@jbrowse/plugin-gtf 4.3.0 → 5.0.0-beta.1

package/esm/util.d.ts

@@ -1,13 +1,83 @@
+import type { SimpleFeatureSerialized } from '@jbrowse/core/util';
 export type Strand = '+' | '-' | '.' | '?';
 export interface FeatureLoc {
-    [key: string]: unknown;
     start: number;
     end: number;
     strand: Strand;
     seq_name: string;
-    child_features?: FeatureLoc[][];
-    data: unknown;
-    derived_features: unknown;
-    attributes: Record<string, unknown[]>;
+    featureType?: string | null;
+    source?: string | null;
+    score?: number | null;
+    frame?: string | null;
+    child_features?: FeatureLoc[];
+    attributes: Record<string, string[]>;
 }
 export declare function featureData(data: FeatureLoc, id?: string): Record<string, unknown>;
+/** A raw GTF feature line; callers may extend it with their own identity. */
+export interface GtfLineRecord {
+    line: string;
+}
+/**
+ * A top-level parsed feature paired with the record it came from. The parser
+ * stamps no identity onto the feature; callers that need a stable per-feature
+ * id (e.g. a tabix byte offset) read it off their own `record`. Mirrors
+ * gff-nostream's `ParsedRecord`.
+ */
+export interface ParsedGtfRecord<R extends GtfLineRecord = GtfLineRecord> {
+    feature: FeatureLoc;
+    record: R;
+}
+/**
+ * Parse an array of records wrapping raw GTF lines into top-level features.
+ * Lines sharing a `transcript_id` are grouped under a transcript feature
+ * (synthesized if the file has no explicit `transcript` line, per the
+ * Cufflinks/StringTie convention); the transcript is spanned to cover its
+ * children. Lines without a `transcript_id` (e.g. a `gene` line) pass through
+ * as standalone features. Each top-level feature is returned paired with the
+ * record that defined it (the explicit line, or a synthesized transcript's
+ * first child), so callers can attach a stable id.
+ */
+export declare function parseGtf<R extends GtfLineRecord>(records: readonly R[]): ParsedGtfRecord<R>[];
+/**
+ * Parse GTF records and serialize each top-level feature. Shared by the
+ * plain-text and tabix adapters, which differ only in how they source records
+ * and derive the uniqueId (parse index vs. tabix byte offset).
+ */
+export declare function parseGtfToFeatures<R extends GtfLineRecord>(records: readonly R[], makeUniqueId: (record: R, index: number) => string): SimpleFeatureSerialized[];
+/**
+ * GTF has no spanning gene line, so a gene is synthesized by grouping related
+ * transcript features and spanning them. An explicit `gene` line is held aside
+ * rather than aggregated — the parser hangs exons off the transcript, so a gene
+ * line is childless and the synthesized parent supersedes it — and emitted only
+ * if no synthesized gene claimed its `gene_id`, which is what keeps a file of
+ * nothing but `gene` rows from rendering as an empty track. A gene line with no
+ * `gene_id` at all (AUGUSTUS writes a bare `g1` in column 9, which parses to no
+ * attributes) has no key to be superseded by, and is dropped.
+ *
+ * Grouping is keyed by `gene_id` where a transcript carries one, falling back to
+ * the aggregate value; `aggregateField` supplies the gene's display `name`. The
+ * two are separate because neither alone works:
+ *
+ * - `gene_name` (the default aggregate field, and the only one suitable as a
+ *   label) is not unique within a reference sequence. A GENCODE chromosome holds
+ *   hundreds of separate genes named `U6`, `Y_RNA` or `5S_rRNA`, and keying on
+ *   the name alone fused all of them into one feature spanning the whole
+ *   chromosome. `gene_id` is unique per gene by the GTF spec, so it splits those
+ *   back apart while still merging the transcripts that really do share a gene.
+ * - Plenty of GTFs carry `gene_id` and no `gene_name` at all — UCSC's
+ *   `genePredToGtf` emits `gene_id "TP53"; transcript_id "NM_000546";` and
+ *   nothing else, and AUGUSTUS is the same shape. Keying only on the aggregate
+ *   value left those with no gene model whatsoever: every transcript passed
+ *   through bare. Keying on `gene_id` gives them genes, named by the `gene_id`
+ *   when there's no better label.
