted as a follow-up).

Version

Bun v1.3.14 (standalone) and Bun v1.3.14 (embedded in OpenCode 1.18.21) on Windows 11 26100, x64.