[ojura]

AltStore Classic 2.3's Set up Remote AltServer… sends AltServer a PairingFileRequest. AltServer-Linux rejected it ("AltServer does not support this request"), so AltStore showed "unable to pair". This PR implements it, following AltServer for macOS.

Depends on jaakkopalvaila/AltServer-Windows#1, which adds the request handler to ClientConnection and the DevicePairingManager interface. The upstream_repo submodule here points at that PR's commit (b0d2846). If that PR is squashed or rebased on merge, this pointer needs updating to the merged commit.

What it does

• src/DevicePairingManager.cpp is a port of DevicePairingManager.swift:
  • finds the device's usbmuxd device_id by UDID
  • creates a usbmuxd provider
  • pairs with idevice's tunnel_pair_usb(); the device shows a Trust prompt for "AltServer on "
  • serializes the RPPairing file
• Differences from the Swift version:
  • It reads USBMUXD_SOCKET_ADDRESS once, by libusbmuxd's rules (UNIX:<path>, <host>:<port>, [<host>]:<port>, otherwise /var/run/usbmuxd), and uses that address for both the device lookup and the pairing, so netmuxd setups keep both on the same muxer. idevice's own reader rejects the UNIX: form.
  • It accepts only the device's USB entry. netmuxd in shim mode lists the same UDID once over USB and once over the network, and pairing must not go over the network entry.
  • The usbmuxd connection used for the lookup is closed before pairing starts.
  • It frees the usbmuxd address when creating the provider fails; the Swift version leaks it on that path.
  • A second request for a device that is already waiting on its Trust prompt fails at once with "AltServer is already pairing with this device". tunnel_pair_usb() can't be cancelled, so there is no timeout.
  • Pairing runs on its own thread with an 8 MiB stack. On a cpprestsdk pool thread, polling idevice's pairing future overflowed musl's 128 KiB default thread stack: a SIGSEGV in tunnel_pair_usb, with rsp about 130 KiB below the thread's stack top.
• libraries/idevice is jkcoxson/idevice v0.1.68, built with usbmuxd,tunnel_tcp_stack,ring.
  • Its static library also defines libplist-compatible plist_* functions, plus the Rust standard library. Both clash with the vendored libplist.
  • So makefiles/idevice-build/idevice.mak prelinks it with ld -r and keeps only the 18 functions in idevice-api.txt global (objcopy --keep-global-symbols).
  • The final binary has one global plist_new_dict, libplist's, and no unresolved symbols.
• Build:
  • IDEVICE_FFI_LIB points make at a prebuilt library. Without it, the Makefile runs cargo on every make and lets cargo decide whether anything changed. idevice.mak sets .DELETE_ON_ERROR.
  • Cross builds go to libraries/idevice/target/cross, which idevice's .gitignore covers.
  • The builder images have no Rust, and rustup ships musl-hosted compilers only for x86_64 and aarch64. So CI cross-compiles idevice on the runner with cargo-zigbuild (build-idevice-cross.sh, image pinned to 0.23.4), one Rust target per matrix entry.
• README-NG documents all of this as Fix 4 and updates the build instructions.

Testing

• AltStore Classic 2.3 on an iPhone 16 Pro running iOS 26.6.2:
  • Remote AltServer setup got the pairing file over USB. The same code, in the Ben-Diehlci port, also paired with netmuxd in shim mode listing the phone over both USB and the network.
  • With Prefer Remote AltServer on and AltServer stopped, AltStore installed an app on the device over LocalDevVPN.
• A PairingFileRequest for a UDID that isn't connected returns DeviceNotFound with the macOS failure reason and recovery suggestion, for USBMUXD_SOCKET_ADDRESS set to 127.0.0.1:27015 and to UNIX:<socket>.
• x86_64, i586, armv7 and aarch64 build and link statically from the first commit of this PR, using the builder images plus the cross-built library. The armv7 and aarch64 binaries were started under QEMU user emulation and found the device through usbmuxd.

🤖 Generated with Claude Code

[jaakkopalvaila]

Thanks for this, and sorry for the slow reply. The pairing support is useful, and the write-up made it easy to follow.

The ng branch moved a lot in October, so the PR now has conflicts:

• Apple's -80045 and -80046 changes. The fork now ports Handle 2FA codes that Apple sends on its own rileytestut/AltSign#59 for pushed 2FA codes, and it defaults to the anisette server's own client info, because only Mac anisette passes the 2FA step since about 21 October. See Fix 4 and Fix 5 in README-NG.md.
• Your AnisetteDataManager: parse X-Apple-I-Client-Time as UTC #1 and dnssd_loader: load libdns_sd.so.1 before libdns_sd.so #2 are merged. Thank you for those too.
• upstream_repo is now at jaakkopalvaila/AltServer-Windows@7358c74, which includes the 2FA port. Your companion PR AltServer: handle PairingFileRequest (Remote AltServer setup) AltServer-Windows#1 merges cleanly on top of it.

Could you rebase this onto the current ng, with upstream_repo pointing at your #1 rebased onto ng-fixes? Then I can merge both, and I will test the Remote AltServer setup with AltStore 2.3 on an iPhone on iOS 27.

If you don't have time for that, just tell me, and I will do the rebase myself and keep you as the author.

[ojura]

hey @jaakkopalvaila, thanks for having a look. yeah, under a bit of a crunch lately. if you have the time to rebase it, would be grateful. (based from my experience, Claude can probably transfer all useful work on this branch to the new base with little intervention. Then it was mostly running rounds of reviews until it saturated.)