[analyze_memory] Skip constants whose linked copy is unknown, detect jump tables from unnamed bytes only

esphome/analyze_memory/__init__.py

@@ -2,6 +2,7 @@
 
 from collections import Counter, defaultdict
 from dataclasses import dataclass, field
+from functools import cached_property
 import logging
 from pathlib import Path
 import re
@@ -55,6 +56,7 @@ _COMPONENT_PREFIX_EXTERNAL = "[external]"
 _COMPONENT_PREFIX_LIB = "[lib]"
 _COMPONENT_CORE = f"{_COMPONENT_PREFIX_ESPHOME}core"
 _COMPONENT_API = f"{_COMPONENT_PREFIX_ESPHOME}api"
+_COMPONENT_ARDUINO_CORE = "arduino_core"
 
 # Placement new storage suffix (generated by codegen Pvariable)
 _PSTORAGE_SUFFIX = "__pstorage"
@@ -83,6 +85,24 @@ _NAMESPACE_STD = "std::"
 # Type alias for symbol information: (symbol_name, size, component)
 SymbolInfoType = tuple[str, int, str]
 
+# Only ESP8266 has this section; its .rodata is copied to RAM at boot like .data
+_ESP8266_FLASH_CODE_SECTION = ".irom0.text"
+
+# readelf -SW row: [Nr] Name Type Address Off Size ES Flg
+_OBJ_SECTION_PATTERN = re.compile(
+    r"^\s*\[\s*(\d+)\]\s+(\S+)\s+(\S+)\s+[\da-f]+\s+[\da-f]+\s+([\da-f]+)\s+[\da-f]+\s+(\S*)"
+)
+_RELOC_SECTION_PATTERN = re.compile(r"^Relocation section '\.rela(\S+)'")
+# Code section; group 1 is the function for -ffunction-sections names, after any
+# placement prefix (.text.startup.<fn>, .iram.text.<fn> for IRAM_ATTR)
+_CODE_SECTION_PATTERN = re.compile(
+    r"^\.(?:literal|text|irom0\.text|iram\.text|iram0\.text)"
+    r"(?:\.(?:startup|unlikely|hot|exit)(?=\.|$))?(?:\.(.+))?$"
+)
+# readelf -x row: "  0x00000000 0000803f 0000803f 0000803f 0000803f ...?...?...?...?"
+_HEX_DUMP_PREFIX = "  0x"
+_HEX_DUMP_COLUMNS = slice(13, 48)
+
 # RAM sections - symbols in these sections consume RAM
 RAM_SECTIONS = frozenset([".data", ".bss"])
 
@@ -119,6 +139,19 @@ class SDKSymbol:
     demangled: str = ""  # Demangled name (populated after analysis)
 
 
+@dataclass
+class UnnamedConstant:
+    """Constant without a symbol in an object's .rodata (RAM on ESP8266)."""
+
+    size: int
+    source_file: str
+    component: str
+    functions: list[str]  # Functions that reference it, mangled
+    jump_table: bool  # Holds code addresses (a switch table), not a value
+    data: bytes  # Leading bytes, to recognize the value
+    strings: list[str] = field(default_factory=list)  # String literals it points to
+
+
 @dataclass
 class ComponentMemory:
     """Tracks memory usage for a component."""
@@ -152,6 +185,146 @@ class ComponentMemory:
         return self.data_size + self.bss_size
 
