@preventive/triage 1.0.0-alpha.15 → 1.0.0-alpha.17

package/common/scan-server.ts

@@ -0,0 +1,13 @@
+export const DEFAULT_SCAN_SERVER = 'http://127.0.0.1:3123/'
+
+// Discovery is optional: malformed scan configuration must never prevent
+// opening local data or recognizing the server's sync protocol.
+export function normalizeScanServer(value: unknown): string | null {
+  if (typeof value !== 'string' || !value.trim()) return null
+  try {
+    const url = new URL(value)
+    if (!['http:', 'https:'].includes(url.protocol) || url.username || url.password || url.search || url.hash) return null
+    url.pathname = url.pathname.replace(/\/+$/u, '') + '/'
+    return url.href
+  } catch { return null }
+}

package/common/server-info.ts

@@ -22,6 +22,7 @@ export interface ManagedServerInfo {
 export interface ServerInfo {
   mode: ServerMode
   managed: ManagedServerInfo | null
+  deepviewScanServer?: string
 }
 
 // The mode-probe route. A client GETs this to learn a server's protocol up

package/common/utf8.d.ts

@@ -11,3 +11,6 @@
 // narrower declaration is safe.
 export function encodeUtf8(str: string): Uint8Array<ArrayBuffer>
 export function decodeUtf8(bytes: Uint8Array | ArrayBuffer | ArrayBufferView): string
+// Bytes `str` takes in UTF-8, measured without encoding when no character
+// is past \xFF. A lone surrogate counts as the U+FFFD it encodes to.
+export function utf8ByteLength(str: string): number

package/common/utf8.js

@@ -43,6 +43,51 @@ export function encodeUtf8(str) {
   return encoder.encode(str)
 }
 
+// ASCII characters in a row after which the count hands back to the
+// native scan. A jump costs a call and a step does not, so it pays only
+// across a real gap: at 1, alternating text (`aéaé…`) takes a jump per
+// character, and past 2, text with an accent every few words steps
+// through ASCII the scan crosses faster. Measured on 50 MB of each, 2 was
+// never more than 1.5x the best of the thresholds tried, 1 to 32.
+const ASCII_RUN = 2
+
+// The number of bytes `str` takes in UTF-8, without encoding it where
+// that can be helped. A string with no character past \xFF is Latin-1:
+// ASCII takes a byte and \x80-\xFF two, so its size is its length plus
+// its high characters. An engine stores such a string a byte per
+// character, where no character past \xFF can be, so the first test is
+// answered without reading it. Only a string with a wider character is
+// encoded to be measured.
+//
+// The high characters are counted without collecting them: a match array
+// takes an entry apiece, which for Latin-1-heavy text outweighs the
+// encoding this avoids. A native scan jumps to the next one — for ASCII
+// it finds none — and a loop counts on from there while they keep coming,
+// handing back to the scan after a run of ASCII. Sparse text is read
+// natively, dense text by the loop, and neither allocates.
+//
+// This sizes text for display, so unlike `encodeUtf8` it does not refuse
+// a lone surrogate: it counts the U+FFFD that TextEncoder writes for one.
+export function utf8ByteLength(str) {
+  if (typeof str !== 'string') {
+    throw new TypeError(`utf8ByteLength expects a string, got ${typeof str}`)
+  }
+  // Code units, not code points: a `u` regexp reads a two-byte string by
+  // code point, several times slower, and a surrogate is past \xFF either way.
+  // eslint-disable-next-line require-unicode-regexp
+  if (/[\u0100-\uFFFF]/.test(str)) return encoder.encode(str).byteLength
+  const high = /[\u0080-\u00FF]/gu
+  let bytes = str.length
+  while (high.test(str)) {
+    let i = high.lastIndex - 1
+    for (let ascii = 0; i < str.length && ascii < ASCII_RUN; i++) {
+      if (str.codePointAt(i) > 0x7F) { bytes++; ascii = 0 } else ascii++
+    }
+    high.lastIndex = i
+  }
+  return bytes
+}
+
 export function decodeUtf8(bytes) {
   // Reject non-BufferSource so a missed destructure / optional field /
   // misnamed property surfaces here instead of silently defaulting to