0.9.0 release working PR#256
Open
sanghoonio wants to merge 37 commits into
Open
Conversation
Enables JS/TS consumers to build a RegionSetList from individual RegionSet objects, unlocking batch pairwiseJaccard() for uploaded collections in bedbase-ui. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds RegionSetListOps trait in gtars-genomicdist with indexed pair operations (pintersect_at, jaccard_at, union_at, setdiff_at, region_count, union_except) so all bindings (wasm, Python, R) can operate on pairs by index without cloning across the FFI boundary. Also adds RegionSetList.fromEntries() to wasm bindings for zero-clone construction directly from BED entry arrays. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add LolaColumnar struct and results_to_columns() in gtars-lola/output.rs so all bindings share a single row→column pivot implementation - Simplify WASM, Python, and R bindings to use results_to_columns() instead of copy-pasting the 22-field iteration (~100 lines each) - Remove empty_to_none/empty_to_na helpers from each binding (now in core) - Remove greet() scaffolding from gtars-wasm/lib.rs - Remove unused write_results_to_file() from gtars-lola/output.rs Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add bulk_union_except() using prefix/suffix arrays: O(n) unions instead of O(n²) for computing per-file unique region counts - Add union_all() and intersect_all() fold methods - Add wasm bindings: bulkUnionExcept (returns union stats + per-file unique counts in a single call), unionAll, intersectAll Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…etListOps and results_to_columns intersect_all() was incorrectly using pintersect (positional pairing) instead of intersect (range-level), producing wrong results when sets have different numbers of regions. Added 21 tests covering all RegionSetListOps trait methods and 3 tests for results_to_columns. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…ze suffix construction - Always push a name in JsRegionSetList::add() to keep names/region_sets aligned - Return None from union_except when skip >= n instead of silently ignoring - Build bulk_union_except suffix array incrementally to avoid redundant clones Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The refactor in eb8ef5a removed this function but r_regiondb_anno still depends on it for converting empty annotation strings to R NA values. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace empty-string-as-sentinel pattern with Option<String> for all annotation fields (cell_type, description, tissue, data_source, antibody, treatment, collection). This eliminates empty_to_na/empty_to_none helpers from all bindings — R gets NA, Python gets None, and WASM gets null natively from the type system. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add RegionSetList bindings and indexed operations
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
WASM builds from the local workspace, not crates.io, so it doesn't need publish-all-crates to succeed first. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Streaming bed
The Rust core genomicdist crate already has both region_distribution_with_bins
and region_distribution_with_chrom_sizes, but only R binding exposed the
chrom_sizes variant. Without reference-derived per-chrom bin sizes, every BED
file gets its own bin width (from max observed coordinate), making outputs
incomparable across files. Server-side SQL aggregation is impossible under
this scheme.
This wires the chrom_sizes variant through CLI, Python, and WASM so that
bin widths are a property of the reference genome, not the BED file.
Changes per binding:
CLI: --chrom-sizes dispatches to with_chrom_sizes. When absent, prints
loud warning to stderr noting that outputs won't be comparable
across files. Existing --chrom-sizes flag (previously only used
for partitions/promoter trimming) now also controls region_dist.
Python: distribution() accepts optional chrom_sizes=None kwarg.
WASM: regionDistribution(n_bins, chrom_sizes) dispatches on
null/undefined. Breaking change for the JS signature.
R: already supported, but updated to handle new result type.
