The fixtures originally landed in `tests/data/modules/trim_galore/` of
the main repo, but `test_modules_run.py` resolves test data via
`<repo>/test-data/data/modules/<module>/` (a separate sibling repo,
MultiQC/test-data). Moving them there in MultiQC/test-data#377.

* Add native MultiQC module for Trim Galore v2.x (Oxidized Edition)

Closes #3529.

Trim Galore v2.x emits a structured `*_trimming_report.json` (schema v1)
alongside the legacy `*_trimming_report.txt` report. The text report
still carries the `"This is cutadapt"` shim for backwards compatibility,
so the existing `cutadapt` module path keeps working unchanged. This new
module parses the JSON natively, which:

- Gets the Software Versions table right ("Trim Galore X.Y.Z" instead of
  the misleading "Cutadapt 4.0" backwards-compat shim)
- Surfaces TrimGalore-specific stats not available from Cutadapt output
  (RRBS truncation counts, poly-A/G trimming, paired-end pair-validation
  outcomes — the latter two are wired through to the data file but not
  yet plotted; happy to add follow-up sections)

## What's plotted

- General stats columns: % adapter, % pass, % q-trimmed, total reads
  (hidden), total bp written (hidden)
- Filtered reads bargraph: passing / too_short / too_long / too_many_n
  / discarded_untrimmed
- Adapter length distribution linegraph (per sample, per adapter when a
  sample has more than one)

## Sample-name handling

PE TrimGalore reports list both R1 and R2 in `input_filenames` (both
JSONs do — Trim Galore preserves the pair context). The parser uses the
JSON's `read_number` field to pick the correct filename, so R1 and R2
become distinct samples.

## Coexistence with the cutadapt module

Both modules will discover their respective files (text vs JSON). With
both enabled, each sample appears in both modules' general-stats
columns. Users who want to disable the cutadapt path on TrimGalore
samples can:

```yaml
disable_modules:
  - cutadapt
```

Documented in the module's class docstring.

## Test fixtures

`tests/data/modules/trim_galore/` contains:
- `sample_R1.fastq.gz_trimming_report.json` — SE example (10K Illumina)
- `BS-seq_10K_R{1,2}.fastq.gz_trimming_report.json` — PE example (BS-seq
  10K, with `pair_validation` populated)

Verified locally: `multiqc -m trim_galore tests/data/modules/trim_galore/`
produces 3/3 reports parsed (1 SE + 2 PE), all sections rendered, data
file written to `multiqc_data/multiqc_trim_galore.txt`.

## Schema reference

JSON schema v1 is documented in the upstream issue thread (linked in
the issue body). The parser version-gates on `schema_version: 1` and
warns + skips files with a different version, so a future schema bump
won't silently misparse.

## Status

Marking as draft. Initial scope is intentionally focused — happy to
extend for poly-A/G trimming sections, pair-validation visualisation,
RRBS-specific stats, or anything else the maintainers want before a
final review pass.

* Apply prettier + ruff format from prek hooks

* Move test fixtures to MultiQC/test-data fork (companion PR)

The fixtures originally landed in `tests/data/modules/trim_galore/` of
the main repo, but `test_modules_run.py` resolves test data via
`<repo>/test-data/data/modules/<module>/` (a separate sibling repo,
MultiQC/test-data). Moving them there in MultiQC/test-data#377.

* Address PR review feedback on trim_galore module

- Move write_data_file to end of __init__ and flatten payload to scalar
  columns so multiqc_trim_galore.txt is machine-readable
- Call add_software_version unconditionally; bail on ignored samples
  inside the parse loop
- Drop module-level docstring and unicode divider comments per project
  style; tone down class docstring
- Drop redundant _strip_fastq_suffix helper in favour of clean_s_name
- Add SampleGroupingConfig so PE pairs collapse cleanly with
  table_sample_merge (weighted-average percentages, sum counts)
- Remove hardcoded bargraph colours; use uniform composite keys in the
  adapter-length plot and continue on zero-adapter samples
- Drop % Q-trim precision to {:,.1f}; surface tg_total_reads by default
- Bump schema_version mismatch to log.error with explicit guidance
- Simplify search_patterns.yaml (drop contents/num_lines shim)
- Group trim_galore adjacent to cutadapt in config_defaults.yaml
- Revert CHANGELOG.md entry (generated from PR titles)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Log debug message when JSON tool field is not Trim Galore

Helps diagnostics if a non-TrimGalore JSON happens to match the
filename glob.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Auto-suppress cutadapt module for Trim Galore v2.x text reports

The cutadapt module's text-report pattern matches any file containing
"This is cutadapt", which also catches the backwards-compatibility shim
that Trim Galore v2.x writes alongside its native JSON report. Result:
every v2.x sample shows up twice — once via cutadapt (as a misleading
"Cutadapt 4.0"), once via trim_galore. Telling users to disable cutadapt
globally also kills parsing of genuine cutadapt logs and legacy
Trim Galore v0/v1 reports, so it isn't a real fix.