 
+@dataclass
+class _RodataSection:
+    """An object's .rodata section and the byte ranges its symbols cover."""
+
+    name: str
+    size: int
+    covered: list[tuple[int, int]] = field(default_factory=list)  # (offset, size)
+
+    @cached_property
+    def _named(self) -> frozenset[int]:
+        return frozenset(
+            offset
+            for start, size in self.covered
+            for offset in range(start, min(start + size, self.size))
+        )
+
+    def is_unnamed(self, offset: int) -> bool:
+        return 0 <= offset < self.size and offset not in self._named
+
+    @property
+    def unnamed_size(self) -> int:
+        return self.size - len(self._named)
+
+
+def _parse_hex_dumps(output: str) -> dict[str, bytearray]:
+    """Collect the bytes of each ``readelf -x`` section dump by section name."""
+    contents: dict[str, bytearray] = defaultdict(bytearray)
+    section: str | None = None
+    for line in output.splitlines():
+        if line.startswith("Hex dump of section"):
+            section = line.split("'")[1]
+        elif section is not None and line.startswith(_HEX_DUMP_PREFIX):
+            contents[section] += bytes.fromhex(line[_HEX_DUMP_COLUMNS].replace(" ", ""))
+        elif line and not line.startswith(" "):
+            section = None
+    return contents
+
+
+def _parse_relocations(output: str) -> list[tuple[str, int, str, int]]:
+    """List (relocated section, offset, target, addend) from ``readelf -rW`` output."""
+    relocations: list[tuple[str, int, str, int]] = []
+    section: str | None = None
+    for line in output.splitlines():
+        if line.startswith("Relocation section"):
+            match = _RELOC_SECTION_PATTERN.match(line)
+            section = match.group(1) if match else None
+        elif line[:1].isdigit() and section is not None:
+            # Offset Info Type Sym.Value Sym.Name + Addend
+            fields = line.split()
+            if len(fields) < 5:
+                continue
+            addend = int(fields[6], 16) if len(fields) > 6 else 0
+            if len(fields) > 6 and fields[5] == "-":
+                addend = -addend
+            relocations.append((section, int(fields[0], 16), fields[4], addend))
+        elif line and line[0] != " ":
+            section = None
+    return relocations
+
+
+def _pointed_literal(
+    referrer: bytes, offset: int, addend: int, strings: bytes
+) -> bytes:
+    """Read the string literal a relocated pointer at ``offset`` refers to."""
+    # Xtensa keeps part of the addend in the relocated word
+    start = addend + int.from_bytes(referrer[offset : offset + 4], "little")
+    end = strings.find(b"\0", start)
+    return bytes(strings[start : end if end != -1 else len(strings)])
+
+
+@dataclass
+class _FunctionDefinitions:
+    """Where each function is defined across the scanned objects."""
+
+    weak: dict[str, set[str]] = field(default_factory=lambda: defaultdict(set))
+    strong: set[str] = field(default_factory=set)  # Global, non-weak
+    count: Counter[str] = field(default_factory=Counter)  # Objects defining it
+
+    def linked_copy_unknown(self, file_path: str, functions: list[str]) -> bool:
+        """True if another object's copy of these functions may be the linked one."""
+        if self.weak[file_path].issuperset(functions):
+            # Inline copies: a strong definition elsewhere replaces them all
+            return not self.strong.isdisjoint(functions)
+        # Same-named local functions in several objects; the ELF can't tell them apart
+        return all(f not in self.strong and self.count[f] > 1 for f in functions)
+
+
+def _parse_unnamed_rodata(
+    output: str,
+) -> tuple[dict[str, dict[str, _RodataSection]], _FunctionDefinitions]:
+    """Find .rodata sections with bytes no symbol covers in ``readelf -SW -sW`` output.
+
+    Returns:
+        Object path -> section index -> section, and where functions are defined
+    """
+    sections: dict[str, dict[str, _RodataSection]] = defaultdict(dict)
+    definitions = _FunctionDefinitions()
+    file_path = ""
+    for line in output.splitlines():
+        if line.startswith("File: "):
+            file_path = line[6:]
+        elif match := _OBJ_SECTION_PATTERN.match(line):
+            index, name, section_type, size_hex, flags = match.groups()
+            # Merged sections (string literals) are shared between objects
+            if (
+                section_type == "PROGBITS"
+                and name.startswith(".rodata")
+                and "A" in flags
+                and "M" not in flags
+            ):
+                sections[file_path][index] = _RodataSection(name, int(size_hex, 16))
+        else:
+            # Num: Value Size Type Bind Vis Ndx Name
+            fields = line.split()
+            if len(fields) < 8:
+                continue
+            if fields[3] == "OBJECT" and (
+                section := sections[file_path].get(fields[6])
+            ):
+                section.covered.append((int(fields[1], 16), int(fields[2], 0)))
+            elif fields[3] == "FUNC" and fields[6] != "UND":
+                definitions.count[fields[7]] += 1
+                if fields[4] == "WEAK":
+                    definitions.weak[file_path].add(fields[7])
+                elif fields[4] == "GLOBAL":
+                    definitions.strong.add(fields[7])
+    unnamed = {
+        path: kept
+        for path, file_sections in sections.items()
+        if (
+            kept := {
+                index: section
+                for index, section in file_sections.items()
+                if section.unnamed_size
+            }
+        )
+    }
+    return unnamed, definitions
+
+
 class MemoryAnalyzer:
     """Analyzes memory usage from ELF files."""
 
@@ -231,6 +404,9 @@ class MemoryAnalyzer:
         self._heuristic_to_lib: dict[str, str] = {}
         # Function call counts: mangled_name -> call_count
         self._function_call_counts: Counter[str] = Counter()
+        # True on ESP8266, where .rodata is RAM; it is then counted as .data
+        self.rodata_in_ram = False
+        self._unnamed_constants: list[UnnamedConstant] = []
 
     def analyze(self) -> dict[str, ComponentMemory]:
         """Analyze the ELF file and return component memory usage."""
@@ -240,6 +416,7 @@ class MemoryAnalyzer:
         self._scan_source_symbols()
         self._categorize_symbols()
         self._analyze_cswtch_symbols()
+        self._analyze_object_rodata()
         self._analyze_sdk_libraries()
         self._analyze_function_calls()
         return dict(self.components)
@@ -253,18 +430,17 @@ class MemoryAnalyzer:
             check=True,
         )
 
-        # Parse section headers
-        for line in result.stdout.splitlines():
-            # Look for section entries
-            if not (match := _READELF_SECTION_PATTERN.match(line)):
-                continue
+        raw_sections = [
+            (match.group(1), int(match.group(2), 16))
+            for line in result.stdout.splitlines()
+            if (match := _READELF_SECTION_PATTERN.match(line))
+        ]
+        self.rodata_in_ram = any(
+            name == _ESP8266_FLASH_CODE_SECTION for name, _ in raw_sections
+        )
 
-            section_name = match.group(1)
-            size_hex = match.group(2)
-            size = int(size_hex, 16)
-
-            # Map to standard section name
-            mapped_section = map_section_name(section_name)
+        for section_name, size in raw_sections:
+            mapped_section = self._standard_section(map_section_name(section_name))
             if not mapped_section:
                 continue
 
@@ -272,6 +448,12 @@ class MemoryAnalyzer:
                 self.sections[mapped_section] = MemorySection(mapped_section)
             self.sections[mapped_section].total_size += size
 
+    def _standard_section(self, section: str | None) -> str | None:
+        """Count .rodata as .data where it is RAM (ESP8266)."""
+        if section == ".rodata" and self.rodata_in_ram:
+            return ".data"
+        return section
+
     def _parse_symbols(self) -> None:
         """Parse symbols from ELF file."""
         result = subprocess.run(
@@ -289,14 +471,18 @@ class MemoryAnalyzer:
                 continue
 
             section, name, size, address = symbol_info
+            section = self._standard_section(section)
 
+            if section not in self.sections:
+                continue
+            # Every alias is in the ELF, even when its size is counted once
+            self._elf_symbol_names.add(name)
             # Skip duplicate symbols at the same address (e.g., C1/C2 constructors)
-            if address in seen_addresses or section not in self.sections:
+            if address in seen_addresses:
                 continue
 
             self.sections[section].symbols.append((name, size, ""))
             self.sections[section].symbol_size += size
-            self._elf_symbol_names.add(name)
             seen_addresses.add(address)
 
     def _categorize_symbols(self) -> None:
@@ -925,10 +1111,12 @@ class MemoryAnalyzer:
             Path to the directory containing .o files, or None if not found.
         """
         # The ELF is typically at .pioenvs/<env>/firmware.elf
-        # Object files are in .pioenvs/<env>/src/ and .pioenvs/<env>/lib*/
+        # PlatformIO puts object files in .pioenvs/<env>/src/ and .pioenvs/<env>/lib*/,
+        # the native ESP8266 build in .pioenvs/<env>/obj/src/ and obj/lib/
         pioenvs_dir = self.elf_path.parent
-        if pioenvs_dir.exists() and any(pioenvs_dir.glob("src/*.o")):
-            return pioenvs_dir
+        for obj_dir in (pioenvs_dir, pioenvs_dir / "obj"):
+            if any(obj_dir.glob("src/*.o")):
+                return obj_dir
         return None
 
