fix(search): read hemisphere letters separated from the number (#1340)

Support spaced hemisphere coordinates while rejecting malformed numeric fragments and ambiguous component assignments. Include regression tests and runtime documentation.

Contribution by @rudycelekli.

changelog.d/1340-search-separated-hemisphere.md

@@ -0,0 +1,3 @@
+- Read a hemisphere letter that is separated from its number, so a search for
+  `40.7128 N 74.0060 W` flies there instead of falling through to the geocoders
+  (Rudy Celekli, @rudycelekli, #1340).

docs/CURRENT-STATE.md

@@ -768,6 +768,12 @@ transports and cancellation through response parsing. Stop and application
 teardown abort pending connections; reconnect requests a new client secret.
 Microphone, radio handoff, tool schemas and default model behavior are preserved.
 
+Coordinate search accepts whitespace-separated decimal degrees with prefix or
+suffix hemisphere letters, including `40.7128 N 74.0060 W`. Components require
+an actual separator and an unambiguous hemisphere assignment; malformed numeric
+fragments and scientific notation fall through to geocoders instead of becoming
+exact coordinate results.
+
 Geospatial lookups are composed through `src/search`: forward/reverse geocoding,
 text/nearby search and routing use configurable providers/endpoints. Annotation,
 HUD and voice consumers share the configured service. Existing Google/Photon

src/search/coordinateParser.js

@@ -47,16 +47,38 @@ function readComponent(text) {
   };
 }
 
-/** Split on a comma, a semicolon, or whitespace — never on anything else. */
+/**
+ * Split on a comma, a semicolon, or whitespace — never on anything else.
+ * A hemisphere letter stays with the number it names, so `40.7 N 74.0 W`
+ * is two components. Splitting on the space before that letter used to
+ * leave three pieces and the search declined the coordinate.
+ */
 function splitComponents(query) {
   const text = String(query).trim();
   if (!text) return null;
-  const parts = /[,;]/.test(text)
+  if (/[,;]/.test(text)) {
-    ? text.split(/[,;]/)
+    const trimmed = text.split(/[,;]/).map((part) => part.trim());
-    : text.split(/\s+(?=[+-]?[.\d]|[NSEW]\s*[+-]?[.\d])/i);
-  const trimmed = parts.map((part) => part.trim());
     return trimmed.length === 2 && trimmed.every(Boolean) ? trimmed : null;
   }
+  const words = text.split(/\s+/);
+  // Each component has at most a number, a degree sign and one hemisphere.
+  if (words.length > 6) return null;
+  // Keep every character: extracting tokens would turn `1e3` or `1.2.3`
+  // into invented coordinates. A lone letter between numbers can name either
+  // component; require an unambiguous split rather than guessing its owner.
+  let result = null;
+  for (let boundary = words.length - 1; boundary > 0; boundary -= 1) {
+    const parts = [
+      words.slice(0, boundary).join(' '),
+      words.slice(boundary).join(' '),
+    ];
+    if (parts.every((part) => readComponent(part))) {
+      if (result) return null;
+      result = parts;
+    }
+  }
+  return result;
+}
 
 /** Present a parsed pair the way a coordinate is normally written. */
 export function formatCoordinateLabel(lat, lon) {

src/search/coordinateParser.test.mjs

@@ -52,6 +52,19 @@ test('hemisphere letters fix the axis, in either order', () => {
     label: '40.0000° N, 74.0000° W',
   });
   assert.deepEqual(parseCoordinateQuery('40° N, 74° W')?.lat, 40);
+  // Written the way a coordinate is usually pasted, with no comma.
+  assert.deepEqual(parseCoordinateQuery('40.7128 N 74.0060 W'), {
+    lat: 40.7128,
+    lon: -74.006,
+    label: '40.7128° N, 74.0060° W',
+  });
+  assert.deepEqual(parseCoordinateQuery('40.7128° N 74.0060° W')?.lon, -74.006);
+  assert.deepEqual(parseCoordinateQuery('N 40.7128 W 74.0060'), {
+    lat: 40.7128,
+    lon: -74.006,
+    label: '40.7128° N, 74.0060° W',
+  });
+  assert.deepEqual(parseCoordinateQuery('74.0060 W 40.7128 N')?.lat, 40.7128);
   // One letter is enough: it fixes its own axis, the other takes what is left.
   assert.deepEqual(parseCoordinateQuery('40 N, -74')?.lon, -74);
   assert.deepEqual(parseCoordinateQuery('-74, 40 N'), {
@@ -66,6 +79,21 @@ test('malformed and ambiguous coordinates are refused, never guessed', () => {
     // The reviewer's case: trailing text must not be discarded.
     '12junk, 34oops',
     '12junk,34',
+    '12junk 34',
+    '1e3 5',
+    '1e3',
+    '1E3',
+    '1e-3',
+    '1.2.3',
+    '12-34',
+    '40 NW 74',
+    '40 N 74',
+    '40 W 74',
+    '40 E 74',
+    '°40 74',
+    '40°° 74',
+    '40 74°°',
+    '40 26 46 N 79 58 56 W',
     '43.1731, -79.0384x',
     'abc, def',
     'Paris',
@@ -196,3 +224,25 @@ test('malformed or unsupported geo URIs are refused without losing their meaning
     assert.equal(parseCoordinateQuery(query), null, query);
   }
 });
+
+test('whitespace boundaries preserve degree signs and hemisphere ownership', () => {
+  for (const query of ['40 ° 74', '40° 74', '40 N, 74', 'N 40 74']) {
+    assert.equal(parseCoordinateQuery(query)?.lat, 40, query);
+    assert.equal(parseCoordinateQuery(query)?.lon, 74, query);
+  }
+  for (const query of ['40 ° N 74 ° W', 'N 40 ° W 74 °', '40\tN\n74\tW']) {
+    assert.equal(parseCoordinateQuery(query)?.lat, 40, query);
+    assert.equal(parseCoordinateQuery(query)?.lon, -74, query);
+  }
+});
+
+test('a signed plain component leaves a spaced letter with the other component', () => {
+  assert.deepEqual(
+    parseCoordinateQuery('-40 N 74'),
+    parseCoordinateQuery('-40, N 74'),
+  );
+  assert.deepEqual(
+    parseCoordinateQuery('40 N -74'),
+    parseCoordinateQuery('40 N, -74'),
+  );
+});