Bonus fix: region_distribution_with_chrom_sizes had a latent bug where
regions whose midpoint fell beyond the stated chromosome size produced
invalid bins (end < start, rid >= n_bins). This happens with assembly
mismatches (e.g. hg38 BED paired with hg19 chrom_sizes). Fixed by:
- Changing return type to RegionDistResult { bins, out_of_range }
- Skipping regions with mid >= chrom_size, tracking per-chrom counts
- CLI warns with per-chrom breakdown when out_of_range is non-empty
- Python/WASM/R surface out_of_range alongside bins
Tests added:
- gtars-genomicdist: 2 new tests covering out-of-range skip + happy path
- gtars-python: 3 new tests covering new result shape + chrom_sizes kwarg
Test results (all green):
cargo test -p gtars-genomicdist: 262 passed (260 original + 2 new)
pytest tests/ (gtars-python): 147 passed (146 original + 1 new)
wasm-pack test --node: 2 passed
CLI smoke: dispatch + warning confirmed on hg38 BED + hg19 chrom_sizes
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Cross-binding alignment for genomicdist functions: - Add ignore_unk_chroms parameter to calc_dinucl_freq (core, Python, R, CLI), matching calc_gc_content's existing behavior for alt chromosomes - Fold calc_dinucl_freq variants into single R-aligned function with raw_counts and ignore_unk_chroms parameters - Align region_distribution to return RegionDistResult with out_of_range tracking across Python, R, and WASM bindings - Simplify Python neighbor_distances/nearest_neighbors (remove dead Option<> wrapping) - Update R type casts, test coverage, and rextendr-generated wrappers Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Set default features to include all subcommands so cargo install
produces a functional binary (previously default = [] gave an empty
CLI with no subcommands visible in --help)
- Fix overlaprs description ("Tokenize data into a universe" → accurate
description of overlap-based tokenization)
- Add --query and --universe long-form flags to overlaprs (previously
only -q/-u short forms existed)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Introduces the .fab binary FASTA format — sequences stored contiguously without line wrapping, enabling mmap + zero-copy &[u8] slices. This gives both instant construction (~ms) and fast per-region access, replacing the tradeoff between HashMap (fast access, slow load) and line-wrapped mmap (fast load, slow access due to newline filtering). Benchmarked at 83.9s for 30 files vs 151.2s with HashMap-per-subprocess. The .fab format eliminates the need for a mixed CLI+Python backend. Core changes: - BinaryGenomeAssembly: mmap-backed .fab reader with SequenceAccess impl - BinaryGenomeAssembly::write_from_fasta: one-time FASTA → .fab conversion - SequenceAccess trait: calc_gc_content/calc_dinucl_freq accept either GenomeAssembly (HashMap fallback) or BinaryGenomeAssembly (.fab) - memmap2 dependency added CLI changes: - `gtars prep --fasta` command creates .fab files - `gtars genomicdist --fasta` auto-detects .fab vs .fa by extension - --fasta help text updated to mention .fab format - --dinucl-freq made opt-in (expensive for wide regions) Binding changes: - Python: BinaryGenomeAssembly exposed, calc_gc_content/calc_dinucl_freq accept either type via PyAny dispatch - R: load_binary_genome_assembly() exposed, with_assembly! macro handles both types via dyn SequenceAccess - Type stubs updated (Union types, missing ignore_unk_chroms parameter) Tests: 265 total (262 existing + 3 new .fab round-trip/bad-magic/GC-parity) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Binary FASTA format, binding alignment, CLI UX
…osomes Previously each chromosome used bin_size = chrom_size / n_bins, giving every chromosome 250 bins regardless of size. This mismatched the compression and UI layers which expect proportional bins (longest chrom gets n_bins, shorter ones get fewer). The fix computes a uniform bin_width from the longest chromosome so that bin index N maps to the same genomic position across all files for a given genome, enabling valid cross-file aggregation. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Fix region distribution binning to use uniform bin width across chrom…
Adds four new per-file summary statistics to GenomicIntervalSetStatistics, rewrites gaps() to require chrom_sizes for correct boundary handling, closes several pre-existing binding asymmetries, and exposes all of this through the CLI, Python, R, and Wasm layers. ## New Rust functions (gtars-genomicdist) - `calc_inter_peak_spacing` → SpacingStats (mean / median / std / IQR / log_mean / log_std) over calc_neighbor_distances(). NaN for empty, singleton, or all-overlapping inputs. - `calc_peak_clusters(radius_bp)` → ClusterStats (n_clusters, n_clustered_peaks, mean/max cluster size over size-gt-1 clusters, fraction_clustered) wrapping the existing cluster() primitive. - `calc_density_vector(chrom_sizes, n_bins)` → DensityVector. Dense zero-padded per-window count vector ordered karyotypically, with chrom_offsets so callers recover per-chromosome slices without re-binning. - `calc_density_homogeneity(chrom_sizes, n_bins)` → DensityHomogeneity (n_windows, n_nonzero_windows, mean_count, variance, cv, Gini). Delegates to calc_density_vector to avoid the zero-window correctness trap present in region_distribution_with_chrom_sizes. 28 new Rust unit tests cover edge cases (empty, singleton, per-chrom, cross-chrom, out-of-bounds, zero-padding). ## gaps() rewritten `IntervalRanges::gaps` now takes a required `&HashMap<String, u32>` of chromosome sizes and emits: - leading gaps from 0 to first peak - inter-region gaps (unchanged) - trailing gaps from last peak to chrom_size - full-chromosome gaps for any chromosome in chrom_sizes with no regions (matches bedtools complement) Output is karyotypically ordered. Regions past chrom_size are clipped. This is a **breaking API change** but corrects a real bug: the old gaps() couldn't include trailing-gap regions (no access to chrom_size) so the output was always wrong at the end of every chromosome. The new signature subsumes the proposed `largest_gaps` feature, which is dropped from scope — callers wanting top-N gaps filter gaps() output themselves. Updates the 3 existing callers (gtars-cli ranges, gtars-r, gtars-wasm) and adds 10 Rust unit tests for the new behavior. ## Pre-existing binding gaps closed - Python: expose `gaps()` on PyRegionSet (previously R/Wasm only) - R: expose `cluster()` as `clusterRegions(maxGap)` (previously Py only) - Wasm: expose `cluster(max_gap)` (previously Py only) ## CLI - `gtars ranges gaps`: new required `--chrom-sizes` flag. - `gtars genomicdist`: four new JSON output fields — `inter_peak_spacing`, `peak_clusters` (array over stitching radii), `density_vector`, `density_homogeneity`. Density fields gated on `--chrom-sizes` (warning when skipped, matching expected_partitions). - New `--cluster-radii` flag (default "500,5000,50000") probes promoter / enhancer / domain scales in one invocation. ## Bindings - **Python** (gtars-python): PyRegionSet gains gaps(), inter_peak_spacing(), peak_clusters(), density_vector(), density_homogeneity(). Four new #[pyclass] result wrappers with getters and __repr__. Module registration + .pyi stubs updated. 11 new pytest cases. - **R** (gtars-r): new S4 generics interPeakSpacing, peakClusters, densityVector, densityHomogeneity. gaps() method updated to take chrom_sizes (start/end retained for IRanges generic compatibility, ignored). extendr-wrappers.R and man/*.Rd regenerated via rextendr::document(). 11 new testthat cases. - **Wasm** (gtars-wasm): cluster(), interPeakSpacing(), peakClusters(), densityVector(), densityHomogeneity() on JsRegionSet. gaps() signature updated. Results via serde_wasm_bindgen. Compiles clean against wasm32-unknown-unknown. ## Side fix: calcDinuclFreq R wrapper arity `rextendr::document()` exposed a pre-existing bug: R/extendr-wrappers.R was checked in with a stale 3-arg signature for r_calc_dinucl_freq, while the Rust source has taken 4 args since commit bd1e5b7. The regenerated wrapper file is now correct, and R/genomicdist.R is updated to pass the 4th ignoreUnknownChroms argument so calcDinuclFreq() dispatches without a runtime arity error. ## Verification - cargo test -p gtars-genomicdist — 296 passed (+28 new) - cargo check --workspace — clean - cargo check -p gtars-js --target wasm32-unknown-unknown — clean - pytest tests/test_regionset.py — 25 passed (+11 new) - Rscript -e 'devtools::test()' — 288 passed (+11 new; 2 pre-existing calcDinuclFreq failures fixed by the side fix above) - End-to-end CLI smoke against ENCODE ENCFF730DLS (23K regions, hg19): all four new JSON fields populate with plausible values. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Addresses review feedback on #254 — no code changes, documentation only. Reviewer found that `n_bins=249` against canonical UCSC hg38.chrom.sizes (455 entries) produced 3585 total windows because every contig shorter than the derived bin_width contributes a single narrow bin. The parameter semantic ("target bin count for the longest chromosome") wasn't clear from the existing docstrings, and the consequence for short contigs wasn't documented at all. Updated docstrings now state: - `n_bins` is the target bin count for the **longest chromosome** in `chrom_sizes`, not the total length of `counts`. Bin width is `max(chrom_sizes) / n_bins` (floored, minimum 1 bp). Total bins returned is `sum(ceil(size / bin_width))` across all chromosomes and can substantially exceed `n_bins`. To target a specific bin width, pass `n_bins = max_chrom_len / desired_bin_width_bp`. - The last bin on each chromosome is narrower than `bin_width` whenever `chrom_size` is not an exact multiple. Chromosomes shorter than `bin_width` (common with UCSC alt / random / unplaced contigs) reduce to a single bin whose effective width equals the chromosome size. Counts are per-bin, not per-bp — bins of different effective widths are not directly comparable as densities. The doc framing is a technical description of behavior, not a "you should pre-filter your chrom_sizes" prescription — users decide what to do with the information. Mirrored the same language across every layer so the caller sees the same explanation regardless of binding: - gtars-genomicdist/src/statistics.rs — trait method docstrings for calc_density_vector and calc_density_homogeneity - gtars-genomicdist/src/models.rs — struct-level docstrings for DensityVector and DensityHomogeneity, with field-level notes on bin_width and n_windows - gtars-python/py_src/gtars/models/__init__.pyi — class-level and method-level docstrings - gtars-python/src/models/region_set.rs — PyRegionSet method docs - gtars-r/R/RegionSet-methods.R — roxygen Details sections on densityVector and densityHomogeneity generics - gtars-r/src/rust/src/genomicdist.rs — extendr doc comments for r_calc_density_vector and r_calc_density_homogeneity (become Rd man pages) - gtars-wasm/src/regionset.rs — wasm-bindgen doc comments for densityVector and densityHomogeneity - gtars-cli/src/genomicdist/cli.rs — --bins help text now explains the semantic and how to compute a value from a target bin width - gtars-cli/src/genomicdist/handlers.rs — GenomicDistOutput field docs for density_vector and density_homogeneity Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Addresses review feedback on #254 — Sanghoon found that calc_peak_clusters mixed two populations: `n_clusters` counted all connected components including singletons, while `mean_cluster_size` averaged only over clusters of size > 1. The naive product `n_clusters * mean_cluster_size` was not `n_clustered_peaks`, which caused real confusion during integration testing. ## Fix Add `min_cluster_size: usize` parameter to `calc_peak_clusters` and apply it **uniformly** to every size-dependent field: - `n_clusters` = count of clusters with size >= min_cluster_size - `n_clustered_peaks` = peaks belonging to those clusters - `mean_cluster_size` = mean size of those clusters - `fraction_clustered` = n_clustered_peaks / total_peaks (raw total) - `max_cluster_size` = max over **all** clusters (inherently unfiltered) This preserves the arithmetic identity `n_clusters * mean_cluster_size == n_clustered_peaks` at any threshold. ## Default value Bindings default `min_cluster_size` to **2**. At this threshold every field describes "clusters of at least 2 peaks" — the scientifically meaningful view matching typical enhancer-clustering / super-enhancer stitching analyses. `fraction_clustered` is then the fraction of peaks with at least one neighbor within `radius_bp`. Callers who want the simple-average view pass `min_cluster_size = 1`: `mean_cluster_size` becomes the simple `total_peaks / n_clusters`, but `n_clustered_peaks` degenerates to `total_peaks` and `fraction_clustered` to `1.0` by definition (every peak is in a cluster of size >= 1). Useful but less informative — the size-count fields become tautological at this threshold. Explored but rejected: - Leaving min_cluster_size to affect only the mean (preserves existing asymmetry). - Defaulting to 1 (simple mean): Sanghoon's identity trap returns because n_clusters and mean_cluster_size use different populations at any non-default value. ## Empty / edge cases - Empty input: all fields zero, mean and fraction NaN (unchanged). - All singletons at default min=2: n_clusters=0, n_clustered_peaks=0, max_cluster_size=1 (inherent max), mean=NaN, fraction=0. - All singletons at min=1: n_clusters=n_peaks, all three peaks each contribute a size-1 cluster, mean=1.0, fraction=1.0. ## Changes by layer - **Rust core** (gtars-genomicdist): trait method signature updated, impl rewritten to apply the filter uniformly, ClusterStats struct docstring rewritten, 8 unit tests updated/added covering default, min=1, min=3, all-singletons-both-thresholds, and cross-chromosome. - **Python** (gtars-python): `peak_clusters(radius_bp, min_cluster_size=2)` with pyo3 default. .pyi stub updated. 3 new pytest cases (default, min=1 simple average, empty). - **R** (gtars-r): S4 generic and methods default `min_cluster_size = 2L`. Rust extendr binding takes the parameter explicitly. testthat tests rewritten. extendr-wrappers.R regenerated via `rextendr::document()`; peakClusters.Rd and r_calc_peak_clusters.Rd also regenerated. - **Wasm** (gtars-wasm): `peakClusters(radius_bp, min_cluster_size)` — no default at the Wasm layer, users pass both. Doc comment updated. - **CLI** (gtars-cli genomicdist): new `--cluster-min-size` flag with default 2, applied uniformly across all --cluster-radii. Handler parses and threads through to `calc_peak_clusters`. ## Verification - cargo test -p gtars-genomicdist — 299 passed (+3 new over last baseline: min=3 case, all-singletons-min=1, all-singletons-min=2) - cargo check --workspace — clean - Rscript -e 'devtools::test()' — 294 passed (no regressions) - pytest tests/test_regionset.py — 26 passed - End-to-end CLI against real ENCFF730DLS (23420 peaks): default (min=2), r=5000: n_clusters=5614, n_clustered=16555, mean=2.949, fraction=0.707 identity: 5614 * 2.949 == 16555 ✓ min=1, r=5000: n_clusters=12479, n_clustered=23420, mean=1.877, fraction=1.000 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add spatial-arrangement stats and fix gaps() chrom_sizes signature