     @staticmethod
@@ -1110,6 +1298,12 @@ class MemoryAnalyzer:
         ):
             return f"{_COMPONENT_PREFIX_LIB}{self._lib_hash_to_name[parts[1]]}"
 
+        # Native ESP8266 build: lib/<owner>__<Name>/... and the Arduino core in core/
+        if len(parts) >= 3 and parts[0] == "lib":
+            return f"{_COMPONENT_PREFIX_LIB}{parts[1].rsplit('__', 1)[-1].lower()}"
+        if len(parts) >= 2 and parts[0] == "core":
+            return _COMPONENT_ARDUINO_CORE
+
         # Other framework/library files - return the first path component
         # e.g., FrameworkArduino/... -> FrameworkArduino
         return parts[0] if parts else source_file
@@ -1191,9 +1385,7 @@ class MemoryAnalyzer:
                 and other_mem is not None
             ):
                 other_mem.add_section_size(section_name, -size)
-                if component not in self.components:
-                    self.components[component] = ComponentMemory(component)
-                self.components[component].add_section_size(section_name, size)
+                self._add_component_size(component, section_name, size)
 
         # Sort by size descending
         self._cswtch_symbols.sort(key=lambda x: x[1], reverse=True)
@@ -1205,6 +1397,146 @@ class MemoryAnalyzer:
             total_size,
         )
 
+    def _analyze_object_rodata(self) -> None:
+        """Attribute .rodata without a symbol, which is RAM on ESP8266, to components."""
+        if not self.rodata_in_ram or (obj_dir := self._find_object_files_dir()) is None:
+            return
+        result = run_tool(
+            [self.readelf_path, "-SW", "-sW"]
+            + [str(f) for f in sorted(obj_dir.rglob("*.o"))]
+        )
+        if result is None or result.returncode != 0:
+            _LOGGER.debug("readelf failed scanning object files for constants")
+            return
+        self._attribute_unnamed_constants(result.stdout, obj_dir)
+        names = {n for c in self._unnamed_constants for n in c.functions}
+        self._batch_demangle_symbols(sorted(names - self._demangle_cache.keys()))
+
+    def _attribute_unnamed_constants(self, output: str, obj_dir: Path) -> None:
+        """Attribute constants GCC copies into objects built from constant initializers."""
+        unnamed, definitions = _parse_unnamed_rodata(output)
+        # Inline (weak) functions are emitted, with their constants, in every
+        # object that uses them; the linker keeps one copy
+        inline_copies: dict[tuple[tuple[str, ...], int], str] = {}
+        for file_path, sections in unnamed.items():
+            rel_path = Path(file_path).relative_to(obj_dir).as_posix()
+            component = self._source_file_to_component(rel_path)
+            for constant in self._scan_object_constants(
+                file_path, sections, rel_path, component
+            ):
+                if definitions.linked_copy_unknown(file_path, constant.functions):
+                    continue
+                if definitions.weak[file_path].issuperset(constant.functions):
+                    key = (tuple(constant.functions), constant.size)
+                    if inline_copies.setdefault(key, file_path) != file_path:
+                        continue
+                self._unnamed_constants.append(constant)
+                self._charge_ram(component, constant.size)
+
+        self._unnamed_constants.sort(key=lambda c: c.size, reverse=True)
+
+    def _scan_object_constants(
+        self,
+        file_path: str,
+        sections: dict[str, _RodataSection],
+        rel_path: str,
+        component: str,
+    ) -> list[UnnamedConstant]:
+        """Return the unnamed constants of one object that the link kept.
+
+        Args:
+            file_path: Object file to read
+            sections: Candidate sections by index
+            rel_path: Object path shown in the report
+            component: Component the object belongs to
+        """
+        result = run_tool(
+            [self.readelf_path, "-rW"]
+            + [arg for index in sections for arg in ("-x", index)]
+            + [file_path]
+        )
+        if result is None or result.returncode != 0:
+            _LOGGER.debug("readelf failed reading constants of %s", file_path)
+            return []
+
+        by_name = {section.name: section for section in sections.values()}
+        contents = _parse_hex_dumps(result.stdout)
+        functions: dict[str, set[str]] = defaultdict(set)
+        jump_tables: set[str] = set()
+        # Section -> (offset, string section, addend) of each pointer to a literal
+        pointers: dict[str, list[tuple[int, str, int]]] = defaultdict(list)
+        for section, offset, target, addend in _parse_relocations(result.stdout):
+            if target in by_name:
+                code = _CODE_SECTION_PATTERN.match(section)
+                if code and code.group(1):
+                    functions[target].add(code.group(1))
+            elif section in by_name and by_name[section].is_unnamed(offset):
+                if _CODE_SECTION_PATTERN.match(target):
+                    jump_tables.add(section)
+                elif ".str" in target:
+                    pointers[section].append((offset, target, addend))
+
+        strings = self._read_pointed_strings(file_path, pointers, contents)
+        return [
+            UnnamedConstant(
+                size=section.unnamed_size,
+                source_file=rel_path,
+                component=component,
+                functions=sorted(functions[name]),
+                jump_table=name in jump_tables,
+                data=bytes(
+                    byte
+                    for offset, byte in enumerate(contents[name])
+                    if section.is_unnamed(offset)
+                )[:16],
+                strings=strings.get(name, []),
+            )
+            for name, section in by_name.items()
+            # Dropped by --gc-sections unless a function that uses it was kept
+            if not self._elf_symbol_names.isdisjoint(functions[name])
+        ]
+
+    def _read_pointed_strings(
+        self,
+        file_path: str,
+        pointers: dict[str, list[tuple[int, str, int]]],
+        contents: dict[str, bytearray],
+    ) -> dict[str, list[str]]:
+        """Resolve each constant's pointers to string literals, in offset order."""
+        if not pointers:
+            return {}
+        targets = sorted({t for refs in pointers.values() for _, t, _ in refs})
+        result = run_tool(
+            [self.readelf_path]
+            + [arg for target in targets for arg in ("-x", target)]
+            + [file_path]
+        )
+        if result is None or result.returncode != 0:
+            _LOGGER.debug("readelf failed reading strings of %s", file_path)
+            return {}
+        literals = _parse_hex_dumps(result.stdout)
+        return {
+            name: [
+                _pointed_literal(
+                    contents[name], offset, addend, literals[target]
+                ).decode("utf-8", "backslashreplace")
+                for offset, target, addend in sorted(refs)
+            ]
+            for name, refs in pointers.items()
+        }
+
+    def _charge_ram(self, component: str, size: int) -> None:
+        """Charge RAM bytes that have no ELF symbol to a component."""
+        self._add_component_size(component, ".data", size)
+        if (data_section := self.sections.get(".data")) is not None:
+            data_section.symbol_size += size
+
+    def _add_component_size(self, component: str, section: str, size: int) -> None:
+        """Add size to a component, creating it under its own name."""
+        if component not in self.components:
+            self.components[component] = ComponentMemory(component)
+        self.components[component].add_section_size(section, size)
+
     def _analyze_function_calls(self) -> None:
         """Count function call sites by parsing disassembly output.
 

