README: -mm=own and Android; release zip Android layout

[ASDAlexander77]

README

• What's new and Contents: entries for -mm=own and the Android targets.
• Compile for Android (new): covers
  • the targets (aarch64-linux-android29, x86_64-linux-android29), the NDK option, and why the API level must be in the triple;
  • which release-zip library goes with which option;
  • an arm64 --emit=exe example and a note on --emit=dll;
  • running the result with adb. It needs chmod +x, because a file pushed from Windows isn't executable and the run fails with exit 126.
• Memory models: an -mm=own row in the table, and a subsection covering
  • moves, borrows and frees at points known at compile time;
  • a two-owners example with the real diagnostic;
  • Shared<T> with .value and Shared.count;
  • --no-default-lib, since there is no own default library yet.
• Build for Android (new): build_gc_release_android.bat, build_tslang_runtime_release_android.bat, and the DefaultLib repo's build_android.bat.

Release zip: Android layout

Get-ChildItem -Path ... .\android_stage\android | Compress-Archive packs that folder's children. So the Windows zip has arm64-v8a\lib\ and x86_64\lib\ at its root, not the android\<abi>\lib\ that the workflow comment and the DefaultLib README document. The fix is to pass .\android_stage, the same way gcdll_stage and resourcedir_stage are passed.

I checked this by running the same pipeline on a small staged tree:

Before	After
arm64-v8a/lib/libgc.a	android/arm64-v8a/lib/libgc.a

v0.0-pre-alpha91's zip was built before this fix, so its Android libraries are at <zip>\<abi>\lib.

Test plan

• Every README example compiled with the current Release tslang.exe.
• On the API 29 x86_64 emulator, hello world printed correctly under gc, rc and own (adb push, chmod +x, run), and the Shared<T> example printed 1 1 2.
• An arm64 gc executable linked to an AArch64 ELF. Not run: no arm64 device.
• The packaging fix was checked on a staged tree, as shown above. Not checked on a real release run yet.

While writing the -mm=own section I found that reassigning a moved-from let is rejected, and filed it as a separate issue.

🤖 Generated with Claude Code