Add an exclude_contents_re to the cutadapt text-report pattern matching
"Trim Galore version: " followed by a major version of 2 or higher.
v0.x / v1.x text reports continue to be picked up by cutadapt; v2.x
text reports are skipped (the sibling JSON is handled by trim_galore);
pure cutadapt logs are unaffected.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Add Pair Validation, Poly-A/G, and RRBS sections to trim_galore

Surface the schema-v1 fields that were already in the data file but not
plotted: pair_validation, poly_a_trimming, poly_g_trimming, rrbs. Each is
a small table with sensible gating.

- Pair Validation: collapses R1/R2 (pair-level data is identical between
  them), drops rows where less than 0.1% of pairs were affected.
- Poly-A/G and RRBS: per-row gating, samples with zero counts are hidden.
- All three sections show a Bootstrap alert listing dropped samples, with
  long lists wrapped in <details> (bases2fastq pattern).
- Defensive try/except around length_distribution int-coercion so a
  malformed key downgrades to a debug log rather than crashing the run.
- Data file flattening extended to all of pair_validation,
  poly_a_trimming, poly_g_trimming and rrbs blocks (25 columns total).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Add support for sample grouping

* Add explicit_groups for deterministic tool-derived sample grouping

Extend SampleGroupingConfig with an `explicit_groups` parameter that
lets modules supply their own ground-truth groups instead of relying on
the user's `table_sample_merge` name patterns. Useful when the tool
output already tells you which samples are related — paired-end
trimmers that emit both filenames in each report, lane manifests,
replicate IDs, etc. The framework silently ignores entries with a
single member so callers don't need to filter them out themselves.

Wire trim_galore to use this. Each JSON's `tuple(input_filenames)` is
a stable pair key (byte-identical between R1 and R2 of the same pair).
Auto-grouping applies to:

- General Stats table — framework path with expand-to-see-individuals
- Pair Validation table — manual collapse keyed on the same pair_key

Filtered Reads bargraph, Poly-A/G and RRBS tables stay per-read because
R1 and R2 stats there can legitimately differ. Users with
`table_sample_merge` configured layer name-pattern grouping on top of
the auto-derived pairs. The `trim_galore_config.auto_group_pairs: false`
flag opts out of auto-grouping entirely.

Replaces the earlier `_apply_grouping` helper that relied on
`config.table_sample_merge` to pre-aggregate filtered_reads / poly /
RRBS — those now stay per-read regardless of grouping config.

Docs updated: developer guide gets a worked example for module authors
with authoritative pair info; user-facing customisation page describes
the auto-grouping behaviour and the opt-out.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Simplify trim_galore module after review pass

- Store pair_key_to_samples and pair_display_by_key on `self` so
  _pair_validation_plot and _derive_auto_groups drop their extra
  parameters
- Add a small _add_filtered_section helper that wraps the
  plot+description+alert+add_section pattern, collapsing three
  near-identical 11-line blocks
- Simplify the gen_stats type annotation from a quadruple Union
  workaround to `Dict[str, Dict[ColumnKey, Any]]` plus a single
  `cast(Any, ...)` at the addcols call site
- Drop the unused `Union` import that fell out of the above
- Tighten narrative comments per CLAUDE.md (keep only WHY)

Net: -30 lines, no behaviour change. Lint / mypy / module tests all
clean across the three grouping scenarios (default, with
table_sample_merge, and opt-out).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Bit of clean up

* Manual review of docs

* Schema version: assume semver, only throw error on major version bump

* Remove some excessively cautious code

* Make code way less defensive still.

If the data is that badly mangled, I'd rather it throw an exception instead of silently default to fake numbers

* Better docstring / docs

* Tidy up descriptions / helptext a bit2

---------

Co-authored-by: Phil Ewels <phil.ewels@seqera.io>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>