Fix per-process randomized string hashing in test-framework codec/similarity selection

MockRandomPostingsFormat, RandomCodec, and RandomSimilarityProvider chose a
per-file/per-field int-stream factory, postings/docvalues format, and similarity
by indexing into a collection with `string.GetHashCode()`. On .NET, string hash
codes are randomized per process, so a choice made when writing an index (e.g. in
the forked child process of TestIndexWriterOnJRECrash) differed from the choice
made when reading it back in another process, producing spurious "codec header
mismatch" / CheckIndex failures. Java is unaffected because String.hashCode() is
deterministic across JVMs.

Use J2N's CharSequenceComparer.Ordinal.GetHashCode(), which is deterministic and
produces the same value as Java's String.hashCode(), keeping writer and reader in
sync. This was a pre-existing test-framework bug (reproduces on master).

Verified: TestIndexWriterOnJRECrash.TestNRTThreads_Mem (previously a deterministic
failure under seed 0x03c0703dd1d74190:0xfc8ded704ed908f9) now passes on fixed and
random seeds; RandomCodec- and similarity-heavy suites remain green. Confirmed
against a native build of upstream Lucene 4.8.1, which passes this test.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: paulirwin

src/Lucene.Net.TestFramework/Codecs/MockRandom/MockRandomPostingsFormat.cs

@@ -1,3 +1,4 @@
+using J2N.Text;
 using Lucene.Net.Codecs.BlockTerms;
 using Lucene.Net.Codecs.Lucene41;
 using Lucene.Net.Codecs.Memory;
@@ -97,7 +98,12 @@ public override Int32IndexInput OpenInput(Directory dir, string fileName, IOCont
             {
                 // Must only use extension, because IW.addIndexes can
                 // rename segment!
-                Int32StreamFactory f = delegates[(Math.Abs(salt ^ GetExtension(fileName).GetHashCode())) % delegates.Count];
+                // LUCENENET specific: use J2N's CharSequenceComparer.Ordinal.GetHashCode(), which produces
+                // the same value as Java's String.hashCode(). The .NET string.GetHashCode() is randomized
+                // per-process, so the factory chosen when writing (e.g. in a forked child process, as in
+                // TestIndexWriterOnJRECrash) would differ from the one chosen when reading in another process,
+                // causing "codec header mismatch" failures. A deterministic hash keeps writer and reader in sync.
+                Int32StreamFactory f = delegates[(Math.Abs(salt ^ CharSequenceComparer.Ordinal.GetHashCode(GetExtension(fileName)))) % delegates.Count];
                 if (LuceneTestCase.Verbose)
                 {
                     Console.WriteLine("MockRandomCodec: read using int factory " + f + " from fileName=" + fileName);
@@ -107,7 +113,8 @@ public override Int32IndexInput OpenInput(Directory dir, string fileName, IOCont
 
             public override Int32IndexOutput CreateOutput(Directory dir, string fileName, IOContext context)
             {
-                Int32StreamFactory f = delegates[(Math.Abs(salt ^ GetExtension(fileName).GetHashCode())) % delegates.Count];
+                // LUCENENET specific: deterministic (Java-parity) hash; see the note in OpenInput().
+                Int32StreamFactory f = delegates[(Math.Abs(salt ^ CharSequenceComparer.Ordinal.GetHashCode(GetExtension(fileName)))) % delegates.Count];
                 if (LuceneTestCase.Verbose)
                 {
                     Console.WriteLine("MockRandomCodec: write using int factory " + f + " to fileName=" + fileName);

