mirror of
https://github.com/esphome/esphome.git
synced 2025-11-08 10:28:40 +00:00
422 lines
15 KiB
Python
422 lines
15 KiB
Python
"""CLI interface for memory analysis with report generation."""
|
|
|
|
from collections import defaultdict
|
|
import sys
|
|
|
|
from . import (
|
|
_COMPONENT_API,
|
|
_COMPONENT_CORE,
|
|
_COMPONENT_PREFIX_ESPHOME,
|
|
_COMPONENT_PREFIX_EXTERNAL,
|
|
MemoryAnalyzer,
|
|
)
|
|
|
|
|
|
class MemoryAnalyzerCLI(MemoryAnalyzer):
|
|
"""Memory analyzer with CLI-specific report generation."""
|
|
|
|
# Column width constants
|
|
COL_COMPONENT: int = 29
|
|
COL_FLASH_TEXT: int = 14
|
|
COL_FLASH_DATA: int = 14
|
|
COL_RAM_DATA: int = 12
|
|
COL_RAM_BSS: int = 12
|
|
COL_TOTAL_FLASH: int = 15
|
|
COL_TOTAL_RAM: int = 12
|
|
COL_SEPARATOR: int = 3 # " | "
|
|
|
|
# Core analysis column widths
|
|
COL_CORE_SUBCATEGORY: int = 30
|
|
COL_CORE_SIZE: int = 12
|
|
COL_CORE_COUNT: int = 6
|
|
COL_CORE_PERCENT: int = 10
|
|
|
|
# Calculate table width once at class level
|
|
TABLE_WIDTH: int = (
|
|
COL_COMPONENT
|
|
+ COL_SEPARATOR
|
|
+ COL_FLASH_TEXT
|
|
+ COL_SEPARATOR
|
|
+ COL_FLASH_DATA
|
|
+ COL_SEPARATOR
|
|
+ COL_RAM_DATA
|
|
+ COL_SEPARATOR
|
|
+ COL_RAM_BSS
|
|
+ COL_SEPARATOR
|
|
+ COL_TOTAL_FLASH
|
|
+ COL_SEPARATOR
|
|
+ COL_TOTAL_RAM
|
|
)
|
|
|
|
@staticmethod
|
|
def _make_separator_line(*widths: int) -> str:
|
|
"""Create a separator line with given column widths.
|
|
|
|
Args:
|
|
widths: Column widths to create separators for
|
|
|
|
Returns:
|
|
Separator line like "----+---------+-----"
|
|
"""
|
|
return "-+-".join("-" * width for width in widths)
|
|
|
|
# Pre-computed separator lines
|
|
MAIN_TABLE_SEPARATOR: str = _make_separator_line(
|
|
COL_COMPONENT,
|
|
COL_FLASH_TEXT,
|
|
COL_FLASH_DATA,
|
|
COL_RAM_DATA,
|
|
COL_RAM_BSS,
|
|
COL_TOTAL_FLASH,
|
|
COL_TOTAL_RAM,
|
|
)
|
|
|
|
CORE_TABLE_SEPARATOR: str = _make_separator_line(
|
|
COL_CORE_SUBCATEGORY,
|
|
COL_CORE_SIZE,
|
|
COL_CORE_COUNT,
|
|
COL_CORE_PERCENT,
|
|
)
|
|
|
|
def generate_report(self, detailed: bool = False) -> str:
|
|
"""Generate a formatted memory report."""
