From 209e4a3ffd41757cc5bb9b53fcf4dcb2b459ae20 Mon Sep 17 00:00:00 2001
From: Thomas Clegg <t@clegg.dev>
Date: Wed, 5 Aug 2026 17:03:34 -0500
Subject: [PATCH] test: measure the parse, validate and format phases
 separately

---
 .../PhoneNumberWorkflowBenchmark.cs           | 62 +++++++++++++++++++
 csharp/PhoneNumbers.PerformanceTest/README.md |  5 +-
 2 files changed, 66 insertions(+), 1 deletion(-)

diff --git a/csharp/PhoneNumbers.PerformanceTest/Benchmarks/PhoneNumberWorkflowBenchmark.cs b/csharp/PhoneNumbers.PerformanceTest/Benchmarks/PhoneNumberWorkflowBenchmark.cs
index 226e24712..521f14ae6 100644
--- a/csharp/PhoneNumbers.PerformanceTest/Benchmarks/PhoneNumberWorkflowBenchmark.cs
+++ b/csharp/PhoneNumbers.PerformanceTest/Benchmarks/PhoneNumberWorkflowBenchmark.cs
@@ -3,12 +3,19 @@
 
 namespace PhoneNumbers.PerformanceTest.Benchmarks
 {
+    /// <summary>
+    /// The end-to-end workflow plus each of its three phases measured separately, so a cost can be
+    /// attributed rather than only totalled. Parse and Format allocate for quite different reasons -
+    /// StringBuilder round trips on one side, regex replacement producing a fresh string per step on
+    /// the other - and those want different fixes.
+    /// </summary>
     [MemoryDiagnoser]
     [SimpleJob(RuntimeMoniker.Net10_0)]
     public class PhoneNumberWorkflowBenchmark
     {
         private PhoneNumberUtil _phoneNumberUtil = null!;
         private PhoneNumberBenchmarkCase[] _phoneNumbers = null!;
+        private PhoneNumber[] _parsedNumbers = null!;
 
         [Params(1000)]
         public int PhoneNumberCount { get; set; }
@@ -18,8 +25,20 @@ public void Setup()
         {
             _phoneNumberUtil = PhoneNumberUtil.GetInstance();
             _phoneNumbers = PhoneNumberBenchmarkData.Create(_phoneNumberUtil, PhoneNumberCount);
+
+            // Pre-parsed, so the validate and format phases measure only their own work.
+            _parsedNumbers = new PhoneNumber[_phoneNumbers.Length];
+            for (var i = 0; i < _phoneNumbers.Length; i++)
+            {
+                _parsedNumbers[i] =
+                    _phoneNumberUtil.Parse(_phoneNumbers[i].NumberToParse, _phoneNumbers[i].DefaultRegion);
+            }
         }
 
+        /// <summary>
+        /// Kept alongside the phases: parsing warms metadata the later phases reuse, so the whole is
+        /// not simply the sum, and this is the figure tracked across previous changes.
+        /// </summary>
         [Benchmark]
         public int ParseValidateAndFormatPhoneNumbers()
         {
@@ -38,5 +57,48 @@ public int ParseValidateAndFormatPhoneNumbers()
 
             return checksum;
         }
+
+        [Benchmark]
+        public int ParseOnly()
+        {
+            var checksum = 0;
+            for (var i = 0; i < _phoneNumbers.Length; i++)
+            {
+                var phoneNumber = _phoneNumbers[i];
+                checksum += _phoneNumberUtil
+                    .Parse(phoneNumber.NumberToParse, phoneNumber.DefaultRegion).CountryCode;
+            }
+
+            return checksum;
+        }
+
+        [Benchmark]
+        public int ValidateOnly()
+        {
+            var checksum = 0;
+            for (var i = 0; i < _parsedNumbers.Length; i++)
+            {
+                if (_phoneNumberUtil.IsValidNumber(_parsedNumbers[i]))
+                    checksum++;
+            }
+
+            return checksum;
+        }
+
+        /// <summary>
+        /// INTERNATIONAL rather than E164: E164 takes an early exit that skips pattern formatting,
+        /// so it would not exercise the regex replacement chain in FormatNsnUsingPattern.
+        /// </summary>
+        [Benchmark]
+        public int FormatOnly()
+        {
+            var checksum = 0;
+            for (var i = 0; i < _parsedNumbers.Length; i++)
+            {
+                checksum += _phoneNumberUtil.Format(_parsedNumbers[i], PhoneNumberFormat.INTERNATIONAL).Length;
+            }
+
+            return checksum;
+        }
     }
 }
diff --git a/csharp/PhoneNumbers.PerformanceTest/README.md b/csharp/PhoneNumbers.PerformanceTest/README.md
index d9c98b531..b98b3e86b 100644
--- a/csharp/PhoneNumbers.PerformanceTest/README.md
+++ b/csharp/PhoneNumbers.PerformanceTest/README.md
@@ -22,7 +22,10 @@ dotnet run -c Release --framework net10.0 -- --filter "*PhoneNumberWorkflowBench
 ```
 
 The `PhoneNumberWorkflowBenchmark` exercises the widest slice of the library; the full suite
-completes in a few minutes on a single runtime.
+completes in a few minutes on a single runtime. Alongside the end-to-end
+`ParseValidateAndFormatPhoneNumbers` it measures `ParseOnly`, `ValidateOnly` and `FormatOnly`
+against the same data, so a cost — allocation especially — can be attributed to a phase instead of
+only being visible as a total.
 
 Other available benchmarks: