Merge branch 'fix/harden_empty_toolchain_output' into 'master'

fix: harden build against empty toolchain output

Closes IDFGH-17820

See merge request espressif/esp-idf!50136

components/esp_common/project_include.cmake

@@ -4,11 +4,7 @@
 idf_build_get_property(target IDF_TARGET)
 
 if(NOT (${target} STREQUAL "linux" OR CMAKE_C_COMPILER_ID MATCHES "Clang"))
-    execute_process(
+    __compiler_query(toolchain_name ${CMAKE_C_COMPILER} -dumpmachine)
-        COMMAND ${CMAKE_C_COMPILER} -dumpmachine
-        OUTPUT_VARIABLE toolchain_name
-        OUTPUT_STRIP_TRAILING_WHITESPACE
-        ERROR_QUIET)
     check_expected_tool_version(${toolchain_name} ${CMAKE_C_COMPILER})
 endif()
 

components/xtensa/project_include.cmake

@@ -7,20 +7,13 @@ endif()
 if(CMAKE_C_COMPILER_ID MATCHES "Clang")
     # without '--target' option 'clang -dumpmachine' prints default target arch and it might be not Xtensa
     # so use `-print-targets` option
-    execute_process(
+    __compiler_query(dump_machine ${CMAKE_C_COMPILER} -print-targets)
-        COMMAND ${CMAKE_C_COMPILER} -print-targets
-        OUTPUT_VARIABLE dump_machine
-    )
 else()
-    execute_process(
+    __compiler_query(dump_machine ${CMAKE_C_COMPILER} -dumpmachine)
-        COMMAND ${CMAKE_C_COMPILER} -dumpmachine
-        OUTPUT_VARIABLE dump_machine
-        OUTPUT_STRIP_TRAILING_WHITESPACE
-    )
 endif()
 message(STATUS "Compiler supported targets: ${dump_machine}")
 
-if(NOT (${CMAKE_SYSTEM_NAME} STREQUAL "Generic" AND ${dump_machine} MATCHES xtensa))
+if(NOT ("${CMAKE_SYSTEM_NAME}" STREQUAL "Generic" AND "${dump_machine}" MATCHES "xtensa"))
     # Sub-projects (e.g. ULP RISC-V on an xtensa target) may use a
     # non-IDF toolchain provided by the parent build. Skip validation
     # when a custom toolchain is explicitly declared.

tools/cmake/compiler_query.cmake

@@ -0,0 +1,40 @@
+# SPDX-FileCopyrightText: 2026 Espressif Systems (Shanghai) CO LTD
+# SPDX-License-Identifier: Apache-2.0
+
+# __compiler_query
+#
+# Run a compiler query command (e.g. "<compiler> -dumpmachine") given in ARGN
+# and store its trimmed stdout in the variable named by "output_var".
+#
+# Fails with an actionable error if the command fails or returns no output. On
+# some Windows systems, antivirus, endpoint-security or DLP/encryption software
+# intercepts short-lived toolchain processes and strips their stdout when it is
+# captured by the build system, while the same command works when run directly
+# in a terminal. Without this guard the empty result collapses the callers'
+# parsing into a cryptic CMake error (see
+# https://github.com/espressif/esp-idf/issues/18727).
+#
+# This module is included by both the cmakev1 and cmakev2 utilities, because the
+# esp_common and xtensa project_include.cmake files that call __compiler_query
+# are shared by both build systems.
+function(__compiler_query output_var)
+    execute_process(
+        COMMAND ${ARGN}
+        OUTPUT_VARIABLE query_output
+        RESULT_VARIABLE query_result
+        OUTPUT_STRIP_TRAILING_WHITESPACE)
+
+    if(NOT query_result EQUAL 0 OR query_output STREQUAL "")
+        string(REPLACE ";" " " query_command "${ARGN}")
+        message(FATAL_ERROR
+            "Failed to query the compiler: '${query_command}' (result: ${query_result}).\n"
+            "The command produced no output when run by the build system. This is usually caused "
+            "by antivirus, endpoint-security or DLP/encryption software intercepting the compiler "
+            "process and discarding its output; the same command often works when run directly in "
+            "a terminal.\n"
+            "Add an exclusion for the ESP-IDF tools directory in that software (on a managed "
+            "machine you may need your IT department), then run 'idf.py fullclean' and build again.")
+    endif()
+
+    set(${output_var} "${query_output}" PARENT_SCOPE)
+endfunction()

tools/cmake/utilities.cmake

@@ -509,3 +509,9 @@ function(add_deprecated_target_alias old_target new_target)
     )
     add_dependencies(${old_target} ${new_target})
 endfunction()
+
+
+# __compiler_query is defined in a standalone module so it can also be included
+# by the cmakev2 utilities; the esp_common and xtensa project_include.cmake
+# files that call it are shared by both build systems.
+include(${CMAKE_CURRENT_LIST_DIR}/compiler_query.cmake)

tools/cmakev2/utilities.cmake

@@ -7,6 +7,10 @@
 # used by toolchain CMake files.
 include(${CMAKE_CURRENT_LIST_DIR}/../cmake/deduplicate_flags.cmake)
 