|
|
components = sorted(
|
|
self.components.items(), key=lambda x: x[1].flash_total, reverse=True
|
|
)
|
|
|
|
# Calculate totals
|
|
total_flash = sum(c.flash_total for _, c in components)
|
|
total_ram = sum(c.ram_total for _, c in components)
|
|
|
|
# Build report
|
|
lines: list[str] = []
|
|
|
|
lines.append("=" * self.TABLE_WIDTH)
|
|
lines.append("Component Memory Analysis".center(self.TABLE_WIDTH))
|
|
lines.append("=" * self.TABLE_WIDTH)
|
|
lines.append("")
|
|
|
|
# Main table - fixed column widths
|
|
lines.append(
|
|
f"{'Component':<{self.COL_COMPONENT}} | {'Flash (text)':>{self.COL_FLASH_TEXT}} | {'Flash (data)':>{self.COL_FLASH_DATA}} | {'RAM (data)':>{self.COL_RAM_DATA}} | {'RAM (bss)':>{self.COL_RAM_BSS}} | {'Total Flash':>{self.COL_TOTAL_FLASH}} | {'Total RAM':>{self.COL_TOTAL_RAM}}"
|
|
)
|
|
lines.append(self.MAIN_TABLE_SEPARATOR)
|
|
|
|
for name, mem in components:
|
|
if mem.flash_total > 0 or mem.ram_total > 0:
|
|
flash_rodata = mem.rodata_size + mem.data_size
|
|
lines.append(
|
|
f"{name:<{self.COL_COMPONENT}} | {mem.text_size:>{self.COL_FLASH_TEXT - 2},} B | {flash_rodata:>{self.COL_FLASH_DATA - 2},} B | "
|
|
f"{mem.data_size:>{self.COL_RAM_DATA - 2},} B | {mem.bss_size:>{self.COL_RAM_BSS - 2},} B | "
|
|
f"{mem.flash_total:>{self.COL_TOTAL_FLASH - 2},} B | {mem.ram_total:>{self.COL_TOTAL_RAM - 2},} B"
|
|
)
|
|
|
|
lines.append(self.MAIN_TABLE_SEPARATOR)
|
|
lines.append(
|
|
f"{'TOTAL':<{self.COL_COMPONENT}} | {' ':>{self.COL_FLASH_TEXT}} | {' ':>{self.COL_FLASH_DATA}} | "
|
|
f"{' ':>{self.COL_RAM_DATA}} | {' ':>{self.COL_RAM_BSS}} | "
|
|
f"{total_flash:>{self.COL_TOTAL_FLASH - 2},} B | {total_ram:>{self.COL_TOTAL_RAM - 2},} B"
|
|
)
|
|
|
|
# Top consumers
|
|
lines.append("")
|
|
lines.append("Top Flash Consumers:")
|
|
for i, (name, mem) in enumerate(components[:25]):
|
|
if mem.flash_total > 0:
|
|
percentage = (
|
|
(mem.flash_total / total_flash * 100) if total_flash > 0 else 0
|
|
)
|
|
lines.append(
|
|
f"{i + 1}. {name} ({mem.flash_total:,} B) - {percentage:.1f}% of analyzed flash"
|
|
)
|
|
|
|
lines.append("")
|
|
lines.append("Top RAM Consumers:")
|
|
ram_components = sorted(components, key=lambda x: x[1].ram_total, reverse=True)
|
|
for i, (name, mem) in enumerate(ram_components[:25]):
|
|
if mem.ram_total > 0:
|
|
percentage = (mem.ram_total / total_ram * 100) if total_ram > 0 else 0
|
|
lines.append(
|
|
f"{i + 1}. {name} ({mem.ram_total:,} B) - {percentage:.1f}% of analyzed RAM"
|
|
)
|
|
|
|
lines.append("")
|
|
lines.append(
|
|
"Note: This analysis covers symbols in the ELF file. Some runtime allocations may not be included."
|
|
)
|
|
lines.append("=" * self.TABLE_WIDTH)
|
|
|
|
# Add ESPHome core detailed analysis if there are core symbols
|
|
if self._esphome_core_symbols:
|
|
lines.append("")
|
|
lines.append("=" * self.TABLE_WIDTH)
|
|
lines.append(
|
|
f"{_COMPONENT_CORE} Detailed Analysis".center(self.TABLE_WIDTH)
|
|
)
|
|
lines.append("=" * self.TABLE_WIDTH)
|
|
lines.append("")
|
|
|
|
# Group core symbols by subcategory
|
|
core_subcategories: dict[str, list[tuple[str, str, int]]] = defaultdict(
|
|
list
|
|
)
|
|
|
|
for symbol, demangled, size in self._esphome_core_symbols:
|
|
# Categorize based on demangled name patterns
|
|
subcategory = self._categorize_esphome_core_symbol(demangled)
|
|
core_subcategories[subcategory].append((symbol, demangled, size))
|
|
|
|
# Sort subcategories by total size
|
|
sorted_subcategories = sorted(
|
|
[
|
|
(name, symbols, sum(s[2] for s in symbols))
|
|
for name, symbols in core_subcategories.items()
|
|
],
|
|
key=lambda x: x[2],
|
|
reverse=True,
|
|
)
|
|
|
|
lines.append(
|
|
f"{'Subcategory':<{self.COL_CORE_SUBCATEGORY}} | {'Size':>{self.COL_CORE_SIZE}} | "
|
|
