- Update FRIDA_VERSION files to 17.13.0
- Add new API bindings: ApiResolver, Cloak, CodeAllocator, CodeSegment, Exceptor, Memory, Query, Registry, SymbolUtil, TLS
- Update existing bindings for compatibility
- Update examples for new API patterns

- Upgrade FRIDA_VERSION from 17.11.0 to 17.12.0
- Add new API bindings: ApiResolver, Cloak, CodeAllocator, CodeSegment, Exceptor, Memory, Query, Registry, SymbolUtil, TLS
- Update existing bindings for compatibility
- Update examples for new API patterns
- Note: 17.13.0 upgrade pending release of devkit binaries

…ust into kkuehl/frida-rust-updates

…ndings

Quality/soundness fixes (verified against frida-gum C source):
- module_map: fix use-after-free in new_with_filter (filter was a dangling
  temporary Box with no destroy-notify, but Frida retains and re-invokes it on
  every update); now owns the closure by value + GDestroyNotify.
- range_details: add missing PageProtection::WriteExecute variant (value 6) and
  stop unwrap()-panicking on unexpected protection bits in an FFI callback.
- process: free the transfer-full string returned by g_get_current_dir.
- module/process: fix &T-as-*mut aliasing (UB) in enumerate_* callbacks.
- exceptor: require Send on the handler closure; document no auto-remove on drop.
- memory_range: MatchPattern::from_string returns None instead of panicking on
  interior-NUL input.
Several agent-flagged "bugs" were verified FALSE POSITIVES and left unchanged
(registry obtain() is transfer-none; gum_memory_read uses g_malloc; GumCodeSlice
refcount is atomic).

Frida + toolchain:
- Bump Frida 17.12.0 -> 17.13.0 (all FRIDA_VERSION files).
- Migrate workspace to edition 2024 (cargo fix --edition + manual unsafe-block
  fixes in feature-gated files the migration did not compile: invocation_listener,
  stalker/observer, stalker/event_sink, backtracer).

New API bindings (Frida 17.10-17.13):
- control_flow_graph.rs: GumControlFlowGraph (new/for_function, dominates,
  enumerate_dominating_sites, block queries, find_instruction_containing).
- Process::find_function_range (works on stripped binaries).
- Interceptor::flush_function / flush_listener.
- Interceptor::attach_with_options + AttachOptions builder.
- unwind_broker.rs: GumUnwindBroker obtain + provider/translator registration.
- frida core: Variant::StringList decodes GVariant "as" so the new argv process
  parameter surfaces via Process::get_parameters().

Wrap remaining pre-existing C API gaps:
- Memory::mprotect / try_mprotect / clear_cache / ensure_code_readable /
  try_alloc_n_pages_near / patch_code_pages.
- DebugSymbol::load_symbols.
- InvocationContext listener data getters (function / thread / invocation).
- Exceptor::exception_details_to_string.
- CodeAllocator::alloc_deflector (+ CodeDeflector type).

Verified (edition 2024, Frida 17.13.0): cargo build (all gum features), clippy
-D warnings (gum all-features + frida + frida-sys), cargo fmt --check, and
cargo test -p frida-gum (10 passed) all clean.

The instruction-writer wrappers for Gum C functions that are declared `void`
were fabricating `-> bool { ...; true }`, giving callers a meaningless success
signal. Change these inherent methods to return `()` to match the C ABI:

- x86_64: put_leave, put_ret, put_ret_imm, put_jmp_short_label,
  put_jmp_near_label, put_mov_reg_address, put_mov_reg_ptr_u32,
  put_mov_reg_ptr_reg, put_mov_reg_reg_ptr, put_push_u32, put_push_imm_ptr,
  put_nop, put_pushfx, put_popfx, put_pushax, put_popax.
- aarch64: put_call_address_with_arguments.

The InstructionWriter trait methods (put_bytes/put_label/put_branch_address/
put_nop/flush) are left returning bool: arm/arm64's underlying C functions
return gboolean, so the shared trait signature must stay. The x86 put_bytes impl
keeps its `true` fabrication only to satisfy that trait.

No callers consumed these return values. Verified: clippy -D warnings (gum
all-features), cargo fmt --check, cargo test -p frida-gum (10 passed).

