Refactor system library claculation to better handle -nostdlib and friends.

This refactoring splits out `get_libs_to_link` into its own function
which can early return once the list is complete.

The follow gcc/clang flags are now (kind of) supported:

- `-nostartupfiles` - don't link crt1 and other startup files
- `-nolibc`         - don't link libc
- `-nodefualtlibs`  - don't link libc or compiler-rt or other stdlibs
                      (but do link startup files)
- `-nostdlib`       - don't link any system libraries or startup files.

Author: sbc100

emcc.py

@@ -2395,7 +2395,7 @@ def get_full_import_name(name):
 @ToolchainProfiler.profile_block('compile inputs')
 def phase_compile_inputs(options, state, newargs, input_files):
   def is_link_flag(flag):
-    if flag.startswith('-nostdlib'):
+    if flag in ('-nostdlib', '-nostartfiles', '-nolibc', '-nodefaultlibs'):
       return True
     return flag.startswith(('-l', '-L', '-Wl,'))
 
@@ -2561,8 +2561,8 @@ def phase_calculate_system_libraries(state, linker_arguments, linker_inputs, new
   if not settings.SIDE_MODULE:
     # Ports are always linked into the main module, never the size module.
     extra_files_to_link += ports.get_libs(settings)
-  if '-nostdlib' not in newargs and '-nodefaultlibs' not in newargs:
-    extra_files_to_link += system_libs.calculate([f for _, f in sorted(linker_inputs)] + extra_files_to_link, forced=state.forced_stdlibs)
+  all_linker_inputs = [f for _, f in sorted(linker_inputs)] + extra_files_to_link
+  extra_files_to_link += system_libs.calculate(all_linker_inputs, newargs, forced=state.forced_stdlibs)
   linker_arguments.extend(extra_files_to_link)
 
 

tests/test_other.py

@@ -10081,6 +10081,8 @@ def test_nostdlib(self):
     libs = ['-lc', '-lcompiler_rt', '-lc_rt']
     self.run_process([EMCC, test_file('unistd/close.c'), '-nostdlib'] + libs)
     self.run_process([EMCC, test_file('unistd/close.c'), '-nodefaultlibs'] + libs)
+    self.run_process([EMCC, test_file('unistd/close.c'), '-nolibc', '-lc'])
+    self.run_process([EMCC, test_file('unistd/close.c'), '-nostartfiles'])
 
   def test_argument_match(self):
     # Verify that emcc arguments match precisely.  We had a bug where only the prefix

tools/system_libs.py

@@ -1635,22 +1635,12 @@ def add_reverse_deps(need):
   for symbols in symbolses:
     add_reverse_deps(symbols)
 
+def get_libs_to_link(args, forced, only_forced):
+  libs_to_link = []
 
-def calculate(input_files, forced):
-  # Setting this will only use the forced libs in EMCC_FORCE_STDLIBS. This avoids spending time checking
-  # for unresolved symbols in your project files, which can speed up linking, but if you do not have
-  # the proper list of actually needed libraries, errors can occur. See below for how we must
-  # export all the symbols in deps_info when using this option.
-  only_forced = os.environ.get('EMCC_ONLY_FORCED_STDLIBS')
-  if only_forced:
-    # One of the purposes EMCC_ONLY_FORCED_STDLIBS was to skip the scanning
-    # of the input files for reverse dependencies.
-    diagnostics.warning('deprecated', 'EMCC_ONLY_FORCED_STDLIBS is deprecated.  Use `-nostdlib` and/or `-s REVERSE_DEPS=none` depending on the desired result')
-    settings.REVERSE_DEPS = 'all'
-
-  handle_reverse_deps(input_files)
+  if '-nostdlib' in args:
+    return libs_to_link
 
-  libs_to_link = []
   already_included = set()
   system_libs_map = Library.get_usable_variations()
 
@@ -1677,91 +1667,116 @@ def add_library(libname):
     need_whole_archive = lib.name in force_include and lib.get_ext() == '.a'
     libs_to_link.append((lib.get_link_flag(), need_whole_archive))
 
-  if settings.USE_PTHREADS:
-    add_library('crtbegin')
+  if '-nostartfiles' not in args:
+    if settings.USE_PTHREADS:
+      add_library('crtbegin')
 
-  if settings.SIDE_MODULE:
-    return [l[0] for l in libs_to_link]
+    if settings.STANDALONE_WASM:
+      if settings.EXPECT_MAIN:
+        add_library('crt1')
+      else:
+        add_library('crt1_reactor')
 
-  if settings.STANDALONE_WASM:
-    if settings.EXPECT_MAIN:
-      add_library('crt1')
-    else:
-      add_library('crt1_reactor')
+  if settings.SIDE_MODULE:
+    return libs_to_link
 
   for forced in force_include:
     if forced not in system_libs_map:
       shared.exit_with_error('invalid forced library: %s', forced)
     add_library(forced)
 
+  if '-nodefaultlibs' in args:
+    return libs_to_link
+
   if only_forced:
     add_library('libc_rt')
     add_library('libcompiler_rt')
-  else:
-    if settings.AUTO_NATIVE_LIBRARIES:
-      add_library('libGL')
-      add_library('libal')
-      add_library('libhtml5')
+    return libs_to_link
+
+  if settings.AUTO_NATIVE_LIBRARIES:
+    add_library('libGL')
+    add_library('libal')
+    add_library('libhtml5')
 
