ESQL: Workaround for RLike handling of empty lang pattern (elastic#128895)

Lucene's `org.apache.lucene.util.automaton.Operations#getSingleton` fails with an Automaton for a `REGEXP_EMPTY` `RegExp`. This adds a workaround for that, to check the type of automaton before calling into that failing method.

Closes elastic#128813

Author: bpintea

docs/changelog/128895.yaml

@@ -0,0 +1,6 @@
+pr: 128895
+summary: Workaround for RLike handling of empty lang pattern
+area: ES|QL
+type: bug
+issues:
+ - 128813

x-pack/plugin/esql-core/src/main/java/org/elasticsearch/xpack/esql/core/expression/predicate/regex/AbstractStringPattern.java

@@ -32,7 +32,11 @@ public boolean matchesAll() {
 
     @Override
     public String exactMatch() {
-        IntsRef singleton = Operations.getSingleton(automaton());
+        Automaton a = automaton();
+        if (a.getNumStates() == 0) { // workaround for https://github.com/elastic/elasticsearch/pull/128887
+            return null; // Empty automaton has no matches
+        }
+        IntsRef singleton = Operations.getSingleton(a);
         return singleton != null ? UnicodeUtil.newString(singleton.ints, singleton.offset, singleton.length) : null;
     }
 }

x-pack/plugin/esql-core/src/test/java/org/elasticsearch/xpack/esql/core/expression/predicate/regex/StringPatternTests.java

@@ -33,8 +33,12 @@ private boolean rlikeMatchesAll(String pattern) {
         return rlike(pattern).matchesAll();
     }
 
+    private String exactMatchRLike(String pattern) {
+        return rlike(pattern).exactMatch();
+    }
+
     private boolean rlikeExactMatch(String pattern) {
-        return pattern.equals(rlike(pattern).exactMatch());
+        return pattern.equals(exactMatchRLike(pattern));
     }
 
     public void testWildcardMatchAll() {
@@ -86,4 +90,20 @@ public void testRegexExactMatch() {
         assertTrue(rlikeExactMatch("abc"));
         assertTrue(rlikeExactMatch("12345"));
     }
+
+    public void testRegexExactMatchWithEmptyMatch() {
+        // As soon as there's one no conditional `#` in the pattern, it'll match nothing
+        assertNull(exactMatchRLike("#"));
+        assertNull(exactMatchRLike("##"));
+        assertNull(exactMatchRLike("#foo"));
+        assertNull(exactMatchRLike("#foo#"));
+        assertNull(exactMatchRLike("f#oo"));
+        assertNull(exactMatchRLike("foo#"));
+        assertNull(exactMatchRLike("#[A-Z]*"));
+        assertNull(exactMatchRLike("foo(#)"));
+
+        assertNotNull(exactMatchRLike("foo#?"));
+        assertNotNull(exactMatchRLike("#|foo"));
+        assertNotNull(exactMatchRLike("foo|#"));
+    }
 }

x-pack/plugin/esql-core/src/test/java/org/elasticsearch/xpack/esql/core/util/TestUtils.java

@@ -66,8 +66,8 @@ public static String stripThrough(String input) {
     /** Returns the input string, but with parts of it having the letter casing changed. */
     public static String randomCasing(String input) {
         StringBuilder sb = new StringBuilder(input.length());
-        for (int i = 0, inputLen = input.length(), step = (int) Math.sqrt(inputLen), chunkEnd; i < inputLen; i += step) {
-            chunkEnd = Math.min(i + step, inputLen);
+        for (int i = 0, inputLen = input.length(), step = (int) Math.sqrt(inputLen); i < inputLen; i += step) {
+            var chunkEnd = Math.min(i + step, inputLen);
             var chunk = input.substring(i, chunkEnd);
             sb.append(randomBoolean() ? chunk.toLowerCase(Locale.ROOT) : chunk.toUpperCase(Locale.ROOT));
         }

x-pack/plugin/esql/qa/server/src/main/java/org/elasticsearch/xpack/esql/qa/rest/RestEsqlTestCase.java

@@ -1440,6 +1440,46 @@ public void testReplaceStringCasingWithInsensitiveWildcardMatch() throws IOExcep
         assertThat(answer.get("values"), equalTo(List.of(List.of("_\"_$_(_)_+_._[_]_^_{_|_}___", "_#_&_<_>___"))));
     }
 
+    public void testRLikeHandlingOfEmptyLanguagePattern() throws IOException {
+        createIndex(testIndexName(), Settings.EMPTY, """
+            {
+                "properties": {
+                    "field": {
+                        "type": "keyword"
+                    }
+                }
+            }
+            """);
+        for (var val : List.of("#", "foo#bar")) {
+            Request doc = new Request("POST", testIndexName() + "/_doc?refresh=true");
+            doc.setJsonEntity("""
+                {
+                    "field": "%s"
+                }
+                """.formatted(val));
+            client().performRequest(doc);
+        }
+        // pushed down, matches nothing
+        var query = "FROM " + testIndexName() + " | WHERE TO_LOWER(field) RLIKE \"#\"";
+        var answer = runEsql(requestObjectBuilder().query(query));
+        assertThat(answer.get("values"), equalTo(List.of()));
+
+        // matches nothing
+        query = "FROM " + testIndexName() + " | WHERE field RLIKE \"#\"";
+        answer = runEsql(requestObjectBuilder().query(query));
+        assertThat(answer.get("values"), equalTo(List.of()));
+
+        // matches one doc
+        query = "FROM " + testIndexName() + " | WHERE field RLIKE \"\\\\#\"";
+        answer = runEsql(requestObjectBuilder().query(query));
+        assertThat(answer.get("values"), equalTo(List.of(List.of("#"))));
+
+        // matches both docs
+        query = "FROM " + testIndexName() + " | WHERE field RLIKE \".*\\\\#.*\" | SORT field";
+        answer = runEsql(requestObjectBuilder().query(query));
+        assertThat(answer.get("values"), equalTo(List.of(List.of("#"), List.of("foo#bar"))));
+    }
+
     protected static Request prepareRequestWithOptions(RequestObjectBuilder requestObject, Mode mode) throws IOException {
         requestObject.build();
         Request request = prepareRequest(mode);

x-pack/plugin/esql/qa/testFixtures/src/main/resources/where-like.csv-spec

@@ -423,3 +423,13 @@ emp_no:integer |first_name:keyword
 10001          |Georgi
 10055          |Georgy
 ;
+
+# test for https://github.com/elastic/elasticsearch/issues/128813
+rlikeWithEmptyLanguagePattern
+required_capability: rlike_with_empty_language_pattern
+ROW x = "abc" | EVAL bool = x RLIKE "#"
+;
+
+x:keyword | bool:boolean
+abc       | false
+;

x-pack/plugin/esql/src/main/java/org/elasticsearch/xpack/esql/action/EsqlCapabilities.java

@@ -1177,6 +1177,11 @@ public enum Cap {
          */
         ENABLE_LOOKUP_JOIN_ON_ALIASES,
 
+        /**
+         * Allows RLIKE to correctly handle the "empty language" flag, `#`.
+         */
+        RLIKE_WITH_EMPTY_LANGUAGE_PATTERN,
+
         /**
          * MATCH PHRASE function
          */