+# __compiler_query is shared with cmakev1. The esp_common and xtensa
+# project_include.cmake files that call it are used by both build systems.
+include(${CMAKE_CURRENT_LIST_DIR}/../cmake/compiler_query.cmake)
+
 # Note: CMake does not support nested lists. The functions idf_die, idf_warn,
 # idf_msg, and idf_dbg use ARGV# values because this is the only way to prevent
 # arguments from being altered by CMake. ARGV and ARGN contain a flattened list

tools/idf_tools.py

@@ -998,7 +998,25 @@ class IDFTool:
                 f'non-zero exit code ({e.returncode}) with message: {e.stderr.decode("utf-8", errors="ignore")}'
             )  # type: ignore
 
-        return self.parse_tool_version(version_cmd_result.decode('utf-8'))
+        version_str = version_cmd_result.decode('utf-8')
+        if not version_str.strip():
+            # The tool ran and exited successfully, but produced no output when its
+            # output was captured. On some Windows systems, antivirus, endpoint-security
+            # or DLP/encryption software intercepts short-lived toolchain processes and
+            # strips their stdout when it is captured through a pipe, while the same
+            # command works when run directly in a terminal. Surface an actionable hint
+            # instead of silently reporting the version as 'unknown', which otherwise
+            # sends users into a fruitless reinstall loop.
+            # See https://github.com/espressif/esp-idf/issues/18727
+            warn(
+                f'tool {self.name} ran but returned no version output. This is usually caused by '
+                'antivirus, endpoint-security or DLP/encryption software stripping the output of '
+                'toolchain processes; the same command often works when run directly in a terminal. '
+                'Add an exclusion for the ESP-IDF tools directory in that software. If the problem '
+                'persists, run the tool manually to check for a missing DLL.'
+            )
+            return UNKNOWN_VERSION
+        return self.parse_tool_version(version_str)
 
     def get_version_from_file(self, version: str) -> str:
         """

tools/ldgen/ldgen.py

@@ -145,6 +145,40 @@ def _update_environment(env, env_file):
         os.environ.update(env_vars)
 
 
+def _run_objdump(objdump, library):
+    """Run ``objdump -h`` on a library and return its output.
+
+    On some Windows systems, antivirus, endpoint-security or DLP/encryption
+    software intercepts short-lived toolchain processes and strips their output
+    when the build captures it, so objdump exits successfully but returns empty
+    output, while the same command works when run by hand. The output is read
+    through a pipe so that this empty result is reliably detectable: it is
+    rejected here with an actionable error instead of being fed to the parser
+    (which would otherwise report it as a confusing pyparsing error). Capturing
+    to a file is deliberately avoided: it would not be guaranteed complete
+    either, and a truncated-but-non-empty result could be parsed into a wrong
+    linker script instead of failing. See
+    https://github.com/espressif/esp-idf/issues/18665 and
+    https://github.com/espressif/esp-idf/issues/18727.
+    """
+    new_env = os.environ.copy()
+    # Force the C locale so objdump emits the English 'In archive' header that
+    # the section parser expects, regardless of the host locale (see
+    # https://github.com/espressif/esp-idf/issues/7903).
+    new_env['LC_ALL'] = 'C'
+
+    output = subprocess.check_output([objdump, '-h', library], env=new_env).decode()
+    if not output.strip():
+        raise LdGenFailure(
+            f"'{objdump} -h {library}' ran successfully but returned no output. The toolchain ran "
+            'but its output was empty when captured by the build system. This is usually caused by '
+            'antivirus, endpoint-security or DLP/encryption software stripping the output of '
+            'toolchain processes; the same command often works when run directly in a terminal. '
+            'Add an exclusion for the ESP-IDF tools directory in that software, then build again.'
+        )
+    return output
+
+
 def _run(
     input_file,
     fragments,
@@ -184,11 +218,22 @@ def _run(
         for library in libraries_file:
             library = library.strip()
             if library:
-                new_env = os.environ.copy()
+                dump = StringIO(_run_objdump(objdump, library))
-                new_env['LC_ALL'] = 'C'
-                dump = StringIO(subprocess.check_output([objdump, '-h', library], env=new_env).decode())
                 dump.name = library
-                sections_infos.add_sections_info(dump)
+                try:
+                    sections_infos.add_sections_info(dump)
+                except ParseException as e:
+                    # Non-empty but unparsable section info (for example truncated or
+                    # corrupted toolchain output) is reported here rather than allowed to
+                    # propagate as a raw pyparsing traceback. The same root cause as the
+                    # empty case in _run_objdump applies.
+                    raise LdGenFailure(
+                        f'failed to parse section information from {library}. The toolchain output '
+                        'is incomplete or corrupted. This can be caused by antivirus, '
+                        'endpoint-security or DLP/encryption software tampering with the output of '
+                        'toolchain processes; the same command often works when run directly in a '
+                        f'terminal. Add an exclusion for the ESP-IDF tools directory, then build again.\n{e}'
+                    )
 
         # Check if we can skip generation entirely — section names and other
         # inputs unchanged since last run.