f"{'Count':>{self.COL_CORE_COUNT}} | {'% of Core':>{self.COL_CORE_PERCENT}}"
|
|
)
|
|
lines.append(self.CORE_TABLE_SEPARATOR)
|
|
|
|
core_total = sum(size for _, _, size in self._esphome_core_symbols)
|
|
|
|
for subcategory, symbols, total_size in sorted_subcategories:
|
|
percentage = (total_size / core_total * 100) if core_total > 0 else 0
|
|
lines.append(
|
|
f"{subcategory:<{self.COL_CORE_SUBCATEGORY}} | {total_size:>{self.COL_CORE_SIZE - 2},} B | "
|
|
f"{len(symbols):>{self.COL_CORE_COUNT}} | {percentage:>{self.COL_CORE_PERCENT - 1}.1f}%"
|
|
)
|
|
|
|
# Top 15 largest core symbols
|
|
lines.append("")
|
|
lines.append(f"Top 15 Largest {_COMPONENT_CORE} Symbols:")
|
|
sorted_core_symbols = sorted(
|
|
self._esphome_core_symbols, key=lambda x: x[2], reverse=True
|
|
)
|
|
|
|
for i, (symbol, demangled, size) in enumerate(sorted_core_symbols[:15]):
|
|
lines.append(f"{i + 1}. {demangled} ({size:,} B)")
|
|
|
|
lines.append("=" * self.TABLE_WIDTH)
|
|
|
|
# Add detailed analysis for top ESPHome and external components
|
|
esphome_components = [
|
|
(name, mem)
|
|
for name, mem in components
|
|
if name.startswith(_COMPONENT_PREFIX_ESPHOME) and name != _COMPONENT_CORE
|
|
]
|
|
external_components = [
|
|
(name, mem)
|
|
for name, mem in components
|
|
if name.startswith(_COMPONENT_PREFIX_EXTERNAL)
|
|
]
|
|
|
|
top_esphome_components = sorted(
|
|
esphome_components, key=lambda x: x[1].flash_total, reverse=True
|
|
)[:30]
|
|
|
|
# Include all external components (they're usually important)
|
|
top_external_components = sorted(
|
|
external_components, key=lambda x: x[1].flash_total, reverse=True
|
|
)
|
|
|
|
# Check if API component exists and ensure it's included
|
|
api_component = None
|
|
for name, mem in components:
|
|
if name == _COMPONENT_API:
|
|
api_component = (name, mem)
|
|
break
|
|
|
|
# Also include wifi_stack and other important system components if they exist
|
|
system_components_to_include = [
|
|
# Empty list - we've finished debugging symbol categorization
|
|
# Add component names here if you need to debug their symbols
|
|
]
|
|
system_components = [
|
|
(name, mem)
|
|
for name, mem in components
|
|
if name in system_components_to_include
|
|
]
|
|
|
|
# Combine all components to analyze: top ESPHome + all external + API if not already included + system components
|
|
components_to_analyze = (
|
|
list(top_esphome_components)
|
|
+ list(top_external_components)
|
|
+ system_components
|
|
)
|
|
if api_component and api_component not in components_to_analyze:
|
|
components_to_analyze.append(api_component)
|
|
|
|
if components_to_analyze:
|
|
for comp_name, comp_mem in components_to_analyze:
|
|
if not (comp_symbols := self._component_symbols.get(comp_name, [])):
|
|
continue
|
|
lines.append("")
|
|
lines.append("=" * self.TABLE_WIDTH)
|
|
lines.append(f"{comp_name} Detailed Analysis".center(self.TABLE_WIDTH))
|
|
lines.append("=" * self.TABLE_WIDTH)
|
|
lines.append("")
|
|
|
|
# Sort symbols by size
|
|
sorted_symbols = sorted(comp_symbols, key=lambda x: x[2], reverse=True)
|
|
|
|
lines.append(f"Total symbols: {len(sorted_symbols)}")
|
|
lines.append(f"Total size: {comp_mem.flash_total:,} B")
|
|
lines.append("")
|
|
|
|
# Show all symbols > 100 bytes for better visibility
|
|
large_symbols = [
|
|
(sym, dem, size) for sym, dem, size in sorted_symbols if size > 100
|
|
]
|
|
|
|
lines.append(
|
|
f"{comp_name} Symbols > 100 B ({len(large_symbols)} symbols):"
|
|
)
|
|
for i, (symbol, demangled, size) in enumerate(large_symbols):
|
|
lines.append(f"{i + 1}. {demangled} ({size:,} B)")
|
|
|
|
lines.append("=" * self.TABLE_WIDTH)
|
|
|
|
return "\n".join(lines)
|
|
|
|
def dump_uncategorized_symbols(self, output_file: str | None = None) -> None:
|
|
"""Dump uncategorized symbols for analysis."""
