GH-141362: Make get_externals handle fetching platform-specific release artifacts (#142405)

Co-authored-by: Steve Dower <steve.dower@microsoft.com>
Co-authored-by: Chris Eibl <138194463+chris-eibl@users.noreply.github.com>

PCbuild/get_external.py

@@ -3,6 +3,7 @@
 import argparse
 import os
 import pathlib
+import platform
 import sys
 import tarfile
 import time
@@ -44,20 +45,38 @@ def fetch_zip(commit_hash, zip_dir, *, org='python', binary=False, verbose):
 
 
 def fetch_release(tag, tarball_dir, *, org='python', verbose=False):
-    url = f'https://github.com/{org}/cpython-bin-deps/releases/download/{tag}/{tag}.tar.xz'
+    arch = os.environ.get('PreferredToolArchitecture')
+    if not arch:
+        machine = platform.machine()
+        arch = 'ARM64' if machine == 'ARM64' else 'AMD64'
+    elif arch.lower() in ('x86', 'x64'):
+        arch = 'AMD64'
     reporthook = None
     if verbose:
         reporthook = print
     tarball_dir.mkdir(parents=True, exist_ok=True)
-    output_path = tarball_dir / f'{tag}.tar.xz'
-    retrieve_with_retries(url, output_path, reporthook)
+
+    arch_filename = f'{tag}-{arch}.tar.xz'
+    arch_url = f'https://github.com/{org}/cpython-bin-deps/releases/download/{tag}/{arch_filename}'
+    try:
+        output_path = tarball_dir / arch_filename
+        retrieve_with_retries(arch_url, output_path, reporthook)
+        return output_path
+    except OSError:
+        if verbose:
+            print(f'{arch_filename} not found, trying generic binary...')
+
+    generic_filename = f'{tag}.tar.xz'
+    generic_url = f'https://github.com/{org}/cpython-bin-deps/releases/download/{tag}/{generic_filename}'
+    output_path = tarball_dir / generic_filename
+    retrieve_with_retries(generic_url, output_path, reporthook)
     return output_path
 
 
 def extract_tarball(externals_dir, tarball_path, tag):
     output_path = externals_dir / tag
     with tarfile.open(tarball_path) as tf:
-        tf.extractall(os.fspath(externals_dir))
+        tf.extractall(os.fspath(externals_dir), filter='data')
     return output_path
 
 

Tools/jit/README.md

@@ -43,6 +43,15 @@ ### Windows
 
 LLVM is downloaded automatically (along with other external binary dependencies) by `PCbuild\build.bat`.
 
+By default, the architecture of the LLVM tools is auto-detected based on the host machine. To override this, set the `PreferredToolArchitecture` environment variable before building:
+
+```sh
+set PreferredToolArchitecture=AMD64
+PCbuild\build.bat --experimental-jit
+```
+
+Valid values are`x64`, `x86` and `ARM64`.
+
 Otherwise, you can install LLVM 21 [by searching for it on LLVM's GitHub releases page](https://github.com/llvm/llvm-project/releases?q=21), clicking on "Assets", downloading the appropriate Windows installer for your platform (likely the file ending with `-win64.exe`), and running it. **When installing, be sure to select the option labeled "Add LLVM to the system PATH".**
 
 Alternatively, you can use [chocolatey](https://chocolatey.org):

Tools/jit/jit_infra.md

@@ -8,21 +8,28 @@ ## Updating LLVM
 
 To update the LLVM release artifact for Windows builds, follow these steps:
 1. Go to the [LLVM releases page](https://github.com/llvm/llvm-project/releases).
-1. Download x86_64 Windows artifact for the desired LLVM version (e.g. `clang+llvm-21.1.4-x86_64-pc-windows-msvc.tar.xz`).
-1. Extract and repackage the tarball with the correct directory structure. For example:
+1. Download Windows artifacts for the desired LLVM version (e.g. `clang+llvm-21.1.4-x86_64-pc-windows-msvc.tar.xz` and `clang+llvm-21.1.4-aarch64-pc-windows-msvc.tar.xz`).
+1. Extract and repackage each tarball with the correct directory structure. For example:
    ```bash
+   # For x86_64 (AMD64)
    tar -xf clang+llvm-21.1.4-x86_64-pc-windows-msvc.tar.xz
    mv clang+llvm-21.1.4-x86_64-pc-windows-msvc llvm-21.1.4.0
-   tar -cf - llvm-21.1.4.0 | pv | xz > llvm-21.1.4.0.tar.xz  
+   tar -cf - llvm-21.1.4.0 | pv | xz > llvm-21.1.4.0-x64.tar.xz
+   rm -rf llvm-21.1.4.0
+
+   # For ARM64
+   tar -xf clang+llvm-21.1.4-aarch64-pc-windows-msvc.tar.xz
+   mv clang+llvm-21.1.4-aarch64-pc-windows-msvc llvm-21.1.4.0
+   tar -cf - llvm-21.1.4.0 | pv | xz > llvm-21.1.4.0-ARM64.tar.xz
    ```
-    The tarball must contain a top-level directory named `llvm-{version}.0/`.
+   Each tarball must contain a top-level directory named `llvm-{version}.0/`.
 1. Go to [cpython-bin-deps](https://github.com/python/cpython-bin-deps).
-1. Create a new release with the updated LLVM artifact.
+1. Create a new release with the LLVM artifacts.
     - Create a new tag to match the LLVM version (e.g. `llvm-21.1.4.0`).
-    - Specify the release title (e.g. `LLVM 21.1.4 for x86_64 Windows`).
-    - Upload the asset (you can leave all other fields the same).
+    - Specify the release title (e.g. `LLVM 21.1.4`).
+    - Upload both platform-specific assets to the same release.
 
 ### Other notes
 - You must make sure that the name of the artifact matches exactly what is expected in `Tools/jit/_llvm.py` and `PCbuild/get_externals.py`.
-- We don't need multiple release artifacts for each architecture because LLVM can cross-compile for different architectures on Windows; x86_64 is sufficient.
+- The artifact filename must include the architecture suffix (e.g. `llvm-21.1.4.0-x64.tar.xz`, `llvm-21.1.4.0-ARM64.tar.xz`).
 - You must have permissions to create releases in the `cpython-bin-deps` repository. If you don't have permissions, you should contact one of the organization admins.