+ *
+ * A feature with neither key still passes through unchanged.
+ */
+export declare function aggregateGtfFeatures({ feats, aggregateField, refName, regionStart, regionEnd, idPrefix, }: {
+    feats: SimpleFeatureSerialized[];
+    aggregateField: string;
+    refName: string;
+    regionStart: number;
+    regionEnd: number;
+    idPrefix: string;
+}): SimpleFeatureSerialized[];

package/esm/util.js

@@ -1,56 +1,469 @@
+import { doesIntersect2 } from '@jbrowse/core/util';
+const strandMap = { '+': 1, '-': -1, '.': 0, '?': undefined };
+// structural output fields an attribute must not overwrite; a clashing
+// attribute key gets a `2` suffix instead (compared lowercase, so camelCase
+// output fields like refName never collide). `subfeatures` is the one that
+// matters beyond tidiness: on a feature with no children nothing overwrites it
+// back, so an attribute of that name would leave a string where every consumer
+// expects the child array
+const reservedFields = new Set([
+    'type',
+    'source',
+    'start',
+    'end',
+    'strand',
+    'score',
+    'phase',
+    'subfeatures',
+]);
 export function featureData(data, id) {
-    const f = { ...data };
-    f.start -= 1;
-    f.strand = { '+': 1, '-': -1, '.': 0, '?': undefined }[data.strand];
-    f.phase = Number(data.frame);
-    f.refName = data.seq_name;
-    if (data.score === null) {
-        f.score = undefined;
-    }
-    if (data.frame === null) {
-        f.score = undefined;
-    }
-    const defaultFields = new Set([
-        'start',
-        'end',
-        'seq_name',
-        'score',
-        'featureType',
-        'source',
-        'frame',
-        'strand',
-    ]);
-    for (const a of Object.keys(data.attributes)) {
-        let b = a.toLowerCase();
-        if (defaultFields.has(b)) {
-            b += '2';
-        }
-        if (data.attributes[a]) {
-            let attr = data.attributes[a];
-            if (Array.isArray(attr) && attr.length === 1) {
-                attr = attr[0].replaceAll(/^"|"$/g, '');
-            }
-            f[b] = attr;
-        }
-    }
-    f.refName = f.seq_name;
-    f.type = f.featureType;
-    if (data.child_features && data.child_features.length > 0) {
-        f.subfeatures = data.child_features.flatMap(childLocs => childLocs.map(childLoc => featureData(childLoc)));
-    }
-    f.child_features = undefined;
-    f.data = undefined;
-    f.derived_features = undefined;
-    f._linehash = undefined;
-    f.attributes = undefined;
-    f.seq_name = undefined;
-    f.featureType = undefined;
-    f.frame = undefined;
-    if (f.transcript_id) {
-        f.name = f.transcript_id;
+    const subfeatures = data.child_features?.length
+        ? data.child_features.map(childLoc => featureData(childLoc))
+        : undefined;
+    // build the output explicitly rather than spreading `data` and clearing its
+    // raw parser fields, which would leave them as `undefined`-valued keys
+    const out = {
+        refName: data.seq_name,
+        type: data.featureType,
+        source: data.source,
+        start: data.start - 1,
+        end: data.end,
+        strand: strandMap[data.strand],
+        score: data.score ?? undefined,
+        phase: data.frame ? Number(data.frame) : undefined,
+    };
+    // lowercase keys, suffix clashes with reserved fields, unwrap single-element
+    // arrays (quotes were already stripped at parse time). Assigned onto the
+    // output rather than built into an intermediate object and spread: the
+    // spread's `Object.entries`/`.map`/`Object.fromEntries` allocated four
+    // throwaway arrays plus a second object per feature, on a path that runs once
+    // per line of the file
+    for (const a in data.attributes) {
+        const vals = data.attributes[a];
+        const lower = a.toLowerCase();
+        out[reservedFields.has(lower) ? `${lower}2` : lower] =
+            vals.length === 1 ? vals[0] : vals;
+    }
+    if (subfeatures !== undefined) {
+        out.subfeatures = subfeatures;
+    }
+    // after the attributes, so an attribute that happens to be called `name`
+    // loses to the transcript_id the renderer labels with — as it did when this
+    // was a spread in this order
+    const transcriptId = out.transcript_id;
+    if (transcriptId) {
+        out.name = transcriptId;
     }
     if (id !== undefined) {
-        f.uniqueId = id;
+        out.uniqueId = id;
+    }
+    return out;
+}
+function toStrand(s) {
+    return s === '+' || s === '-' || s === '?' ? s : '.';
+}
+function nullIfDot(s) {
+    return s === undefined || s === '.' ? null : s;
+}
+/**
+ * True when `[from, to)` opens a double quote it never closes, meaning the ';'
+ * that ended it fell inside the value instead of ending the entry. A GTF value
+ * is either bare or wholly quoted, so a second quote settles it — no need to
+ * scan the rest of the entry.