|
|
# Sort by size descending
|
|
sorted_symbols = sorted(
|
|
self._uncategorized_symbols, key=lambda x: x[2], reverse=True
|
|
)
|
|
|
|
lines = ["Uncategorized Symbols Analysis", "=" * 80]
|
|
lines.append(f"Total uncategorized symbols: {len(sorted_symbols)}")
|
|
lines.append(
|
|
f"Total uncategorized size: {sum(s[2] for s in sorted_symbols):,} bytes"
|
|
)
|
|
lines.append("")
|
|
lines.append(f"{'Size':>10} | {'Symbol':<60} | Demangled")
|
|
lines.append("-" * 10 + "-+-" + "-" * 60 + "-+-" + "-" * 40)
|
|
|
|
for symbol, demangled, size in sorted_symbols[:100]: # Top 100
|
|
demangled_display = (
|
|
demangled[:100] if symbol != demangled else "[not demangled]"
|
|
)
|
|
lines.append(f"{size:>10,} | {symbol[:60]:<60} | {demangled_display}")
|
|
|
|
if len(sorted_symbols) > 100:
|
|
lines.append(f"\n... and {len(sorted_symbols) - 100} more symbols")
|
|
|
|
content = "\n".join(lines)
|
|
|
|
if output_file:
|
|
with open(output_file, "w", encoding="utf-8") as f:
|
|
f.write(content)
|
|
else:
|
|
print(content)
|
|
|
|
|
|
def analyze_elf(
|
|
elf_path: str,
|
|
objdump_path: str | None = None,
|
|
readelf_path: str | None = None,
|
|
detailed: bool = False,
|
|
external_components: set[str] | None = None,
|
|
) -> str:
|
|
"""Analyze an ELF file and return a memory report."""
|
|
analyzer = MemoryAnalyzerCLI(
|
|
elf_path, objdump_path, readelf_path, external_components
|
|
)
|
|
analyzer.analyze()
|
|
return analyzer.generate_report(detailed)
|
|
|
|
|
|
def main():
|
|
"""CLI entrypoint for memory analysis."""
|
|
if len(sys.argv) < 2:
|
|
print("Usage: python -m esphome.analyze_memory <build_directory>")
|
|
print("\nAnalyze memory usage from an ESPHome build directory.")
|
|
print("The build directory should contain firmware.elf and idedata will be")
|
|
print("loaded from ~/.esphome/.internal/idedata/<device>.json")
|
|
print("\nExamples:")
|
|
print(" python -m esphome.analyze_memory ~/.esphome/build/my-device")
|
|
print(" python -m esphome.analyze_memory .esphome/build/my-device")
|
|
print(" python -m esphome.analyze_memory my-device # Short form")
|
|
sys.exit(1)
|
|
|
|
build_dir = sys.argv[1]
|
|
|
|
# Load build directory
|
|
import json
|
|
from pathlib import Path
|
|
|
|
from esphome.platformio_api import IDEData
|
|
|
|
build_path = Path(build_dir)
|
|
|
|
# If no path separator in name, assume it's a device name
|
|
if "/" not in build_dir and not build_path.is_dir():
|
|
# Try current directory first
|
|
cwd_path = Path.cwd() / ".esphome" / "build" / build_dir
|
|
if cwd_path.is_dir():
|
|
build_path = cwd_path
|
|
print(f"Using build directory: {build_path}", file=sys.stderr)
|
|
else:
|
|
# Fall back to home directory
|
|
build_path = Path.home() / ".esphome" / "build" / build_dir
|
|
print(f"Using build directory: {build_path}", file=sys.stderr)
|
|
|
|
if not build_path.is_dir():
|
|
print(f"Error: {build_path} is not a directory", file=sys.stderr)
|
|
sys.exit(1)
|
|
|
|
# Find firmware.elf
|
|
elf_file = None
|
|
for elf_candidate in [
|
|
build_path / "firmware.elf",
|
|
build_path / ".pioenvs" / build_path.name / "firmware.elf",
|
|
]:
|
|
if elf_candidate.exists():
|
|
elf_file = str(elf_candidate)
|
|
break
|
|
|
|
if not elf_file:
|
|
print(f"Error: firmware.elf not found in {build_dir}", file=sys.stderr)
|
|
sys.exit(1)
|
|
|
|
# Find idedata.json - check current directory first, then home
|
|
device_name = build_path.name
|
|
idedata_candidates = [
|
|
Path.cwd() / ".esphome" / "idedata" / f"{device_name}.json",
|
|
Path.home() / ".esphome" / "idedata" / f"{device_name}.json",
|
|
]
|
|
|
|
idedata = None
|
|
for idedata_path in idedata_candidates:
|
|
if not idedata_path.exists():
|
|
continue
|
|
try:
|
|
with open(idedata_path, encoding="utf-8") as f:
|
|
raw_data = json.load(f)
|
|
idedata = IDEData(raw_data)
|
|
print(f"Loaded idedata from: {idedata_path}", file=sys.stderr)
|
|
break
|
|
except (json.JSONDecodeError, OSError) as e:
|
|
print(f"Warning: Failed to load idedata: {e}", file=sys.stderr)
|
|
|
|
if not idedata:
|
|
print(
|
|
f"Warning: idedata not found (searched {idedata_candidates[0]} and {idedata_candidates[1]})",
|
|
file=sys.stderr,
|
|
)
|
|
|
|
analyzer = MemoryAnalyzerCLI(elf_file, idedata=idedata)
|
|
analyzer.analyze()
|
|
report = analyzer.generate_report()
|
|
print(report)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|