esphome/analyze_memory/cli.py

@@ -5,6 +5,7 @@ from __future__ import annotations
 from collections import defaultdict
 from collections.abc import Callable
 import heapq
+import json
 from operator import itemgetter
 from pathlib import Path
 import sys
@@ -204,6 +205,36 @@ class MemoryAnalyzerCLI(MemoryAnalyzer):
                 f"{i + 1:>2}. {size:>7,} B {section_label:<8} {demangled_display:<{self.COL_TOP_SYMBOL_NAME}} {component}"
             )
 
+    def _add_unnamed_constants_analysis(self, lines: list[str]) -> None:
+        """Add the unnamed constants (ESP8266 .rodata in RAM) section."""
+        self._add_section_header(lines, "Unnamed Constants in RAM (.rodata)")
+        total_size = sum(c.size for c in self._unnamed_constants)
+        lines.append(
+            f"Total: {len(self._unnamed_constants)} constant(s), {total_size:,} B. "
+            "GCC copies these into objects built from constant initializers;"
+        )
+        lines.append(
+            "PROGMEM tables, assigned fields or a user-provided constructor avoid the copy."
+        )
+        lines.append("")
+        for constant in self._unnamed_constants:
+            kind = "switch table" if constant.jump_table else "constant"
+            lines.append(
+                f"{constant.size:>6,} B {kind:<12} {constant.component} ({constant.source_file})"
+            )
+            lines.extend(
+                f"{'':>9} used by {self._demangle_symbol(function)}"
+                for function in constant.functions
+            )
+            if constant.strings:
+                shown = ", ".join(f'"{text[:40]}"' for text in constant.strings[:4])
+                more = ", ..." if len(constant.strings) > 4 else ""
+                lines.append(f"{'':>9} points to {shown}{more}")
+            elif constant.data and not constant.jump_table:
+                more = " ..." if constant.size > len(constant.data) else ""
+                lines.append(f"{'':>9} bytes: {constant.data.hex(' ')}{more}")
+        lines.append("")
+
     def _add_cswtch_analysis(self, lines: list[str]) -> None:
         """Add CSWTCH (GCC switch table lookup) analysis section."""
         self._add_section_header(lines, "CSWTCH Analysis (GCC Switch Table Lookups)")
@@ -394,6 +425,12 @@ class MemoryAnalyzerCLI(MemoryAnalyzer):
             f"{total_flash:>{self.COL_TOTAL_FLASH - 2},} B | {total_ram:>{self.COL_TOTAL_RAM - 2},} B"
         )
 
+        if self.rodata_in_ram:
+            lines.append("")
+            lines.append(
+                "ESP8266 keeps .rodata in RAM, so constants are counted in RAM (data)."
+            )
+
         # Show unattributed RAM (SDK/framework overhead)
         unattributed_bss, unattributed_data, unattributed_total = (
             self.get_unattributed_ram()
@@ -657,6 +694,9 @@ class MemoryAnalyzerCLI(MemoryAnalyzer):
                     lines.append(f"    ... and {len(large_ram_syms) - 10} more")
             lines.append("")
 
+        if self._unnamed_constants:
+            self._add_unnamed_constants_analysis(lines)
+
         # CSWTCH (GCC switch table) analysis
         if self._cswtch_symbols:
             self._add_cswtch_analysis(lines)
@@ -738,7 +778,6 @@ def main():
     build_dir = sys.argv[1]
 
     # Load build directory
-    import json
 
     from esphome.platformio.toolchain import IDEData
 

tests/unit_tests/analyze_memory/__init__.py

@@ -1,15 +1,34 @@
 """Tests for the memory analyzer."""
 