+ */
+function hasUnclosedQuote(s, from, to) {
+    const first = s.indexOf('"', from);
+    if (first === -1 || first >= to) {
+        return false;
+    }
+    const second = s.indexOf('"', first + 1);
+    return second === -1 || second >= to;
+}
+// ASCII whitespace, which is what `String.prototype.trim` strips that can occur
+// in a GTF attribute column: space plus \t \n \v \f \r. GTF is an ASCII format,
+// so the Unicode space separators trim also handles cannot appear as padding
+// here — and if one did it would stay in the key or value rather than corrupt
+// the parse
+function isSpace(c) {
+    return c === 32 || (c >= 9 && c <= 13);
+}
+/**
+ * Parse the GTF 9th column (`gene_id "X"; transcript_id "Y";`). Each `key
+ * "value"` entry contributes one value; GTF expresses multiple values per key
+ * via repeated keys (`tag "A"; tag "B"`), not comma separation, so the value is
+ * taken whole (a comma inside it, e.g. `note "a, b"`, stays intact). Surrounding
+ * double-quotes are stripped here.
+ *
+ * A quoted value may also contain the ';' the entries are split on
+ * (`note "a; b"`), so an entry is extended past that ';' before the value is
+ * read. Without that the value is silently truncated at the semicolon, and the
+ * remainder — having no `key value` shape — is dropped rather than erroring.
+ *
+ * Scanned over the string by index rather than `split(';')`-then-trim-then-
+ * `replaceAll(/^"|"$/g)`. This is the GTF load path's single hottest function —
+ * on a GENCODE-shaped 26,870-line file it was more of the profile than every
+ * other function combined, plus the GC for the two intermediate strings and the
+ * regex match it produced per attribute. Only the key and the value are
+ * allocated now, which are the two strings that get kept.
+ */
+function parseGtfAttributes(s, from, to) {
+    const attrs = {};
+    if (from >= to || (to - from === 1 && s.charCodeAt(from) === 46) /* . */) {
+        return attrs;
+    }
+    let i = from;
+    while (i < to) {
+        let end = s.indexOf(';', i);
+        if (end === -1 || end >= to) {
+            end = to;
+        }
+        // an unclosed quote means this ';' was inside the value: take the next one
+        // instead. A trailing entry with no closing quote has nothing to extend
+        // into and is read as-is
+        while (end < to && hasUnclosedQuote(s, i, end)) {
+            const next = s.indexOf(';', end + 1);
+            end = next === -1 || next >= to ? to : next;
+        }
+        // trim the entry by index
+        let es = i;
+        let ee = end;
+        while (es < ee && isSpace(s.charCodeAt(es))) {
+            es++;
+        }
+        while (ee > es && isSpace(s.charCodeAt(ee - 1))) {
+            ee--;
+        }
+        // `key value`, split at the first space — a tab does not separate them, as
+        // it did not when this read `trimmed.indexOf(' ')`
+        const sp = s.indexOf(' ', es);
+        if (sp !== -1 && sp < ee) {
+            let vs = sp + 1;
+            while (vs < ee && isSpace(s.charCodeAt(vs))) {
+                vs++;
+            }
+            let ve = ee;
+            // strip one leading and one trailing quote, independently: `"x"` and a
+            // half-quoted `"x` both yield `x`, and a bare `""` yields the empty
+            // string that is then dropped
+            if (vs < ve && s.charCodeAt(vs) === 34) {
+                vs++;
+            }
+            if (ve > vs && s.charCodeAt(ve - 1) === 34) {
+                ve--;
+            }
+            if (ve > vs) {
+                const key = s.slice(es, sp);
+                (attrs[key] ??= []).push(s.slice(vs, ve));
+            }
+        }
+        i = end + 1;
+    }
+    return attrs;
+}
+/**
+ * A line with fewer than nine columns, which is malformed. Split rather than
+ * scanned, so its absent fields and NaN coordinates come out exactly as they
+ * did when every line was split — the caller drops it on those NaNs.
+ */
+function parseShortGtfLine(line) {
+    const c = line.split('\t');
+    const score = c[5];
+    // a line this short has no ninth column at all, but read it if one is
+    // somehow there rather than differing from the old behaviour twice over
+    const attrs = c[8] ?? '';
+    return {
+        seq_name: c[0] ?? '',
+        source: nullIfDot(c[1]),
+        featureType: nullIfDot(c[2]),
+        start: Number(c[3]),
+        end: Number(c[4]),
+        score: score === undefined || score === '.' ? null : Number(score),
+        strand: toStrand(c[6]),
+        frame: nullIfDot(c[7]),
+        attributes: parseGtfAttributes(attrs, 0, attrs.length),
+    };
+}
+function parseGtfLine(line) {
+    // Scan for the eight tabs bounding the nine columns rather than splitting on
+    // them. Splitting materialized column 9 — by far the longest part of an
+    // annotation-grade line — as its own string, only for the attribute parser to
+    // read it and throw it away; the parser now reads it in place, so the biggest
+    // allocation per line is gone along with the array and the three numeric
+    // columns' strings.
+    //
+    // Taking column 9 as everything after the eighth tab also stops a tab inside
+    // it from truncating the attributes. `split('\t')` put the remainder in a
+    // tenth element that nothing read, so every attribute past such a tab was
+    // silently dropped.
+    const at = [0, 0, 0, 0, 0, 0, 0, 0];
+    let p = -1;
+    for (let i = 0; i < 8; i++) {
+        p = line.indexOf('\t', p + 1);
+        if (p === -1) {
+            return parseShortGtfLine(line);
+        }
+        at[i] = p;
+    }
+    const score = line.slice(at[4] + 1, at[5]);
+    return {
+        seq_name: line.slice(0, at[0]),
+        source: nullIfDot(line.slice(at[0] + 1, at[1])),
+        featureType: nullIfDot(line.slice(at[1] + 1, at[2])),
+        start: Number(line.slice(at[2] + 1, at[3])),
+        end: Number(line.slice(at[3] + 1, at[4])),
+        score: score === '.' ? null : Number(score),
+        strand: toStrand(line.slice(at[5] + 1, at[6])),
+        frame: nullIfDot(line.slice(at[6] + 1, at[7])),
+        attributes: parseGtfAttributes(line, at[7] + 1, line.length),
+    };
+}
+function synthesizeTranscript(child) {
+    return {
+        seq_name: child.seq_name,
+        source: child.source,
+        featureType: 'transcript',
+        start: child.start,
+        end: child.end,
+        score: null,
+        strand: child.strand,
+        frame: null,
+        // seeded from the first child, then narrowed by narrowToCommonAttributes as
+        // the rest arrive
+        attributes: { ...child.attributes },
+        child_features: [],
+    };
+}
+/**
+ * Keep only the attributes a synthesized transcript's children agree on,
+ * dropping any key that is missing from `child` or carries a different value
+ * there. Transcript- and gene-level tags (`transcript_id`, `gene_id`,
+ * `gene_name`, biotypes, StringTie's `ref_gene_name`, …) repeat identically on
+ * every line of a transcript and survive; per-line tags (`exon_number`,
+ * `exon_id`, a CDS-only `protein_id`) differ or are absent and are dropped.
+ *
+ * Deriving the set this way rather than from a fixed whitelist is what lets any
+ * `aggregateField` work: a hardcoded list silently failed to aggregate whenever
+ * the configured field wasn't one of the three names it happened to enumerate.
+ */
+function narrowToCommonAttributes(attributes, child) {
+    for (const key of Object.keys(attributes)) {
+        const mine = attributes[key];
+        const theirs = child.attributes[key];
+        if (theirs === undefined ||
+            theirs.length !== mine.length ||
+            theirs.some((v, i) => v !== mine[i])) {
+            delete attributes[key];
+        }
+    }
+}
+/**
+ * Parse an array of records wrapping raw GTF lines into top-level features.
+ * Lines sharing a `transcript_id` are grouped under a transcript feature
+ * (synthesized if the file has no explicit `transcript` line, per the
+ * Cufflinks/StringTie convention); the transcript is spanned to cover its
+ * children. Lines without a `transcript_id` (e.g. a `gene` line) pass through
+ * as standalone features. Each top-level feature is returned paired with the
+ * record that defined it (the explicit line, or a synthesized transcript's
+ * first child), so callers can attach a stable id.
+ */
+export function parseGtf(records) {
+    const topLevel = [];
+    const byTranscript = new Map();
+    // transcripts with no explicit line of their own, whose attributes are still
+    // being narrowed down to what their children agree on
+    const synthesized = new Set();
+    for (const record of records) {
+        // defensive: both of today's line sources already drop a CRLF terminator
+        // (@gmod/tabix trims it in its line reader "matching htslib's", and the
+        // plaintext path's parseLineByLine trims each line), so this normally does
+        // nothing. It stays because an untrimmed \r is silent — on a line with no
+        // trailing ';' it lands inside the last attribute value, corrupting the
+        // transcript_id that drives both grouping and the feature name
+        const line = record.line.endsWith('\r')
+            ? record.line.slice(0, -1)
+            : record.line;
+        if (line.length === 0 || line.startsWith('#')) {
+            continue;
+        }
+        const feature = parseGtfLine(line);
+        // a line whose coordinate columns aren't numbers is not a feature, and
+        // letting its NaN through is not a local failure: Math.min/Math.max against
+        // NaN poisons the transcript spanning it, and a NaN-bounded transcript then
+        // fails every intersection test, so one truncated line silently removes a
+        // whole gene rather than itself
+        if (Number.isNaN(feature.start) || Number.isNaN(feature.end)) {
+            continue;
+        }
+        const transcriptId = feature.attributes.transcript_id?.[0];
+        if (transcriptId === undefined) {
+            topLevel.push({ feature, record });
+        }
+        else if (feature.featureType === 'transcript') {
+            const existing = byTranscript.get(transcriptId);
+            if (existing) {
+                // explicit transcript line seen after its children: keep the
+                // collected children, but use the explicit line as the container
+                feature.child_features = existing.feature.child_features;
+                feature.start = Math.min(feature.start, existing.feature.start);
+                feature.end = Math.max(feature.end, existing.feature.end);
+                existing.feature = feature;
+                existing.record = record;
+                synthesized.delete(transcriptId);
+            }
+            else {
+                feature.child_features = [];
+                const parsed = { feature, record };
+                topLevel.push(parsed);
+                byTranscript.set(transcriptId, parsed);
+            }
+        }
+        else {
+            let transcript = byTranscript.get(transcriptId);
+            if (transcript) {
+                // an explicit transcript line's own attributes are authoritative, so
+                // only a synthesized one gets narrowed against this child
+                if (synthesized.has(transcriptId)) {
+                    narrowToCommonAttributes(transcript.feature.attributes, feature);
+                }
+            }
+            else {
+                transcript = { feature: synthesizeTranscript(feature), record };
+                topLevel.push(transcript);
+                byTranscript.set(transcriptId, transcript);
+                synthesized.add(transcriptId);
+            }
+            transcript.feature.child_features.push(feature);
+            transcript.feature.start = Math.min(transcript.feature.start, feature.start);
+            transcript.feature.end = Math.max(transcript.feature.end, feature.end);
+        }
+    }
+    return topLevel;
+}
+/**
+ * Parse GTF records and serialize each top-level feature. Shared by the
+ * plain-text and tabix adapters, which differ only in how they source records
+ * and derive the uniqueId (parse index vs. tabix byte offset).
+ */
+export function parseGtfToFeatures(records, makeUniqueId) {
+    return parseGtf(records).map(({ feature, record }, i) => featureData(feature, makeUniqueId(record, i)));
+}
+/** The value if it is a non-empty string, otherwise undefined. */
+function nonEmptyString(value) {
+    return typeof value === 'string' && value.length > 0 ? value : undefined;
+}
+/**
+ * GTF has no spanning gene line, so a gene is synthesized by grouping related
+ * transcript features and spanning them. An explicit `gene` line is held aside
+ * rather than aggregated — the parser hangs exons off the transcript, so a gene
+ * line is childless and the synthesized parent supersedes it — and emitted only
+ * if no synthesized gene claimed its `gene_id`, which is what keeps a file of
+ * nothing but `gene` rows from rendering as an empty track. A gene line with no
+ * `gene_id` at all (AUGUSTUS writes a bare `g1` in column 9, which parses to no
+ * attributes) has no key to be superseded by, and is dropped.
+ *
+ * Grouping is keyed by `gene_id` where a transcript carries one, falling back to
+ * the aggregate value; `aggregateField` supplies the gene's display `name`. The
+ * two are separate because neither alone works:
+ *
+ * - `gene_name` (the default aggregate field, and the only one suitable as a
+ *   label) is not unique within a reference sequence. A GENCODE chromosome holds
+ *   hundreds of separate genes named `U6`, `Y_RNA` or `5S_rRNA`, and keying on
+ *   the name alone fused all of them into one feature spanning the whole
+ *   chromosome. `gene_id` is unique per gene by the GTF spec, so it splits those
+ *   back apart while still merging the transcripts that really do share a gene.
+ * - Plenty of GTFs carry `gene_id` and no `gene_name` at all — UCSC's
+ *   `genePredToGtf` emits `gene_id "TP53"; transcript_id "NM_000546";` and
+ *   nothing else, and AUGUSTUS is the same shape. Keying only on the aggregate
+ *   value left those with no gene model whatsoever: every transcript passed
+ *   through bare. Keying on `gene_id` gives them genes, named by the `gene_id`
+ *   when there's no better label.
+ *
+ * A feature with neither key still passes through unchanged.
+ */
+export function aggregateGtfFeatures({ feats, aggregateField, refName, regionStart, regionEnd, idPrefix, }) {
+    const out = [];
+    const parentAggregation = new Map();
+    const explicitGenes = new Map();
+    for (const feat of feats) {
+        const geneId = nonEmptyString(feat.gene_id);
+        if (feat.type === 'gene') {
+            if (geneId !== undefined) {
+                explicitGenes.set(geneId, feat);
+            }
+            continue;
+        }
+        const name = nonEmptyString(feat[aggregateField]);
+        const key = geneId ?? name;
+        if (key !== undefined) {
+            let group = parentAggregation.get(key);
+            if (group) {
+                // a transcript that carries the aggregate value names the gene even
+                // when the first one grouped under this key didn't
+                group.name ??= name;
+            }
+            else {
+                group = { name, subfeatures: [] };
+                parentAggregation.set(key, group);
+            }
+            group.subfeatures.push(feat);
+        }
+        else if (
+        // a childless transcript with nothing to group on is a stray container
+        // line rather than a feature — AUGUSTUS's bare `transcript` line, whose
+        // real model arrives as the exon/CDS lines below it. One carrying a
+        // gene_id is a real transcript and was aggregated above; dropping those
+        // too made a transcript-only GTF render nothing at all
+        (feat.type !== 'transcript' || feat.subfeatures?.length) &&
+            doesIntersect2(feat.start, feat.end, regionStart, regionEnd)) {
+            // passthrough features (neither a gene_id nor an aggregate value) must
+            // be clipped to the original query too, else a redispatch's expanded
+            // fetch leaks features outside the view (aggregated genes are already
+            // intersection-checked)
+            out.push(feat);
+        }
+    }
+    for (const [geneId, feat] of explicitGenes) {
+        if (!parentAggregation.has(geneId) &&
+            doesIntersect2(feat.start, feat.end, regionStart, regionEnd)) {
+            // labeled the way a synthesized gene is: featureData names a feature from
+            // its transcript_id, which a gene line hasn't got
+            out.push({
+                ...feat,
+                name: nonEmptyString(feat.name) ??