…, R 0.9.0 Prep for next release after PR #254 (spatial-arrangement stats + gaps() signature fix). Genomicdist gets a minor bump for new public API and the breaking gaps() change; bindings and CLI bump to 0.9.0 since they re-export the new functions and inherit the breaking change. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace PyResult<PyObject> with PyResult<Py<PyAny>> (PyObject is a deprecated alias) and Python::with_gil with Python::attach. Both are mechanical renames in pyo3 0.27 with no behavioral change; will become hard errors in pyo3 0.28+. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove unused `wasm_bindgen::prelude::*` import from lib.rs (re-exports don't need it; submodules import it themselves). Comment out the [profile.release] block in Cargo.toml — cargo silently ignores profile sections in non-root workspace members, so this block has been dead since the wasm crate was created. Left commented with a note explaining the correct placement options for any future opt-level work. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
PathBuf is only used inside `from_pretrained`, which is itself `#[cfg(feature = "huggingface")]`. Without huggingface enabled (e.g. in the wasm build, which doesn't use that feature), the import becomes unused and triggers an `unused_imports` warning. Gate the import with the same cfg as its call site. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## master #256 +/- ##
===========================================
+ Coverage 57.62% 82.44% +24.81%
===========================================
Files 91 74 -17
Lines 13049 20117 +7068
===========================================
+ Hits 7520 16585 +9065
+ Misses 5529 3532 -1997 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Surgically reverts the four per-file summary statistics added in #254 (`calc_inter_peak_spacing`, `calc_peak_clusters`, `calc_density_vector`, `calc_density_homogeneity`) and the `--cluster-radii` / `--cluster-min-size` CLI flags that drove them. On review these APIs are either thin scalar reductions over existing primitives (`calc_neighbor_distances`, `region_distribution_with_chrom_sizes`) or — in the case of `calc_peak_clusters` — parameterized by a biologically-underdetermined stitching radius. Not load-bearing for any current consumer. ## Removed - gtars-genomicdist: 4 trait methods and 4 result structs (`SpacingStats`, `ClusterStats`, `DensityVector`, `DensityHomogeneity`), ~28 unit tests. - gtars-cli: `--cluster-radii`, `--cluster-min-size` flags and the four JSON output fields in `gtars genomicdist`. - gtars-python: `inter_peak_spacing()`, `peak_clusters()`, `density_vector()`, `density_homogeneity()` on `PyRegionSet`; 4 `#[pyclass]` result wrappers; `.pyi` stubs; 11 pytest cases. - gtars-r: `interPeakSpacing`, `peakClusters`, `densityVector`, `densityHomogeneity` S4 generics and methods; 4 `r_calc_*` Rust wrappers; 8 `man/*.Rd` files; NAMESPACE exports; testthat cases. - gtars-wasm: `interPeakSpacing`, `peakClusters`, `densityVector`, `densityHomogeneity` methods on `JsRegionSet`. ## Kept (also from #254, orthogonally good) - `IntervalRanges::gaps(chrom_sizes)` rewrite. Real bug fix — the old signature couldn't emit trailing gaps and was silently wrong at every chromosome end. The breaking API change stands. - `cluster()` binding parity: R `clusterRegions(maxGap)` and WASM `cluster(max_gap)`. These expose a pre-existing Rust primitive (`IntervalRanges::cluster` from #241) to R and WASM for parity with Python. Independent of the stats wrapper critique. - `calcDinuclFreq` R wrapper arity fix (side fix in #254). - Version bumps (genomicdist 0.8.0, cli/python/wasm/R 0.9.0). The `gaps()` signature change is breaking on its own and independently warrants the minor bump. ## Verification - `cargo test -p gtars-genomicdist` — 275 passed, 0 failed - `cargo check --workspace` — clean - `cargo check -p gtars-js --target wasm32-unknown-unknown` — clean - `pytest tests/test_regionset.py` — 17 passed - `Rscript -e 'devtools::test()'` — 259 passed, 0 failed Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Remove spatial-arrangement stats from genomicdist
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
3 participants
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.
opening a PR to track work on dev