263 lines
12 KiB
TypeScript
263 lines
12 KiB
TypeScript
/* PHP symbol index + reference lookup.
|
|
*
|
|
* Two jobs, both on ripgrep:
|
|
* - the index: every class/interface/trait/enum DECLARED in the project AND in
|
|
* vendor (so a framework class is clickable too), as
|
|
* name → {path, line}. The renderer needs the bare name list to decide which
|
|
* tokens it may underline, and that decision is per visible token per frame,
|
|
* so it cannot be a search. One rg pass answers it for every token at once.
|
|
* - the lookup: on ⌘-click, the declarations of one name plus every reference
|
|
* to it, found there and then. Nothing about usages is cached.
|
|
*
|
|
* The index is built lazily and never on the startup path: the first caller
|
|
* starts it and later callers join the same promise. A file change drops it, so
|
|
* the next caller pays for the rebuild (~100 ms) instead of the window opening.
|
|
*/
|
|
import { spawn } from 'node:child_process'
|
|
import { relative, sep } from 'node:path'
|
|
import { getConfig } from './config'
|
|
import { log } from './logger'
|
|
|
|
export interface SymbolDef {
|
|
name: string; path: string; line: number; kind: string
|
|
/** The file's `namespace`, so the popup can name the class in full. */
|
|
ns: string
|
|
/** Parents named on the declaration line: `extends A`, `implements B, C`. */
|
|
parents: string[]
|
|
interfaces: string[]
|
|
/** Traits mixed in by an indented `use A, B;` inside this declaration's body. */
|
|
traits: string[]
|
|
}
|
|
export interface RefHit { no: number; ln: string; ix: number }
|
|
export interface RefGroup { path: string; hits: RefHit[] }
|
|
export interface SymbolLookup { name: string; defs: SymbolDef[]; refs: RefGroup[]; refCount: number }
|
|
|
|
/** Same resolution dance as search-service: @vscode/ripgrep is ESM, and the
|
|
* packaged binary lives outside the asar. */
|
|
const rgPathPromise: Promise<string | null> = (async () => {
|
|
try {
|
|
const mod = await import('@vscode/ripgrep')
|
|
let p = (mod as { rgPath: string }).rgPath
|
|
if (p) p = p.replace(/\bapp\.asar\b/, 'app.asar.unpacked')
|
|
return p || null
|
|
} catch (e) {
|
|
log('error', 'symbols', 'ripgrep unavailable', (e as Error).message)
|
|
return null
|
|
}
|
|
})()
|
|
|
|
const BASE_IGNORE = [
|
|
'node_modules', '.git', 'out', 'dist', 'build', '.cache',
|
|
'vendor', 'coverage', '.helder', '.next', '.nuxt', '.turbo', '.idea', '.vscode',
|
|
]
|
|
/** `vendor/` is out of the reference search — nobody wants to read how Symfony
|
|
* uses Throwable — but it IS indexed for declarations, so a framework class is
|
|
* still a name we can underline and open. Measured on a Laravel app: 48 project
|
|
* classes in 30 ms, 7k more with vendor in 170 ms. */
|
|
const DECL_IGNORE = BASE_IGNORE.filter((d) => d !== 'vendor')
|
|
|
|
/** Declarations only. `readonly`/`final`/`abstract` may precede the keyword, and
|
|
* an enum may carry a backing type. Anchored at the line start (with optional
|
|
* indent) so `$x instanceof class` shaped text cannot match. */
|
|
const DECL_RE = /^[ \t]*(?:(?:final|abstract|readonly)[ \t]+)*(class|interface|trait|enum)[ \t]+([A-Za-z_]\w*)(.*)$/
|
|
/** Declarations and the file's namespace in one pass: ripgrep emits a file's
|
|
* matches together and in line order, and `namespace` always precedes the
|
|
* declarations it covers, so the last one seen is the right one. */
|
|
const DECL_RG = '^\\s*(namespace\\s+[A-Za-z_\\\\][\\w\\\\]*\\s*;|(final\\s+|abstract\\s+|readonly\\s+)*(class|interface|trait|enum)\\s+\\w+)'
|
|
const NS_RE = /^[ \t]*namespace[ \t]+([A-Za-z_\\][\w\\]*)[ \t]*;/
|
|
/** An indented `use` — a trait mixed into a class body, not a namespace import. */
|
|
const TRAIT_RE = /^[ \t]+use[ \t]+([A-Za-z_\\][\w\\ \t,]*?)[ \t]*[;{]/
|
|
const TRAIT_RG = '^\\s+use\\s+[A-Za-z_\\\\][\\w\\\\ \t,]*[;{]'
|
|
|
|
const MAX_DEFS = 25
|
|
const MAX_REF_FILES = 300
|
|
const MAX_LINE = 1000
|
|
|
|
function ignoreArgs(dirs: string[] = BASE_IGNORE): string[] {
|
|
const cfg = getConfig()
|
|
const args: string[] = []
|
|
for (const d of dirs) args.push('--glob', `!${d}`)
|
|
for (const g of cfg.files.exclude) if (g) args.push('--glob', `!${g}`)
|
|
if (!cfg.files.followGitignore) args.push('--no-ignore')
|
|
return args
|
|
}
|
|
|
|
function toRel(root: string, p: string): string {
|
|
return relative(root, p).split(sep).join('/')
|
|
}
|
|
|
|
/** Bare class names out of an `extends`/`implements` clause: split on commas and
|
|
* drop the namespace, since the index is keyed on the short name. */
