[clang][python][test] Move python binding tests to lit framework

As discussed in PR #142353, the current testsuite of the `clang` Python
bindings has several issues:

- It `libclang.so` cannot be loaded into `python` to run the testsuite, the
  whole `ninja check-all` aborts.
- The result of running the testsuite isn't report like the `lit`-based
  tests, rendering them almost invisible.
- The testsuite is disabled in a non-obvious way (`RUN_PYTHON_TESTS`) in
  `tests/CMakeLists.txt`, which again doesn't show up in the test results.

All these issues can be avoided by integrating the Python bindings tests
with `lit`, which is what this patch does:

- The actual test lives in `clang/test/bindings/python/bindings.sh` and is
  run by `lit`.
- The current `clang/bindings/python/tests` directory (minus the
  now-subperfluous `CMakeLists.txt`) is moved into the same directory.
- The check if `libclang` is loadable (originally from PR #142353) is now
  handled via a new `lit` feature, `libclang-loadable`.
- The various ways to disable the tests have been turned into `XFAIL`s as
  appropriate.  This isn't complete and not completely tested yet.

Tested on `sparc-sun-solaris2.11`, `sparcv9-sun-solaris2.11`,
`i386-pc-solaris2.11`, `amd64-pc-solaris2.11`, `i686-pc-linux-gnu`, and
`x86_64-pc-linux-gnu`.

Author: rorth

.github/workflows/libclang-python-tests.yml

@@ -10,15 +10,15 @@ on:
       - 'main'
     paths:
       - 'clang/bindings/python/**'
+      - 'clang/test/bindings/python/**'
       - 'clang/tools/libclang/**'
-      - 'clang/CMakeList.txt'
       - '.github/workflows/libclang-python-tests.yml'
       - '.github/workflows/llvm-project-tests.yml'
   pull_request:
     paths:
       - 'clang/bindings/python/**'
+      - 'clang/test/bindings/python/**'
       - 'clang/tools/libclang/**'
-      - 'clang/CMakeList.txt'
       - '.github/workflows/libclang-python-tests.yml'
       - '.github/workflows/llvm-project-tests.yml'
 

clang/CMakeLists.txt

@@ -530,7 +530,6 @@ if( CLANG_INCLUDE_TESTS )
     clang_unit_site_config=${CMAKE_CURRENT_BINARY_DIR}/test/Unit/lit.site.cfg
   )
   add_subdirectory(test)
-  add_subdirectory(bindings/python/tests)
 
   if(CLANG_BUILT_STANDALONE)
     umbrella_lit_testsuite_end(check-all)

clang/bindings/python/tests/CMakeLists.txt

@@ -205,6 +205,17 @@ add_custom_target(clang-test)
 add_dependencies(clang-test check-clang)
 set_target_properties(clang-test PROPERTIES FOLDER "Clang/Tests")
 
+# Allow running Clang Python binding tests separately from CI.
+add_lit_testsuite(check-clang-python "Running the Clang Python tests"
+  ${CMAKE_CURRENT_BINARY_DIR}
+  #LIT ${LLVM_LIT}
+  PARAMS ${CLANG_TEST_PARAMS}
+  DEPENDS ${CLANG_TEST_DEPS}
+  ARGS ${CLANG_TEST_EXTRA_ARGS} --filter=bindings.sh
+  # Avoid running tests twice.
+  EXCLUDE_FROM_CHECK_ALL
+  )
+
 # FIXME: This logic can be removed once all buildbots have moved
 # debuginfo-test from clang/test to llvm/projects or monorepo.
 if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/debuginfo-tests)

clang/test/CMakeLists.txt

@@ -0,0 +1,32 @@
+#!/bin/sh
+
+# UNSUPPORTED: !libclang-loadable
+
+# Tests fail on Windows, and need someone knowledgeable to fix.
+# It's not clear whether it's a test or a valid binding problem.
+# XFAIL: target={{.*windows.*}}
+
+# The Python FFI interface is broken on AIX: https://bugs.python.org/issue38628.
+# XFAIL: target={{.*-aix.*}}
+
+# Hexagon has known test failures that need to be addressed.
+# https://reviews.llvm.org/D52840#1265716
+# XFAIL: target={{hexagon-.*}}
+# python SEGVs on Linux/sparc64 when loading libclang.so.  Seems to be an FFI
+# issue, too.
+# XFAIL: target={{sparc.*-.*-linux.*}}
+
+# Tests will fail if cross-compiling for a different target, as tests will try
+# to use the host Python3_EXECUTABLE and make FFI calls to functions in target
+# libraries.
+#
+# FIXME: Consider a solution that allows better control over these tests in
+# a crosscompiling scenario. e.g. registering them with lit to allow them to
+# be explicitly skipped via appropriate LIT_ARGS, or adding a mechanism to
+# allow specifying a python interpreter compiled for the target that could
+# be executed using qemu-user.
+# REQUIRES: native
+
+# RUN: env PYTHONPATH=%S/../../../bindings/python \
+# RUN:   CLANG_LIBRARY_PATH=%libdir \
+# RUN:   %python -m unittest discover -s %S/tests

clang/test/bindings/python/bindings.sh

@@ -0,0 +1,41 @@
+def is_libclang_loadable():
+    # Do not try to run if libclang was built with sanitizers because
+    # the sanitizer library will likely be loaded too late to perform
+    # interception and will then fail.
+    # We could use LD_PRELOAD/DYLD_INSERT_LIBRARIES but this isn't
+    # portable so its easier just to not run the tests when building
+    # with ASan.
+    if config.llvm_use_sanitizer != "":
+        return False
+    try:
+        sys.path.append(os.path.join(config.clang_src_dir, "bindings/python"))
+        from clang.cindex import Config
+        conf = Config()
+        Config.set_library_path(config.clang_lib_dir)
+        conf.lib
+        return True
+    except Exception as e:
+        # Expected failure modes are considered benign when nothing can be
+        # done about them.
+        #
+        # Cannot load a 32-bit libclang.so into a 64-bit python.
+        if "wrong ELF class: ELFCLASS32" in str(e):
+            return False
+        # If libclang.so is missing, it must have been disabled intentionally,
+        # e.g. by building with LLVM_ENABLE_PIC=OFF.
+        elif "No such file or directory" in str(e):
+            return False
+        # Unexpected failure modes need to be investigated to either fix an
+        # underlying bug or accept the failure, so return True.  This causes
+        # tests to run and FAIL, drawing developer attention.
+        else:
+            print("warning: unhandled failure in is_libclang_loadable: "
+                  + str(e), file=sys.stderr)
+            return True
+
+if is_libclang_loadable():
+    config.available_features.add("libclang-loadable")
+
+config.substitutions.append(('%libdir', config.clang_lib_dir))
+
+config.suffixes = ['.sh']