PR Author 983d6a1566 Support bounded feedback loops in the DAG execution engine ... Allow sampler nodes' internal iteration variables (e.g. step_index) to flow back upstream through ComfyMathExpression nodes to control per-step parameters (cfg, s_noise, eta, r) without triggering a dependency cycle error. Architecture: Two-level cycle handling - Static validation: _is_bounded_feedback_cycle() allows cycles where any node declares BOUNDED_FEEDBACK - Graph building: _is_feedback_output() skips strong links for declared feedback sockets, records them in feedback_links Multi-hop chain walking via _build_feedback_fns() resolves expression->CFGGuider/Sampler chains with simple_eval + MATH_FUNCTIONS, composing per-step fn(step, total_steps) callables. Sampler functions now re-read s_noise/eta/r each iteration via _init_dynamic_options() / _refresh_dynamic_params() / _apply_dynamic_s_noise(). KSAMPLER.sample() conditionally injects mutable extra_options ref. Safety: _dynamic_sampler_options popped at function top before model() calls. One-line opt-in: BOUNDED_FEEDBACK = {'step_index'} on any node.		2026-06-19 18:56:13 +08:00
..
deis.py	Lint all unused variables (#5989)	2024-12-12 17:59:16 -05:00
sa_solver.py	Add SA-Solver sampler (#8834)	2025-07-08 16:17:06 -04:00
sampling.py	Support bounded feedback loops in the DAG execution engine	2026-06-19 18:56:13 +08:00
utils.py	Remove useless code.	2023-06-13 02:40:58 -04:00