Modular branch specific defaults (#14234)
* init
* Ensure branch-specific input defaults in ConditionalPipelineBlocks
When sibling blocks of a ConditionalPipelineBlocks declare different
defaults for the same input, combine_inputs now merges the default to
None and records the per-block defaults in a new
InputParam.defaults_by_block field, instead of silently promoting the
non-None default. get_block_state falls back to the block's own declared
default when the state value is None, so each branch resolves its own
default when it actually runs and a branch using None as a "user didn't
pass this" sentinel is no longer polluted by a sibling's default.
Docstrings render conflicted defaults as e.g. "defaults to None or 189,
depending on the workflow".
Fixes #14132
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Remove stale note about kwargs_type removal — the feature is staying
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Clarify get_block_state default-resolution comment, drop getattr guards
defaults_by_block is a dataclass field so it always exists; the checks
are already inside the hasattr(param, "required") InputParam guard.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Add example comments to combine_inputs and formatters, revert description block restructure
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Comment the duplicate-input path in combine_inputs with the strength example
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Rework branch-defaults tests to reuse the image block fixtures
strength on img2img (0.3) / inpaint (0.9999), text2img as the None
sentinel branch; drop the separate video fixtures. Also add examples to
the combine_inputs accumulator comments.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Simplify branch-defaults tests: drop t2i sentinel and resolved_strength output
Assert on state.get("strength") directly (write-back records the branch's
resolved value); cover None-vs-non-None merge with a direct combine_inputs
test instead.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Move combine_inputs tests to module level as test_combine_inputs_*
block-a/block-b names, whole-list assertions, plus new coverage:
first-occurrence-wins, disagreement records every block, direct nested
prefixing.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Tighten merge assertions: unique strength param + full InputParam equality
Also group combine_inputs tests under TestConditionalBlocksInputs.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Assert workflow in runtime tests, fold doc rendering into merge tests
workflow is now a declared intermediate output on the fixture blocks so
each runtime test shows which branch ran; the merge tests also assert
the rendered docstring (nested case covers distinct-value dedupe).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* docs: declare defaults in InputParam, not inside __call__
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* docs: side-by-side default example in modular.md, test reference in combine_inputs
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Fix doc-builder formatting; regenerate cosmos3 use_system_prompt docstring
The combine_inputs docstring example is now a fenced code block so
doc-builder doesn't reflow it.
The cosmos3 use_system_prompt docstring change reflects a real behavior
fix this PR makes: the transfer/action branches declare default=True
while the standard text branch declares default=None as a sentinel that
falls back to the default_use_system_prompt config. Previously the
merged default True leaked into state before branch selection, so the
config fallback never fired for the standard workflow; each branch now
resolves its own default and the config is honored again.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Regenerate cosmos3 auto docstrings
Mechanical output of utils/modular_auto_docstring.py for the cosmos
files this PR touches; regeneration of the remaining stale pipelines
moved to its own PR.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
