Tags: NativeScript/v8-buildscripts

fix(ios): export the public V8 API from the prebuilt archives

The matrix build sets v8_monolithic=false, which stopped triggering the
gni/v8.gni swap from Chromium's default -fvisibility=hidden to default
visibility. Every symbol in the shipped archives became private-external,
so NativeScript.framework exports no v8:: symbols and plugins that link
the public V8 API through it (e.g. @nativescript/canvas) fail the app
link with undefined v8::* symbols on runtime 9.x.

v8_expose_public_symbols=true defines BUILDING_V8_SHARED, giving the
V8_EXPORT surface (~2.7k symbols incl. cppgc and the v8::internal
helpers referenced by inline header code) default visibility while
keeping internals hidden and dead-strippable: ~+0.5 MB per slice versus
~+3.4 MB for restoring global default visibility.

Android is unaffected: it builds with v8_monolithic=true, which still
triggers the visibility swap.

fix(patch): skip stale queue entries when rooting pending finalizers

* IterateFinalizerHandlesAsRoots visited every queued node unconditionally,
  but an entry can go stale between queueing and the drain: callbacks running
  outside the pause may reset or re-arm a queued handle, clearing the pending
  bit -- and a reset frees the node, whose zapped location must never be
  visited as a root.
* Mirror the drain's re-check of the pending bit and compact stale entries
  out of the queue while iterating.

fix(ios): disable the cppgc caged heap

cppgc_enable_caged_heap defaults to true on arm64 and does not track
v8_enable_pointer_compression, so it stayed on despite that being off.
Enabling it forces cppgc pointer compression, whose reservation asks for
twice the cage to satisfy alignment, and V8::Initialize() calls
cppgc::InitializeProcess unconditionally -- so every iOS app reserved
8 GiB of 4 GiB-aligned address space at startup without using cppgc.

iOS budgets virtual address space per process by device RAM. On smaller
devices no such hole exists, the four reservation attempts all fail, and
Oilpan's fatal OOM handler aborts before any JS runs: an iPad Air crashed
on launch where an iPad Pro was fine.

Costs cppgc pointer compression, its young generation and a write-barrier
fast path. Nothing in the cppgc API changes and CppHeap is unaffected.
Android keeps the cage; 64-bit has the address space and 32-bit cannot
enable it anyway.

fix(patch): allow JS execution in finalizer callbacks

V8 14.9 holds a DisallowJavascriptExecution across the whole collection,
the epilogue included, and turns entering JS from a GC callback into a
GRACEFUL_FATAL. 10.3 ran the second-pass callbacks under an explicit
AllowJavascriptExecution, which the iOS runtime relies on: disposing an
ObjC wrapper runs -dealloc, and any JS-backed override that teardown
reaches re-enters JS through ArgConverter::MethodCallback.

The scope is lifted around the finalizer drain alone, so the second-pass
callbacks below it still satisfy their own assertion that JS is
disallowed. Verified by compiling global-handles.o for arm64-simulator.

V8 14.9.207.39, build 1

First release of the unified matrix: four Android ABIs and five Apple
variants, built in CI from a pinned V8 revision with the resurrecting
finalizer patch applied.

Consumed by NativeScript/android and NativeScript/ns-v8ios-runtime via
their download_v8.sh.