Speed up automatic module pin handling

[julianspeith]

Assigning gates to a module recomputes its pins, and both halves of that were slower than they needed to be: the lookup that decides whether a net already has a pin scanned linearly, and every individual pin change was announced to the GUI as its own event.

Measured on a sha256 netlist with 7144 top level gates, creating and deleting 24 modules emitted 43040 pin_changed events. Because every event a Python script triggers is handed to the GUI thread over a blocking queued connection, the cost of an event is a thread round trip rather than the handler itself. A sample of that workload showed 91% of the script runtime was the Python thread waiting — 68% in the qApp->processEvents() that follows each emit, 23% blocked on the semaphore of the blocking connection. The netlist work itself was 5%.

What is in here

Characterization tests first. enable_automatic_net_checks() and update_nets() had no test at all, and the existing check_pins test covered the manual pin API but almost none of the automatic path. Six tests pin down the behavior that has to survive the change: pins appearing and disappearing as gates enter and leave a module, direction transitions through input/inout/output with a stable pin ID, pins propagating to ancestor modules and moving on re-parenting, the effect of marking a net global, and what happens with the automatic checks switched off.

Index module pins by net. get_pin_by_net() scanned all pins of a module, on the hot path of Module::check_net(), which runs for every net of every gate entering or leaving a module. The net of a pin never changes, so the index only needs maintaining when a pin is created or deleted.

Coalesce the events of bulk operations. A bulk operation changes a module's pins wholesale and a listener has to re-read them anyway, so PinChangedBulkScope collects the pin events of all affected modules and sends a single PinEvent::PinsReload per module. That takes the workload above from 43040 to 48 pin events.

Interactive pin changes such as renaming or reordering a pin group are deliberately not covered by a bulk scope and keep their fine grained events, so the pins tree still updates incrementally while the user edits pins. ModulePinsTree preserves expanded pin groups and the current selection across a reload, so a bulk change no longer collapses the tree.

Notes for review

• The new PinEvent::PinsReload widens the pin event encoding; the 4-bit packing is now guarded by a static_assert.
• The event coalescing is opt-in per call site, so nothing changes for code that does not open a scope.

🤖 Generated with Claude Code