Merge branch 'feature/add_compiler_no_merge_constants' into 'master'

feat: add compiler config for not merging const sections

See merge request espressif/esp-idf!32077

CMakeLists.txt

@@ -136,6 +136,10 @@ if(CONFIG_COMPILER_OPTIMIZATION_ASSERTIONS_DISABLE)
     list(APPEND compile_definitions "-DNDEBUG")
 endif()
 
+if(CONFIG_COMPILER_NO_MERGE_CONSTANTS)
+    list(APPEND compile_options "-fno-merge-constants")
+endif()
+
 if(CONFIG_COMPILER_STACK_CHECK_MODE_NORM)
     list(APPEND compile_options "-fstack-protector")
 elseif(CONFIG_COMPILER_STACK_CHECK_MODE_STRONG)

Kconfig

@@ -510,6 +510,15 @@ mainmenu "Espressif IoT Development Framework Configuration"
             help
                 Stack smashing protection.
 
+        config COMPILER_NO_MERGE_CONSTANTS
+            bool "Disable merging const sections"
+            depends on IDF_TOOLCHAIN_GCC
+            help
+                Disable merging identical constants (string/floating-point) across compilation units.
+                This helps in better size analysis of the application binary as the rodata section
+                distribution is more uniform across libraries. On downside, it may increase
+                the binary size and hence should be used during development phase only.
+
         config COMPILER_WARN_WRITE_STRINGS
             bool "Enable -Wwrite-strings warning flag"
             default "n"

docs/en/api-guides/tools/idf-size.rst

@@ -136,7 +136,7 @@ The table is sorted in descending order of the total contribution of the static
 
 .. note::
 
-    The size of the ``.rodata`` section in the ``Flash Data`` memory type may appear very large for a single archive. This occurs due to linker relaxations. The linker may attempt to combine object file sections with ``MERGE`` and ``STRINGS`` flags from all archives into one to perform tail string optimization. Consequently, one archive may end up with a very large ``.rodata`` section, containing string literals from other archives. This is evident in the ``.rodata`` section of the ``libesp_app_format.a`` archive.
+    The size of the ``.rodata`` section in the ``Flash Data`` memory type may appear very large for a single archive. This occurs due to linker relaxations. The linker may attempt to combine object file sections with ``MERGE`` and ``STRINGS`` flags from all archives into one to perform tail string optimization. Consequently, one archive may end up with a very large ``.rodata`` section, containing string literals from other archives. This is evident in the ``.rodata`` section of the ``libesp_app_format.a`` archive. The specific compiler behavior here can be turned off by enabling :ref:`CONFIG_COMPILER_NO_MERGE_CONSTANTS` option (only for GCC toolchain), please read help for more details.
 
 
 Source File Usage Summary ``idf.py size-files``

tools/test_apps/system/build_test/sdkconfig.ci.no_merge_constants

@@ -0,0 +1 @@
+CONFIG_COMPILER_NO_MERGE_CONSTANTS=y