+import subprocess
 from unittest.mock import patch
 
 from esphome.analyze_memory import MemoryAnalyzer
 
 
-def make_analyzer(external_components: set[str] | None = None) -> MemoryAnalyzer:
-    """Create a MemoryAnalyzer through its real constructor, without an ELF or a toolchain."""
+def make_analyzer[T: MemoryAnalyzer](
+    external_components: set[str] | None = None, cls: type[T] = MemoryAnalyzer
+) -> T:
+    """Create an analyzer through its real constructor, without an ELF or a toolchain."""
     with (
         patch("esphome.analyze_memory.resolve_tool_path", return_value=None),
         patch("esphome.analyze_memory.find_tool", return_value="nm"),
     ):
         # Any existing file passes the ELF check; nothing reads it
-        return MemoryAnalyzer(__file__, external_components=external_components)
+        return cls(__file__, external_components=external_components)
+
+
+def completed(stdout: str) -> subprocess.CompletedProcess[str]:
+    """A successful tool run printing ``stdout``."""
+    return subprocess.CompletedProcess([], 0, stdout=stdout)
+
+
+def hex_dump(section: str, data: bytes, address: int = 0) -> str:
+    """Format bytes the way ``readelf -x`` does."""
+    lines = [f"Hex dump of section '{section}':"]
+    for start in range(0, len(data), 16):
+        chunk = data[start : start + 16]
+        words = " ".join(chunk[i : i + 4].hex() for i in range(0, len(chunk), 4))
+        text = "".join(chr(b) if 32 <= b < 127 else "." for b in chunk)
+        lines.append(f"  0x{address + start:08x} {words:<35} {text}")
+    return "\n".join(lines) + "\n\n"

tests/unit_tests/analyze_memory/test_source_file_attribution.py

@@ -1,5 +1,7 @@
 """Tests for source-file-to-component attribution in memory analyzer."""
 
+import pytest
+
 from . import make_analyzer
 
 
@@ -30,3 +32,19 @@ def test_source_file_to_component_known_component() -> None:
         "src/esphome/components/wifi/wifi_component.cpp.o"
     )
     assert result == "[esphome]wifi"
+
+
+@pytest.mark.parametrize(
+    ("rel_path", "component"),
+    [
+        (
+            "lib/esp32async__ESPAsyncWebServer/WebHandlers.cpp.o",
+            "[lib]espasyncwebserver",
+        ),
+        ("lib/ESP8266WiFi/ESP8266WiFi.cpp.o", "[lib]esp8266wifi"),
+        ("core/core_esp8266_main.cpp.o", "arduino_core"),
+    ],
+)
+def test_native_library_objects(rel_path: str, component: str) -> None:
+    """Objects of the native ESP8266 build's libraries and core get a component."""
+    assert make_analyzer()._source_file_to_component(rel_path) == component