src/Lucene.Net.TestFramework/Index/RandomCodec.cs

@@ -1,3 +1,4 @@
+using J2N.Text;
 using Lucene.Net.Codecs;
 using Lucene.Net.Codecs.Asserting;
 using Lucene.Net.Codecs.Bloom;
@@ -84,11 +85,16 @@ public override PostingsFormat GetPostingsFormatForField(string name)
         {
             if (!previousMappings.TryGetValue(name, out PostingsFormat codec) || codec is null)
             {
-                codec = formats[Math.Abs(perFieldSeed ^ name.GetHashCode()) % formats.Count];
+                // LUCENENET specific: use J2N's CharSequenceComparer.Ordinal.GetHashCode(), which produces
+                // the same value as Java's String.hashCode(). The .NET string.GetHashCode() is randomized
+                // per-process, so a field's format chosen when writing (e.g. in a forked child process, as
+                // in TestIndexWriterOnJRECrash) would differ from the one chosen when reading in another
+                // process, causing "codec header mismatch" failures. A deterministic hash keeps them in sync.
+                codec = formats[Math.Abs(perFieldSeed ^ CharSequenceComparer.Ordinal.GetHashCode(name)) % formats.Count];
                 if (codec is SimpleTextPostingsFormat && perFieldSeed % 5 != 0)
                 {
                     // make simpletext rarer, choose again
-                    codec = formats[Math.Abs(perFieldSeed ^ name.ToUpperInvariant().GetHashCode()) % formats.Count];
+                    codec = formats[Math.Abs(perFieldSeed ^ CharSequenceComparer.Ordinal.GetHashCode(name.ToUpperInvariant())) % formats.Count];
                 }
                 previousMappings[name] = codec;
                 // Safety:
@@ -107,11 +113,12 @@ public override DocValuesFormat GetDocValuesFormatForField(string name)
         {
             if (!previousDVMappings.TryGetValue(name, out DocValuesFormat codec) || codec is null)
             {
-                codec = dvFormats[Math.Abs(perFieldSeed ^ name.GetHashCode()) % dvFormats.Count];
+                // LUCENENET specific: deterministic (Java-parity) hash; see the note in GetPostingsFormatForField().
+                codec = dvFormats[Math.Abs(perFieldSeed ^ CharSequenceComparer.Ordinal.GetHashCode(name)) % dvFormats.Count];
                 if (codec is SimpleTextDocValuesFormat && perFieldSeed % 5 != 0)
                 {
                     // make simpletext rarer, choose again
-                    codec = dvFormats[Math.Abs(perFieldSeed ^ name.ToUpperInvariant().GetHashCode()) % dvFormats.Count];
+                    codec = dvFormats[Math.Abs(perFieldSeed ^ CharSequenceComparer.Ordinal.GetHashCode(name.ToUpperInvariant())) % dvFormats.Count];
                 }
                 previousDVMappings[name] = codec;
                 // Safety:

src/Lucene.Net.TestFramework/Search/RandomSimilarityProvider.cs

@@ -1,4 +1,5 @@
 using J2N.Collections.Generic.Extensions;
+using J2N.Text;
 using Lucene.Net.Diagnostics;
 using Lucene.Net.Search.Similarities;
 using Lucene.Net.Support.Threading;
@@ -78,7 +79,12 @@ public override Similarity Get(string field)
                 if (Debugging.AssertsEnabled) Debugging.Assert(field != null);
                 if (!previousMappings.TryGetValue(field, out Similarity sim) || sim is null)
                 {
-                    sim = knownSims[Math.Max(0, Math.Abs(perFieldSeed ^ field.GetHashCode())) % knownSims.Count];
+                    // LUCENENET specific: use J2N's CharSequenceComparer.Ordinal.GetHashCode(), which produces
+                    // the same value as Java's String.hashCode(). The .NET string.GetHashCode() is randomized
+                    // per-process, so the per-field similarity chosen in one process would differ from another,
+                    // breaking reproducibility across processes (e.g. the forked TestIndexWriterOnJRECrash) and
+                    // diverging from Java. A deterministic hash keeps the choice stable.
+                    sim = knownSims[Math.Max(0, Math.Abs(perFieldSeed ^ CharSequenceComparer.Ordinal.GetHashCode(field))) % knownSims.Count];
                     previousMappings[field] = sim;
                 }
                 return sim;