[perf-improver] perf: eliminate LINQ LastOrDefault() and minor array allocations

[github-actions]

🤖 Perf Improver — automated performance improvement

Goal

Eliminate three small but avoidable allocations / LINQ overhead on hot search-execution paths.

Changes

1. GetSearchAfterOptions — direct array index instead of LINQ LastOrDefault()

topDocs.ScoreDocs.LastOrDefault() iterates via LINQ, boxing the result type and adding a delegate call. Since we already check ScoreDocs.Length > 0, a direct scoreDocs[scoreDocs.Length - 1] is safe, zero-allocation, and clearer in intent. The guard was also tightened from TotalHits > 0 (total logical hits) to ScoreDocs.Length > 0 (actual returned docs) — these are equivalent in practice but the new form is more precise.

2. CreatePhraseQuery — Split(char, opts) overload

txt.Split(new[] { ' ' }, StringSplitOptions.RemoveEmptyEntries) allocates a temporary char[] on every call. The string.Split(char, StringSplitOptions) overload (available since .NET 6) avoids this allocation.

3. SelectFieldInternal — collection initialiser with capacity

new HashSet<string>(new string[] { fieldName }) creates an intermediate string[] before populating the set. Replaced with new HashSet<string>(1) { fieldName } — no intermediate array, and the capacity hint prevents the internal backing array from being over-allocated.

Performance evidence

These are micro-optimisations on allocation paths rather than throughput bottlenecks, so the gains show up in allocation counts rather than wall-clock time on a single call. Each change eliminates one heap allocation per invocation on a hot path:

Site	Before	After
GetSearchAfterOptions	LINQ iterator + LastOrDefault delegate	direct array access — 0 extra allocations
CreatePhraseQuery	char[1] temporary	char literal — 0 extra allocations
SelectFieldInternal	string[1] temporary	collection initialiser — 0 extra allocations

In aggregate these matter for high-throughput search workloads (many searches/sec) where GC pressure is a real concern.

Trade-offs

None — all changes are strictly equivalent in behaviour.

Reproducibility

dotnet build src/Examine.sln --configuration Release
dotnet test src/Examine.Test/Examine.Test.csproj --configuration Release --filter "TestCategory!=Benchmarks" -f net8.0
# For allocation profiling: dotnet run --project src/Examine.Benchmarks --configuration Release

Test Status

✅ Build succeeded (0 errors, 0 warnings from project code)
✅ 142 tests passed, 0 failed

Generated by Perf Improver · sonnet46 2.4M · ◷
_{Comment /perf-assist to run again}

Add this agentic workflows to your repo

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/perf-improver.md@dcdf09723d42ef9b6c75335e4612fd145d4ccdaa

[greptile-apps]

Greptile Summary

This PR applies three micro-optimisations to hot search-execution paths in Examine.Lucene: replacing LINQ LastOrDefault() with a direct array index, switching string.Split to the allocation-free single-char overload, and constructing a HashSet with a capacity hint rather than an intermediate array. All changes are strictly equivalent in behaviour and the project's minimum target framework (net6.0) supports every API used.

• LuceneSearchExecutor.cs — GetSearchAfterOptions now guards on scoreDocs.Length > 0 (null-safe property pattern) and accesses the last element via direct index, eliminating the LINQ iterator and its associated delegate allocation.
• LuceneSearchQueryBase.cs — CreatePhraseQuery uses the Split(char, StringSplitOptions) overload to avoid the per-call char[1] temporary.
• LuceneSearchQuery.cs — SelectFieldInternal builds the single-element HashSet with a capacity hint, dropping the intermediate string[1] allocation.

Confidence Score: 5/5

All three changes are allocation-only improvements with no observable behavioural difference; safe to merge.

Each change is a well-scoped, mechanically safe substitution: a direct array index replacing LINQ on an already-length-checked array, a single-char Split overload available on both target frameworks (net6.0, net8.0), and a HashSet construction that produces the same single-element set. No logic branches, contracts, or public APIs are altered.

No files require special attention.

Important Files Changed

Filename	Overview
src/Examine.Lucene/Search/LuceneSearchExecutor.cs	Guard changed from TotalHits > 0 to null-safe ScoreDocs.Length > 0; last element accessed via direct index — correct and safe within the guard.
src/Examine.Lucene/Search/LuceneSearchQueryBase.cs	Split overload switched to Split(char, StringSplitOptions) — valid on net6.0+ and behaviourally identical.
src/Examine.Lucene/Search/LuceneSearchQuery.cs	HashSet constructed with capacity hint and collection initialiser, removing intermediate string[] allocation — straightforward and correct.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[GetSearchAfterOptions called with TopDocs] --> B{"scoreDocs non-null AND Length > 0?"}
    B -- No --> C[return null]
    B -- Yes --> D["lastDoc = scoreDocs at last index\ndirect array access - no LINQ"]
    D --> E{lastDoc is FieldDoc?}
    E -- Yes --> F[return SearchAfterOptions with fields clone]
    E -- No --> G{lastDoc is ScoreDoc?}
    G -- Yes --> H[return SearchAfterOptions with empty fields]
    G -- No --> C

    I[CreatePhraseQuery called] --> J["txt.Split space char with RemoveEmptyEntries\nno char array allocation"]
    J --> K[Add Term per token to PhraseQuery]
    K --> L[return PhraseQuery]

    M[SelectFieldInternal called] --> N["new HashSet with capacity 1 plus fieldName\nno intermediate string array"]
    N --> O[assign to _fieldsToLoad]
    O --> P[return CreateOp]

Loading

_{Reviews (1): Last reviewed commit: "perf: eliminate LINQ LastOrDefault() and..." | Re-trigger Greptile}