|
|
function clauseNames(clause: string): string[] {
|
|
return clause.split(',')
|
|
.map((part) => (part.trim().split('\\').pop() ?? '').trim())
|
|
.filter((n) => /^[A-Za-z_]\w*$/.test(n))
|
|
}
|
|
|
|
/** The namespace a `namespace X;` line declares, or null. */
|
|
export function parseNamespace(text: string): string | null {
|
|
const m = NS_RE.exec(text)
|
|
return m ? m[1].replace(/^\\/, '') : null
|
|
}
|
|
|
|
/** The traits an indented `use` line mixes in, or [] for anything else. A
|
|
* `use function …` or a closure's `use ($x)` is not a trait. */
|
|
export function parseTraitUse(text: string): string[] {
|
|
const m = TRAIT_RE.exec(text)
|
|
if (!m || /^\s*use\s+(function|const)\b/.test(text)) return []
|
|
return clauseNames(m[1])
|
|
}
|
|
|
|
/** The declaration on one line, or null. Exported for the unit tests.
|
|
* Only this line is read, so an `implements` list wrapped onto the next line is
|
|
* not seen — the popup then shows fewer parents, never wrong ones. */
|
|
export function parseDeclaration(text: string): { kind: string; name: string; parents: string[]; interfaces: string[] } | null {
|
|
const m = DECL_RE.exec(text)
|
|
if (!m) return null
|
|
const rest = m[3] ?? ''
|
|
const ext = /\bextends\s+([^{]*?)(?:\s+implements\b|\s*\{|$)/.exec(rest)
|
|
const impl = /\bimplements\s+([^{]*?)(?:\s*\{|$)/.exec(rest)
|
|
return {
|
|
kind: m[1],
|
|
name: m[2],
|
|
parents: ext ? clauseNames(ext[1]) : [],
|
|
interfaces: impl ? clauseNames(impl[1]) : [],
|
|
}
|
|
}
|
|
|
|
/** A namespace import — `use App\Models\Agent;` at column 0, with an optional
|
|
* alias or a group body. A `use` INSIDE a class body is indented and mixes a
|
|
* trait in, which is a real reference, so the indent is what separates them. */
|
|
export function isImportLine(text: string): boolean {
|
|
return /^use[ \t]+[^;]*;?[ \t\r]*$/.test(text)
|
|
}
|
|
|
|
/** Run rg with --json and hand every match line to `onMatch`. */
|
|
function rgJson(args: string[], onMatch: (path: string, line: number, text: string, col: number) => void): Promise<void> {
|
|
return new Promise((resolve) => {
|
|
rgPathPromise.then((rgPath) => {
|
|
if (!rgPath) { resolve(); return }
|
|
const child = spawn(rgPath, args)
|
|
let buf = ''
|
|
let done = false
|
|
const finish = (): void => { if (done) return; done = true; resolve() }
|
|
child.stdout.on('data', (chunk: Buffer) => {
|
|
buf += chunk.toString()
|
|
let nl: number
|
|
while ((nl = buf.indexOf('\n')) >= 0) {
|
|
const line = buf.slice(0, nl); buf = buf.slice(nl + 1)
|
|
if (!line) continue
|
|
let msg: { type: string; data: { path?: { text?: string }; lines?: { text?: string }; line_number?: number; submatches?: { start: number }[] } }
|
|
try { msg = JSON.parse(line) } catch { continue }
|
|
if (msg.type !== 'match') continue
|
|
const abs = msg.data.path?.text
|
|
const text = msg.data.lines?.text
|
|
if (!abs || text == null) continue
|
|
onMatch(abs, msg.data.line_number || 0, text.replace(/\n$/, ''), msg.data.submatches?.[0]?.start ?? 0)
|
|
}
|
|
})
|
|
child.on('close', finish)
|
|
child.on('error', (e) => { log('error', 'symbols', 'ripgrep failed', (e as Error).message); finish() })
|
|
})
|
|
})
|
|
}
|
|
|
|
interface Index { root: string; byName: Map<string, SymbolDef[]> }
|
|
let index: Index | null = null
|
|
let building: Promise<Index> | null = null
|
|
|
|
async function build(root: string): Promise<Index> {
|
|
const t0 = Date.now()
|
|
const byName = new Map<string, SymbolDef[]>()
|
|
const byFile = new Map<string, SymbolDef[]>()
|
|
const nsByFile = new Map<string, string>()
|
|
await rgJson(
|
|
['--json', '--hidden', '--glob', '*.php', ...ignoreArgs(DECL_IGNORE), '-e', DECL_RG, '--', root],
|
|
(abs, line, text) => {
|
|
const rel = toRel(root, abs)
|
|
const ns = parseNamespace(text)
|
|
if (ns != null) { nsByFile.set(rel, ns); return }
|
|
const d = parseDeclaration(text)
|
|
if (!d) return
|
|
const def: SymbolDef = {
|
|
name: d.name, path: rel, line, kind: d.kind, ns: nsByFile.get(rel) ?? '',
|
|
parents: d.parents, interfaces: d.interfaces, traits: [],
|
|
}
|
|
const list = byName.get(d.name)
|
|
if (list) list.push(def)
|
|
else byName.set(d.name, [def])
|
|
const inFile = byFile.get(def.path)
|
|
if (inFile) inFile.push(def)
|
|
else byFile.set(def.path, [def])
|
|
},
|
|
)
|
|
// Traits live in the body, not on the declaration line, so they need their own
|
|
// pass. A `use` belongs to the last declaration above it in the same file —
|
|
// which is also why the declaration pass has to run first.
