Release 1.3.1: fix TI ARM CGT big-endian detection in the C library, render README on PyPI

.github/workflows/ci.yml

@@ -69,3 +69,34 @@ jobs:
     - name: Run tests
       run: |
         make test --no-print-directory -s
+
+  big-endian:
+
+    runs-on: ubuntu-24.04
+
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Set up QEMU
+      uses: docker/setup-qemu-action@v4
+      with:
+        platforms: s390x
+
+    - name: Run endian tests on s390x
+      run: |
+        docker run --rm --platform linux/s390x \
+          -v "${{ github.workspace }}:/work" \
+          -w /work \
+          python:3.13-bookworm \
+          bash -euxo pipefail -c '
+            apt-get update
+            apt-get install -y --no-install-recommends gcc make
+            python -m pip install -r compiler/requirements.txt -r tests/requirements_tests.txt
+            printf "%s\n" \
+              "#!/usr/bin/env sh" \
+              "PYTHONPATH=/work/compiler exec python -m bitproto._main \"\$@\"" \
+              >/usr/local/bin/bitproto
+            chmod +x /usr/local/bin/bitproto
+            PYTHONPATH=/work/compiler:/work/lib/py python -c "import sys; assert sys.byteorder == \"big\", sys.byteorder"
+            PYTHONPATH=/work/compiler:/work/lib/py python -m pytest tests/test_encoding/test_endian.py -v
+          '

README.md

@@ -31,7 +31,7 @@ of 4 bytes after encoding.
 
 This image shows the layout of data fields in the encoded bytes buffer:
 
-![](docs/_static/images/data-encoding-sample.png)
+![](https://raw.githubusercontent.com/hit9/bitproto/master/docs/_static/images/data-encoding-sample.png)
 
 ## Code Example
 

changes.rst

@@ -1,5 +1,21 @@
 .. currentmodule:: bitproto
 
+Version 1.3.1
+-------------
+
+.. _version-1.3.1:
+
+- Bugfix: Broaden ``BP_BIG_ENDIAN`` auto-detection beyond GCC/Clang's
+  ``__BYTE_ORDER__`` to also cover ACLE-compatible Arm toolchains such as TI
+  Arm Clang (``__ARM_BIG_ENDIAN``), legacy TI ARM CGT (``__big_endian__``),
+  other toolchains that define ``__BIG_ENDIAN__``, and IAR
+  (``__LITTLE_ENDIAN__ == 0``), in both the optimization-mode generated code and
+  the standard-mode C library. Previously these big-endian targets fell through
+  to the little-endian path and produced incorrect results for multi-byte
+  fields. PR #84.
+- Packaging: Show the project README as the long description on the PyPI project
+  page for the ``bitproto`` package.
+
 Version 1.3.0
 -------------
 

compiler/README.md

@@ -0,0 +1 @@
+../README.md

compiler/bitproto/__init__.py

@@ -8,5 +8,5 @@
 
 """
 
-__version__ = "1.3.0"
+__version__ = "1.3.1"
 __description__ = "bit level data interchange format."

compiler/bitproto/renderer/impls/c/renderer_c.py

@@ -412,12 +412,18 @@ def render(self) -> None:
             # Auto-detect a big-endian host so the default just works without
             # the user defining BP_BIG_ENDIAN; still overridable via -D.
             # __BYTE_ORDER__ / __ORDER_BIG_ENDIAN__ covers GCC/Clang.
-            # __BIG_ENDIAN__ covers TI ARM CGT (--be32) and some other toolchains.
+            # __ARM_BIG_ENDIAN covers ACLE-compatible Arm toolchains, including
+            # TI Arm Clang. __big_endian__ covers legacy TI ARM CGT (armcl).
+            # __BIG_ENDIAN__ covers some other toolchains.
+            # __LITTLE_ENDIAN__ == 0 covers IAR (which always defines it).
             self.push("#if !defined(BP_BIG_ENDIAN) && (\\")
             self.push(
                 "    (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)) || \\"
             )
-            self.push("    defined(__BIG_ENDIAN__))")
+            self.push("    defined(__ARM_BIG_ENDIAN) || \\")
+            self.push("    defined(__big_endian__) || \\")
+            self.push("    defined(__BIG_ENDIAN__) || \\")
+            self.push("    (defined(__LITTLE_ENDIAN__) && (__LITTLE_ENDIAN__ == 0)))")
             self.push("#define BP_BIG_ENDIAN 1")
             self.push("#endif")
             # <string.h> is only needed for memset in the big-endian path.

compiler/pyproject.toml

@@ -4,8 +4,9 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "bitproto"
-version = "1.3.0"
+version = "1.3.1"
 description = "bit level data interchange format."
+readme = "README.md"
 authors = [
     {name = "Chao Wang", email = "hit9@icloud.com"},
 ]

docs/endianness.rst

@@ -47,10 +47,13 @@ about C.
 C standard mode
 ---------------
 
-The C library (``lib/c/bitproto.c``) detects a big-endian host automatically via
-``__BYTE_ORDER__`` and selects an endian-neutral code path. You may also force it
-by defining ``BP_BIG_ENDIAN`` when compiling the library, for toolchains that do
-not define ``__BYTE_ORDER__``:
+The C library (``lib/c/bitproto.c``) detects a big-endian host automatically
+(via ``__BYTE_ORDER__`` for GCC/Clang, ``__ARM_BIG_ENDIAN`` for
+ACLE-compatible Arm toolchains such as TI Arm Clang, ``__big_endian__`` for
+legacy TI ARM CGT, ``__BIG_ENDIAN__`` for other toolchains, and
+``__LITTLE_ENDIAN__ == 0`` for IAR) and selects an endian-neutral code path.
+You may also force it by defining ``BP_BIG_ENDIAN`` when compiling the library,
+for toolchains that expose none of these macros:
 
 .. sourcecode:: bash
 
@@ -77,7 +80,10 @@ emits two equivalent code paths guarded by a preprocessor macro:
        // portable bit-shift path (big-endian hosts)
    #endif
 
-A big-endian host is auto-detected via ``__BYTE_ORDER__``; you may also force the
+A big-endian host is auto-detected (via ``__BYTE_ORDER__`` for GCC/Clang,
+``__ARM_BIG_ENDIAN`` for ACLE-compatible Arm toolchains such as TI Arm Clang,
+``__big_endian__`` for legacy TI ARM CGT, ``__BIG_ENDIAN__`` for other
+toolchains, and ``__LITTLE_ENDIAN__ == 0`` for IAR); you may also force the
 big-endian path by defining ``BP_BIG_ENDIAN`` when compiling the generated code.
 The little-endian path is identical to earlier bitproto releases, so there is no
 performance change on little-endian targets — only the generated source is a