+                    nonEmptyString(feat[aggregateField]) ??
+                    geneId,
+            });
+        }
+    }
+    for (const [key, { name, subfeatures }] of parentAggregation) {
+        let start = Number.POSITIVE_INFINITY;
+        let end = Number.NEGATIVE_INFINITY;
+        for (const f of subfeatures) {
+            if (f.start < start) {
+                start = f.start;
+            }
+            if (f.end > end) {
+                end = f.end;
+            }
+        }
+        if (doesIntersect2(start, end, regionStart, regionEnd)) {
+            out.push({
+                type: 'gene',
+                subfeatures,
+                strand: subfeatures[0].strand,
+                // falls back to the grouping key, so a gene_id-only file (UCSC,
+                // AUGUSTUS) still gets a labeled gene rather than an unnamed one
+                name: name ?? key,
+                start,
+                end,
+                refName,
+                // stable across fetch windows: the same gene keeps one id while panning,
+                // unlike a subfeature-derived id where the "first" transcript (and thus
+                // the id) shifts with whatever the current window happened to pull in.
+                // the grouping key is unique within a ref, so refName+key is unique
+                uniqueId: `${idPrefix}-${refName}-gene-${key}`,
+            });
+        }
     }
-    return f;
+    return out;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jbrowse/plugin-gtf",
-  "version": "4.3.0",
+  "version": "5.0.0-beta.1",
   "type": "module",
   "description": "JBrowse 2 gtf feature adapter",
   "keywords": [
@@ -21,19 +21,13 @@
     "esm"
   ],
   "dependencies": {
-    "@flatten-js/interval-tree": "^2.0.3",
-    "@gmod/bgzf-filehandle": "^6.0.19",
-    "@jbrowse/mobx-state-tree": "^5.6.0",
-    "@mui/material": "^7.3.11",
-    "gtf-nostream": "^1.3.4",
-    "mobx": "^6.15.3",
-    "mobx-react": "^9.2.1",
+    "@gmod/tabix": "^3.8.2",
+    "@jbrowse/mobx-state-tree": "^6.5.1",
     "rxjs": "^7.8.2",
-    "@jbrowse/core": "^4.3.0",
-    "@jbrowse/plugin-linear-genome-view": "^4.3.0"
+    "@jbrowse/core": "^5.0.0-beta.1"
   },
   "peerDependencies": {
-    "react": ">=18.0.0"
+    "react": ">=19.0.0"
   },
   "publishConfig": {
     "access": "public"
@@ -41,16 +35,13 @@
   "sideEffects": false,
   "scripts": {
     "build": "pnpm run /^build:/",
-    "test": "cd ../..; jest --passWithNoTests plugins/gtf --passWithNoTests",
+    "test": "cd ../..; jest --passWithNoTests plugins/gtf",
     "prebuild": "pnpm clean",
-    "build:esm": "tsc -p tsconfig.build.esm.json",
+    "build:esm": "node ../../node_modules/typescript7/bin/tsc --build tsconfig.build.esm.json",
     "clean": "rimraf esm *.tsbuildinfo"
   },
   "types": "esm/index.d.ts",
   "exports": {
-    ".": {
-      "types": "./esm/index.d.ts",
-      "import": "./esm/index.js"
-    }
+    ".": "./esm/index.js"
   }
 }

package/esm/GuessAdapter/index.d.ts

@@ -1,2 +0,0 @@
-import type PluginManager from '@jbrowse/core/PluginManager';
-export default function GuessAdapterF(pluginManager: PluginManager): void;