Edition 2024 + platform differences the Windows/x86_64 build never exercised:

- unsafe extern: aarch64/relocator.rs (7 blocks) and process.rs linux/freebsd
  block were still plain `extern "C"`.
- Enum discriminant signedness: bindgen types C enums as i32 on MSVC but u32 on
  clang. Use `as _` so the cast is correct on both (and not flagged as a
  same-type cast on Windows): transformer MemoryAccess, process
  TeardownRequirement, exceptor ExceptorMode + ExceptionType, aarch64 IndexMode,
  and the jcc_short/jcc_near condition args in the x86 writer.
- thread WatchConditions bitflags backed by gum_sys::GumWatchConditions so
  bits() matches the FFI parameter on every platform.
- gchar pointer signedness: CString/CStr pointers passed to gum_* APIs typed as
  *const gchar (i8 on clang, differs from cstr_core's c_char) now use .cast()
  in api_resolver, symbol_util, debug_symbol.
- GLib symbol naming: g_free/g_error_free/g_array_free/g_ptr_array_* are
  _frida_-prefixed on the Linux/FreeBSD/iOS static devkits but bare g_* on
  Windows/macOS. Added a central cfg-gated glib_compat module (mirroring the
  existing _frida_g_get_*_dir handling) and routed all call sites through it.

Verified on Windows: clippy -D warnings (gum all-features) clean.

FRIDA_VERSION: bump 17.9.5 -> 17.14.0 across all FRIDA_VERSION files.

Edition 2024 (workspace Cargo.toml): `unsafe extern "C"` required for all
extern blocks; `unsafe_op_in_unsafe_fn` now deny-by-default requiring explicit
`unsafe {}` blocks inside unsafe fns; `ref` in let-else patterns removed.
Applied via `cargo fix --edition` where possible, remaining blockers fixed
manually first.

API changes for 17.13.0/17.14.0:
- gum_interceptor_replace/replace_fast: new options param added; pass null.
- gum_interceptor_attach: options param replaces separate listener_data arg.
- gum_stalker_activate_experimental_unwind_support: removed; drop wrapper.
- frida-build: fix redundant-reference lints in format!/println! macros.
- backtracer: fix double-reference in fuzzy_with_context.

Cross-platform (Linux/iOS devkits prefix GLib symbols with _frida_):
- variant.rs `ref` patterns updated for edition 2024 match ergonomics.
- process.rs Linux/FreeBSD extern block marked unsafe extern.
- aarch64/relocator.rs + x86_64/relocator.rs extern blocks marked unsafe.

- Add unsafe keyword to all extern "C" blocks in aarch64 relocator
- Update FRIDA_VERSION from 17.14.0 to 17.14.1 in both frida-sys and frida-gum-sys
- Fixes compilation errors with Rust edition 2024 requirement for unsafe extern blocks

Only uses reqwest::blocking::get with blocking + rustls-tls features,
all unchanged in 0.13. Lets downstream workspaces unify on reqwest 0.13
instead of compiling both 0.12 and 0.13.

Only uses reqwest::blocking::get with blocking + rustls-tls features,
all unchanged in 0.13. Lets downstream workspaces unify on reqwest 0.13
instead of compiling both 0.12 and 0.13.

…tls')

reqwest 0.13 renamed the rustls-tls feature to rustls; the old name no
longer resolves. Plain 'rustls' bundles a crypto provider (unlike
'rustls-no-provider').

…tls')

reqwest 0.13 renamed the rustls-tls feature to rustls; the old name no
longer resolves. Plain 'rustls' bundles a crypto provider (unlike
'rustls-no-provider').

Rust 2024 edition requires unsafe operations inside `unsafe fn` bodies
to be wrapped in explicit `unsafe {}` blocks. Add inner unsafe blocks to
create_cb, js_msg, and load_cb to satisfy unsafe_op_in_unsafe_fn lint.

…ition-2024

Keep frida-rust-updates Rust source (has more API bindings and CI fixes).
Take FRIDA_VERSION 17.15.0 from frida-17.13-edition-2024 branch.

New frida-gum/src/elf_module.rs wraps gum_elf_module_* APIs added in
17.15.0: ElfModule::from_file/from_memory constructors, property accessors
(pointer_size, byte_order, os_abi_version, mapped_size, base/preferred
address, entrypoint, interpreter, source_path), and
enumerate_dynamic_entries/dynamic_entries for iterating ELF .dynamic section.
Linux/Android/FreeBSD only (cfg-guarded against Windows and macOS/iOS).

Document all changes, new features, and remaining gaps:
- Breaking changes (removed page allocation APIs)
- New features (Darwin modules, Kernel APIs)
- API coverage analysis by category
- Notable missing APIs with priorities
- Comparison with other language bindings
- Recommendations for reaching parity

Current status: ~50% coverage, target ~80% for parity with other bindings.

Add new instructions from Frida 17.16.1:

x86_64 Writer (AVX-512):
- kmovq (k-register mask operations)
- vextracti64x4 (extract 256-bit from 512-bit ZMM)
- vinserti64x4 (insert 256-bit into 512-bit ZMM)
- vmovdqu64 (unaligned 512-bit move)

ARM64 Writer:
- movk (move with keep for building 64-bit constants)
- pacia (Pointer Authentication Code for ARMv8.3-A)

Note: Memory::allocate/allocate_near already existed, no changes needed.