tests/unit_tests/analyze_memory/test_unnamed_constants.py

@@ -0,0 +1,475 @@
+"""Tests for ESP8266 .rodata accounting and unnamed constant attribution."""
+
+from collections.abc import Callable
+from pathlib import Path
+import subprocess
+from unittest.mock import patch
+
+import pytest
+
+from esphome.analyze_memory import (
+    _CODE_SECTION_PATTERN,
+    MemoryAnalyzer,
+    MemorySection,
+    UnnamedConstant,
+    _FunctionDefinitions,
+    _parse_unnamed_rodata,
+    _RodataSection,
+)
+from esphome.analyze_memory.cli import MemoryAnalyzerCLI
+
+from . import completed, hex_dump, make_analyzer
+
+LIGHT_OBJ = "src/esphome/components/light/light_state.cpp.o"
+WEB_OBJ = "src/esphome/components/web_server/web_server.cpp.o"
+SETUP = "_ZN7esphome5light10LightState5setupEv"
+SAVE = "_ZN7esphome5light10LightState19save_remote_values_Ev"
+MATCH_URL = "_ZN7esphome10web_serverL9match_urlEPKcjbb"
+
+# readelf -SW -sW over all objects; the web section index and size vary per test
+HEADERS_AND_SYMBOLS = """\
+File: {light}
+
+Section Headers:
+  [Nr] Name              Type            Addr     Off    Size   ES Flg Lk Inf Al
+  [124] .rodata._ZN7esphome5light10LightStateC2EPNS0_11LightOutputE.str1.1 PROGBITS        00000000 00067d 000001 01 AMS  0   0  1
+  [174] .rodata           PROGBITS        00000000 00139c 00002c 00   A  0   0  4
+  [221] .rodata.str1.1    PROGBITS        00000000 001f39 000005 01 AMS  0   0  1
+  [225] .rodata.BUILTIN_LED PROGBITS        00000000 001f54 000004 00   A  0   0  4
+
+Symbol table '.symtab' contains 348 entries:
+   Num:    Value  Size Type    Bind   Vis      Ndx Name
+     0: 00000000     0 NOTYPE  LOCAL  DEFAULT  UND
+   300: 00000000    64 FUNC    {bind:<6} DEFAULT  12 {setup}
+   301: 00000000    64 FUNC    {bind:<6} DEFAULT  13 {save}
+   347: 00000000     4 OBJECT  WEAK   DEFAULT  225 BUILTIN_LED
+
+File: {web}
+
+Section Headers:
+  [Nr] Name              Type            Addr     Off    Size   ES Flg Lk Inf Al
+  [{web_index:>3}] .rodata           PROGBITS        00000000 000400 {web_size:06x} 00   A  0   0  4
+
+Symbol table '.symtab' contains 1 entries:
+   Num:    Value  Size Type    Bind   Vis      Ndx Name
+     0: 00000000     0 NOTYPE  LOCAL  DEFAULT  UND
+{web_symbols}"""
+
+LIGHT_OBJECT_DUMP = f"""\
+Relocation section '.rela.text.{SETUP}' at offset 0x7928 contains 2 entries:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+00000000  00003a01 R_XTENSA_32            00000000   .rodata + 0
+00000094  00003b14 R_XTENSA_SLOT0_OP      00000000   .text.{SETUP} + 82
+
+Relocation section '.rela.text.{SAVE}' at offset 0x7d28 contains 1 entry:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+00000000  00003a01 R_XTENSA_32            00000000   .rodata + 0
+
+Relocation section '.rela.xt.prop' at offset 0x8000 contains 1 entry:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+000007d4  00003a01 R_XTENSA_32            00000000   .rodata + 0
+
+Hex dump of section '.rodata':
+  0x00000000 0000803f 0000803f 0000803f 0000803f ...?...?...?...?
+  0x00000010 0000803f 0000803f 0000803f 0000803f ...?...?...?...?
+  0x00000020 0000803f 00000000 00000000          ...?........
+"""
+
+WEB_OBJECT_DUMP = f"""\
+Relocation section '.rela.text.{MATCH_URL}' at offset 0x100 contains 1 entry:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+00000000  00003a01 R_XTENSA_32            00000000   .rodata + 0
+
+Hex dump of section '.rodata':
+  0x00000000 00000000 00000000 00000000 00000000 ................
+"""
+
+
+@pytest.fixture
+def obj_dir(tmp_path: Path) -> Path:
+    """Native ESP8266 build layout: objects under obj/, ELF next to it."""
+    for rel in (LIGHT_OBJ, WEB_OBJ, "src/main.cpp.o"):
+        path = tmp_path / "obj" / rel
+        path.parent.mkdir(parents=True, exist_ok=True)
+        path.touch()
+    return tmp_path / "obj"
+
+
+def _esp8266_analyzer(
+    obj_dir: Path,
+    kept: set[str],
+    web_index: int = 9,
+    web_size: int = 0x1C,
+    web_dump: str = WEB_OBJECT_DUMP,
+    bind: str = "GLOBAL",
+    web_symbols: str = "",
+) -> tuple[MemoryAnalyzer, Callable[..., subprocess.CompletedProcess[str]]]:
+    analyzer = make_analyzer()
+    analyzer.elf_path = obj_dir.parent / "firmware.elf"
+    analyzer.rodata_in_ram = True
+    analyzer.sections[".data"] = MemorySection(".data", total_size=1000)
+    analyzer._elf_symbol_names = kept
+    light = str(obj_dir / LIGHT_OBJ)
+    web = str(obj_dir / WEB_OBJ)
+
+    def run_tool(cmd: list[str]) -> subprocess.CompletedProcess[str]:
+        if "-SW" in cmd:
+            return completed(
+                HEADERS_AND_SYMBOLS.format(
+                    light=light,
+                    web=web,
+                    web_index=web_index,
+                    web_size=web_size,
+                    bind=bind,
+                    setup=SETUP,
+                    save=SAVE,
+                    web_symbols=web_symbols,
+                )
+            )
+        # Only sections with uncovered bytes are dumped
+        if cmd[-1] == light:
+            assert cmd[1:-1] == ["-rW", "-x", "174"]
+            return completed(LIGHT_OBJECT_DUMP)
+        return completed(web_dump)
+
+    return analyzer, run_tool
+
+
+def test_find_object_files_dir_native_layout(obj_dir: Path) -> None:
+    """The native ESP8266 build keeps objects under obj/."""
+    analyzer = make_analyzer()
+    analyzer.elf_path = obj_dir.parent / "firmware.elf"
+    assert analyzer._find_object_files_dir() == obj_dir
+
+
+def test_find_object_files_dir_platformio_layout(tmp_path: Path) -> None:
+    """PlatformIO keeps objects next to the ELF."""
+    (tmp_path / "src").mkdir()
+    (tmp_path / "src" / "main.cpp.o").touch()
+    analyzer = make_analyzer()
+    analyzer.elf_path = tmp_path / "firmware.elf"
+    assert analyzer._find_object_files_dir() == tmp_path
+
+
+def _parse_sections(code_section: str) -> MemoryAnalyzer:
+    readelf = f"""\
+  [ 1] .data             PROGBITS        3ffe8000 001000 000668 00 WAX  0   0 16
+  [ 4] {code_section}       PROGBITS        40201010 00b010 06612c 00 WAX  0   0 16
+  [ 6] .rodata           PROGBITS        3ffe86b0 0016b0 000b90 00   A  0   0 16
+"""
+    analyzer = make_analyzer()
+    with patch(
+        "esphome.analyze_memory.subprocess.run", return_value=completed(readelf)
+    ):
+        analyzer._parse_sections()
+    return analyzer
+
+
+def test_parse_sections_esp8266_counts_rodata_as_data() -> None:
