parser: set clang -x<language> option at the low level

Have a single point of truth for setting the clang -x<language> option,
to unify parsing across the extension, tests, and cli.

Choose the language primarily based on the domain, and secondarily based
on the filename extension to differentiate headers. Using "-header" is
necessary for some cases such as "#pragma once", where plain "-xc" or
"-xc++" would interpret the file as a main file.

This unconditionally bypasses libclang automatic detection. The user and
tests can still override the parser selected language, as the user
provided clang_args are appended.

Fixes: #208

src/hawkmoth/__init__.py

@@ -53,9 +53,7 @@ def __display_parser_diagnostics(self, errors):
                             location=(self.env.docname, self.lineno))
 
     def __get_clang_args(self):
-        # Always pass `-xc++` to the compiler on 'cpp' domain as the first
-        # option so that the user can override it.
-        clang_args = ['-xc++'] if self._domain == 'cpp' else []
+        clang_args = []
 
         clang_args.extend(self.env.config.hawkmoth_clang.copy())
 

src/hawkmoth/parser.py

@@ -805,6 +805,18 @@ def _parse_undocumented_block(domain, comments, errors, cursor, nest):
 
     return ret
 
+def _language_option(filename, domain):
+    """Return clang -x<language> option depending on domain and filename."""
+    if domain == 'cpp':
+        language = '-xc++'
+    else:
+        language = '-xc'
+
+    if os.path.splitext(filename)[1] in ['.h', '.H', '.hh', '.hpp', '.hxx']:
+        language += '-header'
+
+    return language
+
 # Parse a file and return a tree of docstring.Docstring objects.
 def parse(filename, domain=None, clang_args=None):
     # Empty root comment with just children
@@ -813,8 +825,14 @@ def parse(filename, domain=None, clang_args=None):
     errors = []
     index = Index.create()
 
+    # Note: Preserve the passed in clang_args in RootDocstring, as it's used for
+    # filtering by the callers
+    full_args = [_language_option(filename, domain)]
+    if clang_args:
+        full_args.extend(clang_args)
+
     try:
-        tu = index.parse(filename, args=clang_args,
+        tu = index.parse(filename, args=full_args,
                          options=TranslationUnit.PARSE_DETAILED_PROCESSING_RECORD |
                          TranslationUnit.PARSE_SKIP_FUNCTION_BODIES)
     except TranslationUnitLoadError as e:

test/testenv.py

@@ -47,10 +47,7 @@ def get_directive_string(self):
         return directive_str
 
     def get_clang_args(self):
-        # Default to compile as C++ if the test is for the C++ domain so that we
-        # can use C sources for C++ tests. The yaml may override this in cases
-        # where we want to force a mismatch.
-        clang_args = ['-xc++'] if self.domain == 'cpp' else ['-xc']
+        clang_args = []
 
         clang_args.extend(_clang_include_args.copy())