The update component had the same infinite loop issue as the OTA component
when network read errors occurred. If container->read() returned an error
(negative value), it would be added to read_index and the loop would continue
indefinitely since get_bytes_read() would never reach content_length.
This fix breaks out of the read loop on any read error (read_bytes <= 0),
preventing watchdog resets and infinite loops during manifest downloads.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
This commit fixes an issue where the http_request component would enter
an infinite loop when an HTTP server doesn't send a Content-Length header
or closes the connection prematurely.
The read loop was assuming read operations would always return data, but:
1. When the stream pointer becomes invalid (connection closed), read() returns -1
2. When no more data is available, read() returns 0
Without these checks, the loop would continue indefinitely, causing:
- "Stream pointer vanished!" errors (Arduino platform)
- CPU spinning on zero-byte reads
- Watchdog timeouts
The fix adds validation checks to break out of read loops when
read() returns <= 0 (covering both error and end-of-stream conditions).
This is applied to:
- Response capture loops (http_request.h)
- OTA firmware download loop (ota_http_request.cpp)
- MD5 verification download loop (ota_http_request.cpp)
This allows graceful handling of non-compliant HTTP servers while
maintaining compatibility with properly formatted responses.
Fixesesphome/issues#6682🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Fix polling_component_schema to use update_interval validator when
default_update_interval is None (was using None as validator)
- Replace hardcoded 4294967295 with SCHEDULER_DONT_RUN constant in
update_interval function
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
get_build_info() now always returns current time instead of preserving
the existing build_time. The timestamp preservation logic is now handled
in copy_src_tree() based on sources_changed flag.
- Make copy_file_if_changed() return bool indicating if file was copied
- Track sources_changed in copy_src_tree() to detect when source files change
- Only update build_info timestamp when sources/config/version change
- Exclude generated files (build_info_data.h) from sources_changed tracking
- Add build_info_data.h to ignore_targets to prevent copying from resources
- Track changes to generated headers (defines.h, esphome.h, version.h)
- Check for config_hash or version changes to trigger rebuild
- Pretty-print build_info.json with indentation and trailing newline
- Update mock_copy_file_if_changed to return True by default
This fixes the issue where changing a source file would recompile the .o
file but not relink the final program executable.
J. Nick Koston