|
|
await rgJson(
|
|
['--json', '--hidden', '--glob', '*.php', ...ignoreArgs(DECL_IGNORE), '-e', TRAIT_RG, '--', root],
|
|
(abs, line, text) => {
|
|
const names = parseTraitUse(text)
|
|
if (!names.length) return
|
|
const decls = byFile.get(toRel(root, abs))
|
|
if (!decls) return
|
|
let owner: SymbolDef | null = null
|
|
for (const d of decls) if (d.line < line && (!owner || d.line > owner.line)) owner = d
|
|
if (owner) for (const n of names) if (!owner.traits.includes(n)) owner.traits.push(n)
|
|
},
|
|
)
|
|
log('info', 'symbols', 'index built', { classes: byName.size, ms: Date.now() - t0 })
|
|
return { root, byName }
|
|
}
|
|
|
|
/** The index for `root`, built on first use. Concurrent callers share one build. */
|
|
function getIndex(root: string): Promise<Index> {
|
|
if (index && index.root === root) return Promise.resolve(index)
|
|
if (building) return building
|
|
building = build(root).then((ix) => { index = ix; building = null; return ix })
|
|
.catch((e) => { building = null; throw e })
|
|
return building
|
|
}
|
|
|
|
/** Drop the index after a file change. The next caller rebuilds it. */
|
|
export function invalidateSymbols(): void {
|
|
index = null
|
|
}
|
|
|
|
/** Every declared name. The renderer keeps this as a Set to mark tokens. */
|
|
export async function symbolNames(root: string): Promise<string[]> {
|
|
const ix = await getIndex(root)
|
|
return [...ix.byName.keys()]
|
|
}
|
|
|
|
/** Declarations of `name` plus every reference to it, grouped by file. Namespace
|
|
* imports are dropped (noise), and so are the declaration lines themselves —
|
|
* they are already the first section of the popup. */
|
|
export async function lookupSymbol(root: string, name: string): Promise<SymbolLookup> {
|
|
if (!/^[A-Za-z_]\w*$/.test(name)) return { name, defs: [], refs: [], refCount: 0 }
|
|
const ix = await getIndex(root)
|
|
// The project's own declaration comes first: a common name like `Handler` is
|
|
// declared a dozen times inside vendor, and none of those is what was clicked.
|
|
const defs = (ix.byName.get(name) ?? [])
|
|
.slice()
|
|
.sort((a, b) => Number(a.path.startsWith('vendor/')) - Number(b.path.startsWith('vendor/')) || a.path.localeCompare(b.path))
|
|
.slice(0, MAX_DEFS)
|
|
const declared = new Set(defs.map((d) => d.path + ':' + d.line))
|
|
const order: string[] = []
|
|
const groups = new Map<string, RefGroup>()
|
|
let refCount = 0
|
|
await rgJson(
|
|
['--json', '--hidden', '--word-regexp', '--glob', '*.php', ...ignoreArgs(), '-e', name, '--', root],
|
|
(abs, line, text, col) => {
|
|
const rel = toRel(root, abs)
|
|
if (declared.has(rel + ':' + line) || isImportLine(text)) return
|
|
let g = groups.get(rel)
|
|
if (!g) { if (groups.size >= MAX_REF_FILES) return; g = { path: rel, hits: [] }; groups.set(rel, g); order.push(rel) }
|
|
const ln = text.slice(0, MAX_LINE)
|
|
g.hits.push({ no: line, ln, ix: Math.min(col, ln.length) })
|
|
refCount++
|
|
},
|
|
)
|
|
return { name, defs, refs: order.map((p) => groups.get(p) as RefGroup), refCount }
|
|
}
|