refactor: reorganize the partition module by codepath - #940
Merged
Conversation
- Move dense, sparse, discrete state types under partition/storage/ so the representation payloads cluster instead of sorting apart from the code that consumes them. - Rename discrete_states to storage::discrete for a consistent module name.
…odule - Move optimization contribution, dense coefficients, and sparse coefficients under partition/optimize/ so the optimization concern reads as one unit. - Qualify consumers through the parent optimize module for a stable reference path.
- Move the Fitch partition into fitch/partition and its config into fitch/config so the parsimony codepath reads as one unit.
- Move the Augur ancestral node-data serialization out of the partition root into partition/io/ so output serialization is separate from partition math.
…codepath - Move the dense/discrete marginal pass framework and the all-codepath log-likelihood helpers into marginal/shared so both representations import one framework module. - Merge the single-consumer sparse chain (former marginal_sparse, marginal_passes, marginal_helpers) into a marginal/sparse directory with execution-ordered stage files: partition, message, pass, reconstruct, count, reroot. - Relocate normalize_1d_inplace to the sparse message stage where its only callers live.
…odule - Split the dense marginal partition into a marginal/dense directory: partition holds the type and its trait surface, pass holds the message-passing hooks. - Move the discrete marginal partition to marginal/discrete as a single cohesive module. - Both representations now sit beside the shared framework and sparse codepath under marginal/.
- Move the shared-framework tests beside marginal/shared, the sparse partition and transition-counting tests beside marginal/sparse, and leave the indexed-pass tests at the partition root. - Rename the former marginal_core test module to marginal_shared to match its module.
- Describe the reorganized partition submodules: storage, marginal codepaths, fitch, optimize, and io.
… submodules Group the flat shared module by concern: `data` holds `MarginalData` and the partition traits, `pass` holds the indexed backward and forward passes, and `normalize` holds the profile normalization and cavity-message helpers. Drop the non-indexed `marginal_process_node_backward`/`_forward` functions, which had no callers in the workspace.
…omment submodules Separate the discrete partition type and its trait impls from the leaf-profile setup (`input`) and the Newick comment provider (`comment`).
…rd submodules Separate the pass drivers from the per-node backward and forward reconstruction so each traversal direction reads on its own.
Move the generic parallel DAG scheduler (`run_dependency_queue`, `validate_dependency_graph`, and its worker pool) into its own module. The engine operates only on index-based prerequisites and successors, so isolating it lets a reader trace the indexed pass without crossing the concurrency machinery.
Marginal inference runs entirely through the indexed pass, so the `MarginalPartition::leaf_profile`, `backward_internal_pre`, and `forward_post` methods had no call sites. Drop them from the trait and from the dense and discrete implementations.
…directory Move the propagate_raw_per_site tests out of a per-file message submodule into sparse/__tests__, alongside the other sparse tests, and rename them to the test_message_* convention.
Make `partition::timetree` a directory module matching sibling partitions, with the dispatch enum grouped by operational concern (branch, optimize, marginal) so each capability is traced in one file.
…ules Each pass orchestrator drove a single per-node worker in a sibling module and had a single caller, forming a single-consumer chain across three files. Colocating each orchestrator with its worker lets a reader trace a pass end to end in one file and makes the worker a private helper.
…-graph - Relocate the payload-agnostic dependency-queue scheduler and the graph-generic indexed pass beside the frontier traversals they complement, so callers depend on the lower, more stable crate. - Move the pass tests to treetime-graph with crate-local fixtures.
… at the sparse root The backward and forward marginal pass modules live directly under the sparse module, with no intermediate pass subdirectory, so the partition dispatch calls them without an extra path segment.
ivan-aksamentov
changed the title
6 tasks
Base automatically changed from
feat/timetree-ancestral-sequence-output
to
rust
August 31, 2026 17:40
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Update:
Follow-up to: #939
The partition module had grown into a flat directory where representation storage, marginal reconstruction, optimization, parsimony, and serialization all sat side by side, and several filenames hid which codepath they served. The sparse marginal reconstruction in particular was scattered across four files whose names did not reveal that they were sparse-only, so tracing one reconstruction from its caller meant opening a shifting set of unrelated files.
This change groups the module by concern and, within marginal reconstruction, by codepath. Representation payloads move under
partition/storage/. Marginal reconstruction moves underpartition/marginal/, split into a shared pass framework [src] and one directory per representation:sparse/with execution-ordered stages (partition, message, pass, reconstruct, count, reroot),dense/, and a singlediscretemodule. Branch-length optimization, Fitch parsimony, and Augur serialization move underoptimize/,fitch/, andio/. Tests move beside the codepath they exercise. Module paths now read general to specific, so a filename or path names its codepath.The single-consumer sparse chain (formerly three separate modules) is merged into the sparse directory as sequential pipeline stages rather than kept as remote modules, so following the sparse reconstruction is a linear descent through one directory instead of a search across the module. No behavior changes: the full test suite passes unchanged.
Work items
partition/storage/(dense, sparse, discrete) [src]partition/marginal/[src]marginal/sparse/[src]optimize/,fitch/, andio/[src]Possible improvements