+    """ESP8266 (.irom0.text) copies .rodata to RAM, so it is counted as .data."""
+    analyzer = _parse_sections(".irom0.text")
+    assert analyzer.rodata_in_ram
+    assert ".rodata" not in analyzer.sections
+    assert analyzer.sections[".data"].total_size == 0x668 + 0xB90
+
+
+def test_parse_sections_other_platforms_keep_rodata() -> None:
+    """Other platforms keep .rodata in flash."""
+    analyzer = _parse_sections(".flash.text")
+    assert not analyzer.rodata_in_ram
+    assert analyzer.sections[".rodata"].total_size == 0xB90
+    assert analyzer.sections[".data"].total_size == 0x668
+
+
+def test_unnamed_constants_attributed_to_component(obj_dir: Path) -> None:
+    """Unnamed .rodata in a kept function is listed and counted as the component's RAM."""
+    analyzer, run_tool = _esp8266_analyzer(obj_dir, {SETUP, MATCH_URL})
+    with patch("esphome.analyze_memory.run_tool", side_effect=run_tool):
+        analyzer._analyze_object_rodata()
+
+    light, web = analyzer._unnamed_constants
+    assert light == UnnamedConstant(
+        size=44,
+        source_file=LIGHT_OBJ,
+        component="[esphome]light",
+        functions=[SAVE, SETUP],
+        jump_table=False,
+        data=bytes.fromhex("0000803f") * 4,
+    )
+    assert (web.size, web.component, web.functions) == (
+        28,
+        "[esphome]web_server",
+        [MATCH_URL],
+    )
+    assert analyzer.components["[esphome]light"].data_size == 44
+    assert analyzer.components["[esphome]web_server"].ram_total == 28
+    # Attributed bytes no longer count as unattributed RAM
+    assert analyzer.sections[".data"].symbol_size == 72
+
+
+def test_unnamed_constants_skip_garbage_collected(obj_dir: Path) -> None:
+    """A constant whose functions the linker dropped is not in the firmware."""
+    analyzer, run_tool = _esp8266_analyzer(obj_dir, {MATCH_URL})
+    with patch("esphome.analyze_memory.run_tool", side_effect=run_tool):
+        analyzer._analyze_object_rodata()
+    assert [c.component for c in analyzer._unnamed_constants] == ["[esphome]web_server"]
+
+
+def test_unnamed_constants_skipped_when_rodata_in_flash(obj_dir: Path) -> None:
+    """Other platforms keep .rodata in flash, so there is nothing to scan."""
+    analyzer, run_tool = _esp8266_analyzer(obj_dir, {SETUP, MATCH_URL})
+    analyzer.rodata_in_ram = False
+    with patch("esphome.analyze_memory.run_tool", side_effect=run_tool) as mock:
+        analyzer._analyze_object_rodata()
+    mock.assert_not_called()
+    assert analyzer._unnamed_constants == []
+
+
+@pytest.mark.parametrize(("bind", "expected"), [("WEAK", 1), ("GLOBAL", 2)])
+def test_unnamed_constants_inline_copies_counted_once(
+    obj_dir: Path, bind: str, expected: int
+) -> None:
+    """Only weak (inline) functions bring the same constant from several objects."""
+    # The second object emits the same functions, with the same binding
+    analyzer, run_tool = _esp8266_analyzer(
+        obj_dir,
+        {SETUP},
+        web_index=174,
+        web_size=0x2C,
+        web_dump=LIGHT_OBJECT_DUMP,
+        bind=bind,
+        web_symbols=(
+            f"   300: 00000000    64 FUNC    {bind:<6} DEFAULT  12 {SETUP}\n"
+            f"   301: 00000000    64 FUNC    {bind:<6} DEFAULT  13 {SAVE}\n"
+        ),
+    )
+    with patch("esphome.analyze_memory.run_tool", side_effect=run_tool):
+        analyzer._analyze_object_rodata()
+    assert len(analyzer._unnamed_constants) == expected
+    assert analyzer.sections[".data"].symbol_size == 44 * expected
+
+
+def test_parse_symbols_keeps_alias_names() -> None:
+    """C1/C2 constructor aliases share an address; both names are in the ELF."""
+    objdump = """4020b558 g     F .irom0.text\t0000007e _ZN7esphome3FooC1Ev
+4020b558 g     F .irom0.text\t0000007e _ZN7esphome3FooC2Ev
+"""
+    analyzer = make_analyzer()
+    analyzer.sections[".text"] = MemorySection(".text")
+    with patch(
+        "esphome.analyze_memory.subprocess.run", return_value=completed(objdump)
+    ):
+        analyzer._parse_symbols()
+    assert {"_ZN7esphome3FooC1Ev", "_ZN7esphome3FooC2Ev"} <= analyzer._elf_symbol_names
+    assert analyzer.sections[".text"].symbol_size == 0x7E
+
+
+def test_unnamed_constants_skip_named_bytes() -> None:
+    """In a section shared with a named object, only the unnamed bytes are shown."""
+    analyzer = make_analyzer()
+    analyzer._elf_symbol_names = {"_Z3fooi"}
+    dump = """Relocation section '.rela.text._Z3fooi' at offset 0x100 contains 1 entry:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+00000000  00003a01 R_XTENSA_32            00000000   .rodata + 4
+
+Hex dump of section '.rodata':
+  0x00000000 aaaaaaaa 01020304                   ........
+"""
+    section = _RodataSection(".rodata", 8, covered=[(0, 4)])
+    with patch("esphome.analyze_memory.run_tool", return_value=completed(dump)):
+        (constant,) = analyzer._scan_object_constants(
+            "foo.o", {"5": section}, "src/foo.o", "[esphome]foo"
+        )
+    assert constant.size == 4
+    assert constant.data == bytes([1, 2, 3, 4])
+
+
+def test_unnamed_constants_jump_table() -> None:
+    """A .rodata section that points into code is a switch jump table."""
+    analyzer = make_analyzer()
+    analyzer._elf_symbol_names = {"_Z3fooi"}
+    dump = """\
+Relocation section '.rela.text._Z3fooi' at offset 0x100 contains 1 entry:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+00000000  00003a01 R_XTENSA_32            00000000   .rodata + 0
+
+Relocation section '.rela.rodata' at offset 0x200 contains 2 entries:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+00000000  00003b01 R_XTENSA_32            00000000   .text._Z3fooi + 12
+00000004  00003b01 R_XTENSA_32            00000000   .text._Z3fooi + 20
+"""
+    with patch("esphome.analyze_memory.run_tool", return_value=completed(dump)):
+        (constant,) = analyzer._scan_object_constants(
+            "foo.o", {"5": _RodataSection(".rodata", 8)}, "src/foo.o", "[esphome]foo"
+        )
+    assert constant.jump_table
+    assert constant.functions == ["_Z3fooi"]
+
+
+def test_report_lists_unnamed_constants() -> None:
+    """The report names the component, the object, the functions and the bytes."""
+    analyzer = make_analyzer(cls=MemoryAnalyzerCLI)
+    analyzer.rodata_in_ram = True
+    analyzer._demangle_cache = {SETUP: "esphome::light::LightState::setup()"}
+    analyzer._unnamed_constants = [