-    sanitize = settings.USE_LSAN or settings.USE_ASAN or settings.UBSAN_RUNTIME
+  sanitize = settings.USE_LSAN or settings.USE_ASAN or settings.UBSAN_RUNTIME
 
-    # JS math must come before anything else, so that it overrides the normal
-    # libc math.
-    if settings.JS_MATH:
-      add_library('libjsmath')
+  # JS math must come before anything else, so that it overrides the normal
+  # libc math.
+  if settings.JS_MATH:
+    add_library('libjsmath')
 
-    # to override the normal libc printf, we must come before it
-    if settings.PRINTF_LONG_DOUBLE:
-      add_library('libprintf_long_double')
+  # to override the normal libc printf, we must come before it
+  if settings.PRINTF_LONG_DOUBLE:
+    add_library('libprintf_long_double')
 
-    if settings.ALLOW_UNIMPLEMENTED_SYSCALLS:
-      add_library('libstubs')
+  if settings.ALLOW_UNIMPLEMENTED_SYSCALLS:
+    add_library('libstubs')
+  if '-nolibc' not in args:
     add_library('libc')
-    add_library('libcompiler_rt')
-    if settings.LINK_AS_CXX:
-      add_library('libc++')
-    if settings.LINK_AS_CXX or sanitize:
-      add_library('libc++abi')
-      if settings.EXCEPTION_HANDLING:
-        add_library('libunwind')
     if settings.MALLOC != 'none':
       add_library('libmalloc')
-    if settings.STANDALONE_WASM:
-      add_library('libstandalonewasm')
-    add_library('libc_rt')
+  add_library('libcompiler_rt')
+  if settings.LINK_AS_CXX:
+    add_library('libc++')
+  if settings.LINK_AS_CXX or sanitize:
+    add_library('libc++abi')
+    if settings.EXCEPTION_HANDLING:
+      add_library('libunwind')
+  if settings.STANDALONE_WASM:
+    add_library('libstandalonewasm')
+  add_library('libc_rt')
 
-    if settings.USE_LSAN:
-      force_include.add('liblsan_rt')
-      add_library('liblsan_rt')
+  if settings.USE_LSAN:
+    force_include.add('liblsan_rt')
+    add_library('liblsan_rt')
 
-    if settings.USE_ASAN:
-      force_include.add('libasan_rt')
-      add_library('libasan_rt')
-      add_library('libasan_js')
+  if settings.USE_ASAN:
+    force_include.add('libasan_rt')
+    add_library('libasan_rt')
+    add_library('libasan_js')
 
-    if settings.UBSAN_RUNTIME == 1:
-      add_library('libubsan_minimal_rt')
-    elif settings.UBSAN_RUNTIME == 2:
-      add_library('libubsan_rt')
+  if settings.UBSAN_RUNTIME == 1:
+    add_library('libubsan_minimal_rt')
+  elif settings.UBSAN_RUNTIME == 2:
+    add_library('libubsan_rt')
 
-    if settings.USE_LSAN or settings.USE_ASAN:
-      add_library('liblsan_common_rt')
+  if settings.USE_LSAN or settings.USE_ASAN:
+    add_library('liblsan_common_rt')
 
-    if sanitize:
-      add_library('libsanitizer_common_rt')
+  if sanitize:
+    add_library('libsanitizer_common_rt')
 
-    if settings.PROXY_POSIX_SOCKETS:
-      add_library('libsockets_proxy')
-    else:
-      add_library('libsockets')
+  if settings.PROXY_POSIX_SOCKETS:
+    add_library('libsockets_proxy')
+  else:
+    add_library('libsockets')
+
+  if settings.USE_WEBGPU:
+    add_library('libwebgpu_cpp')
+
+  return libs_to_link
+
+
+def calculate(input_files, args, forced):
+  # Setting this will only use the forced libs in EMCC_FORCE_STDLIBS. This avoids spending time checking
+  # for unresolved symbols in your project files, which can speed up linking, but if you do not have
+  # the proper list of actually needed libraries, errors can occur. See below for how we must
+  # export all the symbols in deps_info when using this option.
+  only_forced = os.environ.get('EMCC_ONLY_FORCED_STDLIBS')
+  if only_forced:
+    # One of the purposes EMCC_ONLY_FORCED_STDLIBS was to skip the scanning
+    # of the input files for reverse dependencies.
+    diagnostics.warning('deprecated', 'EMCC_ONLY_FORCED_STDLIBS is deprecated.  Use `-nostdlib` and/or `-s REVERSE_DEPS=none` depending on the desired result')
+    settings.REVERSE_DEPS = 'all'
+
+  handle_reverse_deps(input_files)
 
-    if settings.USE_WEBGPU:
-      add_library('libwebgpu_cpp')
+  libs_to_link = get_libs_to_link(args, forced, only_forced)
 
   # When LINKABLE is set the entire link command line is wrapped in --whole-archive by
   # building.link_ldd.  And since --whole-archive/--no-whole-archive processing does not nest we
   # shouldn't add any extra `--no-whole-archive` or we will undo the intent of building.link_ldd.
-  if settings.LINKABLE:
+  if settings.LINKABLE or settings.SIDE_MODULE:
     return [l[0] for l in libs_to_link]
 
   # Wrap libraries in --whole-archive, as needed.  We need to do this last