- Added AVX-512 x86_64 instructions (+6)
- Added ARM64 PAC/MOVK instructions (+2)
- Memory allocate APIs already existed
- Updated coverage: ~52% overall
- x86_64 Writer: 88% → 92%
- ARM64 Writer: 30% → 32%
- Memory: 85% → 100%

Add 25 new ARM64 instructions:
- Branch/control: ret, br, blr, b_imm, b_label, bl_label, cbz, cbnz, tbz, tbnz
- PAC variants: br_reg_no_auth, blr_reg_no_auth
- Stack: push, pop, push_all_x_registers, pop_all_x_registers, push_all_q_registers, pop_all_q_registers
- Load/Store: ldr (reg/u32/u64), str, ldp, stp
- Arithmetic: mov, add, sub, and
- Condition flags: mov_reg_nzcv, mov_nzcv_reg

ARM64 Writer: 23 → 48 instructions (32% → 68% coverage)

The types (GumInterceptorScenario, GumRelocationPolicy, GumAttachOptions, etc.)
DO exist in Frida 17.16.1 devkit headers. Previous CI failures were due to
missing devkit download in docs job (now fixed).

build.rs expects headers in frida-gum-sys/include/ when DOCS_RS=1.
Extract devkit tarball to that location.

The committed header is used for docs builds and needs to be kept in sync
with FRIDA_VERSION. Updated from old version to 17.16.1 to include new types
like GumInterceptorScenario, GumRelocationPolicy, GumAttachOptions, etc.

cargo fmt fixed formatting issues in darwin_module.rs and kernel.rs

Removed 14 duplicate function definitions in Aarch64InstructionWriter:
- put_br_reg, put_b_label, put_push_reg_reg, put_pop_reg_reg
- put_ldr_reg_address, put_ldr_reg_u64, put_ldr_reg_reg_offset
- put_str_reg_reg_offset, put_ldp_reg_reg_reg_offset
- put_stp_reg_reg_reg_offset, put_mov_reg_reg
- put_add_reg_reg_imm, put_add_reg_reg_reg, put_sub_reg_reg_imm

These functions were defined twice - once in the existing codebase
(lines 88-256) and again in my additions for 17.16.1 (lines 353+).
Kept the original definitions and removed duplicates from my additions.

New functions like put_br_reg_no_auth, put_blr_reg, put_cbz_reg_imm,
etc. are retained as they don't have duplicates.

x86_64 writer:
- Use .try_into().unwrap() for GumArgType and GumCallingConvention enums
  (they're i32 on Windows, u32 on macOS - cast needed for cross-platform)

darwin_module (macOS-specific):
- Cast protection field from i32 to u32
- Use .as_ptr() for array fields (segment_name, section_name)
- Access nested offset field via __bindgen_anon_1.__bindgen_anon_1.offset
- Cast library_ordinal (i32 → i16) and symbol_flags (u8 → i8)
- Add new parameters to gum_darwin_module_new_from_file/memory:
  GUM_CPU_INVALID, GUM_PTRAUTH_INVALID, 0 (API changed in 17.16.1)
- Fix segments field access - it's now *mut GArray, not embedded GArray

aarch64 writer:
- Fix put_movk_reg_imm parameter order: (reg, shift: u16, imm: u32)
  (was incorrectly (reg, imm: u32, shift: u8))

kernel.rs:
- Cast CString pointer to *const c_char for gum_kernel_enumerate_module_ranges
  (cstr_core produces *const u8, but iOS expects *const i8)

- Remove underscore prefix from GumCpuType and GumPtrauthSupport constants
  (bindgen generates GumCpuType_GUM_CPU_INVALID, not _GumCpuType_GUM_CPU_INVALID)
- Apply cargo fmt to x86_64 writer (break long lines)

Instead of .try_into().unwrap(), use 'as u32' cast which works on both:
- Windows: enums are i32, cast to u32
- Linux/macOS: enums are u32, no-op cast

This is the idiomatic Rust solution for platform-dependent enum types.

Windows: enums are i32, need cast to u32
Linux/macOS: enums are u32, no cast needed

Applied to all functions using:
- _GumArgType_GUM_ARG_REGISTER
- _GumArgType_GUM_ARG_ADDRESS
- _GumCallingConvention_GUM_CALL_CAPI

This eliminates clippy warnings on Linux while maintaining correctness on Windows.

- Import InstructionWriter trait (needed for ::new())
- Import Aarch64Register from correct path
- Combine imports on one line

On 32-bit platforms (i686), gsize is u32, not u64.
Changed all size/offset casts from 'as u64' to 'as gum_sys::gsize':

- memory.rs: n_read variable, gum_memory_write len, gum_memory_find_pointers values
- code_allocator.rs: slice_size, alignment, max_distance
- code_segment.rs: size, source_offset, source_size, max_distance
- thread.rs: watchpoint size

This fixes compilation on i686-unknown-linux-gnu target.