+        UnnamedConstant(
+            size=44,
+            source_file=LIGHT_OBJ,
+            component="[esphome]light",
+            functions=[SETUP],
+            jump_table=False,
+            data=bytes.fromhex("0000803f") * 4,
+        )
+    ]
+    report = analyzer.generate_report()
+    assert "ESP8266 keeps .rodata in RAM" in report
+    assert "Unnamed Constants in RAM (.rodata)" in report
+    assert f"44 B constant     [esphome]light ({LIGHT_OBJ})" in report
+    assert "used by esphome::light::LightState::setup()" in report
+    assert "bytes: 00 00 80 3f 00 00 80 3f 00 00 80 3f 00 00 80 3f ..." in report
+
+
+def test_unnamed_constants_resolve_string_pointers() -> None:
+    """A constant of pointers to literals lists the strings, so it can be recognized."""
+    analyzer = make_analyzer()
+    analyzer._elf_symbol_names = {"setup"}
+    relocations = """\
+Relocation section '.rela.text.setup' at offset 0x100 contains 1 entry:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+00000000  00003a01 R_XTENSA_32            00000000   .rodata + 0
+
+Relocation section '.rela.rodata' at offset 0xe6fc contains 3 entries:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+00000000  00000501 R_XTENSA_32            00000000   .rodata.str1.1 + 0
+00000004  00000501 R_XTENSA_32            00000000   .rodata.str1.1 + 0
+00000008  00000501 R_XTENSA_32            00000000   .rodata.str1.1 + 0
+
+Hex dump of section '.rodata':
+  0x00000000 00000000 0f000000 1e000000          ............
+"""
+    literals = hex_dump(
+        ".rodata.str1.1", b"0.pool.ntp.org\0001.pool.ntp.org\0002.pool.ntp.org\0"
+    )
+
+    def run_tool(cmd: list[str]) -> subprocess.CompletedProcess[str]:
+        if "-rW" not in cmd:
+            assert cmd[1:-1] == ["-x", ".rodata.str1.1"]
+            return completed(literals)
+        return completed(relocations)
+
+    with patch("esphome.analyze_memory.run_tool", side_effect=run_tool):
+        (constant,) = analyzer._scan_object_constants(
+            "main.cpp.o",
+            {"5": _RodataSection(".rodata", 12)},
+            "src/main.cpp.o",
+            "[esphome]core",
+        )
+    assert constant.strings == ["0.pool.ntp.org", "1.pool.ntp.org", "2.pool.ntp.org"]
+
+    cli = make_analyzer(cls=MemoryAnalyzerCLI)
+    cli._unnamed_constants = [constant]
+    assert (
+        'points to "0.pool.ntp.org", "1.pool.ntp.org", "2.pool.ntp.org"'
+        in cli.generate_report()
+    )
+
+
+@pytest.mark.parametrize(
+    ("section", "function"),
+    [
+        (".text._Z3foov", "_Z3foov"),
+        (".literal._Z3foov", "_Z3foov"),
+        (".irom0.text._Z3foov", "_Z3foov"),
+        (".text.startup._GLOBAL__sub_I_setup", "_GLOBAL__sub_I_setup"),
+        (".text.unlikely._Z3foov", "_Z3foov"),
+        (".iram.text._Z3foov", "_Z3foov"),
+        (".text.hot_path", "hot_path"),
+        (".text", None),
+    ],
+)
+def test_code_section_function(section: str, function: str | None) -> None:
+    """Placement prefixes and IRAM code still resolve to the function."""
+    match = _CODE_SECTION_PATTERN.match(section)
+    assert match is not None
+    assert match.group(1) == function
+
+
+def test_unnamed_constants_same_object_not_merged(obj_dir: Path) -> None:
+    """Two equal constants of one inline function in one object are both counted."""
+    analyzer, run_tool = _esp8266_analyzer(obj_dir, {SETUP}, bind="WEAK")
+    constant = UnnamedConstant(
+        size=8,
+        source_file=LIGHT_OBJ,
+        component="[esphome]light",
+        functions=[SAVE, SETUP],
+        jump_table=False,
+        data=b"",
+    )
+    light = str(obj_dir / LIGHT_OBJ)
+    with (
+        patch("esphome.analyze_memory.run_tool", side_effect=run_tool),
+        patch.object(
+            analyzer,
+            "_scan_object_constants",
+            side_effect=lambda path, *_: [constant, constant] if path == light else [],
+        ),
+    ):
+        analyzer._analyze_object_rodata()
+    assert len(analyzer._unnamed_constants) == 2
+
+
+def test_unnamed_constants_jump_table_only_from_unnamed_bytes() -> None:
+    """Code pointers in a named table at the start do not make the rest a jump table."""
+    analyzer = make_analyzer()
+    analyzer._elf_symbol_names = {"_Z3fooi"}
+    dump = """\
+Relocation section '.rela.text._Z3fooi' at offset 0x100 contains 1 entry:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+00000000  00003a01 R_XTENSA_32            00000000   .rodata + 4
+
+Relocation section '.rela.rodata' at offset 0x200 contains 1 entry:
+ Offset     Info    Type                Sym. Value  Symbol's Name + Addend
+00000000  00003b01 R_XTENSA_32            00000000   .text._Z3fooi + 12
+
+"""
+    section = _RodataSection(".rodata", 8, covered=[(0, 4)])
+    with patch("esphome.analyze_memory.run_tool", return_value=completed(dump)):
+        (constant,) = analyzer._scan_object_constants(
+            "foo.o", {"5": section}, "src/foo.o", "[esphome]foo"
+        )
+    assert not constant.jump_table
+
+
+def _definitions(rows: dict[str, list[tuple[str, str]]]) -> _FunctionDefinitions:
+    """Parse symbol rows (binding, function) per object, as readelf -sW prints them."""
+    output = "".join(
+        f"File: {path}\n"
+        + "".join(
+            f"   {i}: 00000000    64 FUNC    {bind:<6} DEFAULT   12 {name}\n"
+            for i, (bind, name) in enumerate(symbols)
+        )
+        for path, symbols in rows.items()
+    )
+    return _parse_unnamed_rodata(output)[1]
+
+
+def test_linked_copy_unknown_weak_overridden_by_strong() -> None:
+    """A strong definition replaces every weak copy, so their constants are not linked."""
+    definitions = _definitions(
+        {"a.o": [("WEAK", "_Z3foov")], "b.o": [("GLOBAL", "_Z3foov")]}
+    )
+    assert definitions.linked_copy_unknown("a.o", ["_Z3foov"])
+    assert not definitions.linked_copy_unknown("b.o", ["_Z3foov"])
+
+
+def test_linked_copy_unknown_same_named_locals() -> None:
+    """Local functions with one name in two objects can't be told apart in the ELF."""
+    definitions = _definitions(
+        {
+            "a.o": [("LOCAL", "_ZL3barv"), ("LOCAL", "_ZL3bazv")],
+            "b.o": [("LOCAL", "_ZL3barv")],
+        }
+    )
+    assert definitions.linked_copy_unknown("a.o", ["_ZL3barv"])
+    # Another, unique user decides it
+    assert not definitions.linked_copy_unknown("a.o", ["_ZL3barv", "_ZL3bazv"])