v2.4.0

v2.4.0

Eighteen merges since v2.3.1. Four new tool families land — symbol library management, symbol property editing, lib_id replacement for library migration, and update-from-library refresh — alongside a process-wide caching layer for symbol discovery. Two fixes restore basic operation for whole classes of users: every .kicad_sym and schematic write was broken on Python 3.9, and JLCPCB part search could not find hyphenated MPNs.

Highlights

New tool families:

• Library management: import_symbol / export_symbol / rename_symbol (#317) -- copy a symbol between .kicad_sym libraries (with optional rename/overwrite; target created if missing), extract one symbol to a standalone file, and rename a symbol including its sub-symbol shards and any (extends ...) references from derived symbols in the same library
• Symbol property tools (#308) -- add_symbol_property sets custom BOM fields (Manufacturer, MPN, LCSC, ...) on a symbol in a .kicad_sym library file; add_library_symbol_property does the same on a schematic's cached definition
• update_symbol_from_library (#291) -- refresh cached lib_symbols definitions in one schematic, a list, or every project under a directory, preserving placed instances -- the programmatic equivalent of KiCad's Update Symbol from Library
• replace_instance_lib_ids (#316) -- swap lib_id references per an explicit old-to-new mapping for migrating schematics between libraries, with automatic angle correction for the Eagle importer's mirror-variant suffixes

Performance:

• Process-wide symbol discovery caches (#320) -- library directories, resolved paths, extracted symbol blocks, and parsed symbol lists were rebuilt on every component add, re-reading multi-MB .kicad_sym files each time; they are now cached process-wide with staleness guards (mtime_ns tracking, path revalidation, explicit invalidation on every mutating write)
• sync_schematic_to_board caches its LibraryManager (#310) -- no longer re-parses the fp-lib-table on every call

Bug fixes:

• Python 3.9 writes restored (#328) -- Path.write_text(newline=) is only accepted from Python 3.10, so on the project's declared 3.9 floor every library-management, Eagle-prettify, and symbol-schematic write raised TypeError; all sites now open the file handle explicitly, preserving forced LF line endings
• JLCPCB search finds hyphenated MPNs (#327) -- FTS5 reads - as a column/NOT operator, so searching a real MPN like SHT41-AD1F-R2 raised OperationalError, which a broad except swallowed into an empty result -- exact-MPN searches silently found nothing; each term is now emitted as a quoted prefix phrase so FTS punctuation matches as literal text
• Eagle import writes KiCad 10 headers (#330, closes #321) -- generated .kicad_sch files carried the KiCad 9 20250114 token; they now carry the canonical KiCad 10 header, verified by exporting the importer's output to PDF and running ERC with real kicad-cli 10.0
• Component placement snaps to the 1.27 mm grid (#319, closes #299) -- library pins sit at multiples of 1.27 mm from the symbol origin, so an off-grid origin left every pin off-grid: wires and net labels could not bind, ERC reported endpoint_off_grid, and the netlist came up empty
• import_ses no longer creates phantom slashless nets (#309, fixes #246) -- Freerouting strips the leading / from global net names; routed tracks now bind to the original nets
• export_dsn/autoroute keep .kicad_pro net classes (#306, fixes #302) -- power/ground clearance and width rules were silently dropped from the Specctra DSN
• _find_pins_on_net locates pins on rotated symbols (#304, fixes #303) -- used a local transform instead of world coordinates
• add_sheet_pin finds sheets regardless of line formatting (#307, fixes #298)
• Test-suite state pollution eliminated (#312, fixes #287) -- a leaked sys.modules entry caused order-dependent pin-position failures
• Freerouting runs headless (#313) -- autoroute no longer opens a GUI window

Tooling:

• pathlib migration, first slice (#254) -- kicad_interface.py and schematic_handlers.py now use pathlib.Path; note the mypy target is now 3.10, so the declared 3.9 floor is no longer type-checker-verified (the #328 crash is exactly the class of breakage this leaves to review)
• Interface construction smoke test -- a handler class that references an unimported function now fails tests instead of crashing the server at startup (#308 shipped exactly that)
• CHANGELOG lint (#322) -- inline code spans crossing line breaks tripped prettier's list-indentation logic on every reformat; a regression test now guards it

Issues closed

#246, #287, #298, #299, #302, #303, #321

Full details in CHANGELOG.md.