em 01: scaffold ExtractMethod tool + Demo fixture

- tools/ExtractMethod: net10.0 console app (Microsoft.CodeAnalysis.CSharp
  5.9.0, pinned latest stable). CLI: <file.cs> <startLine> <endLine>.
  Parses the file, builds a scratch compilation with refs from
  TRUSTED_PLATFORM_ASSEMBLIES, and reports how many whole statements the
  line range covers (clean error otherwise; exit codes: 0 ok, 1
  resolution error, 2 usage).
- Tooling/CompilationLoader: shared parse + compilation path for CLI and
  tests (tests exercise the exact loading path the CLI uses).
- Tooling/SelectionResolver: snaps a 1-based inclusive line range to
  whole statements in the enclosing method body block; boundary checks
  never split a statement; nested/blank-line ranges handled cleanly.
- tests/: ExtractMethod/Fixtures/Demo.cs checked-in fixture exercising
  every bucket of the parent spec (read-only local, written+read-later
  return, scratch local, param read, field+property access, indexer +
  method invocation, multi-statement range incl. a for-loop); excluded
  from project compilation, copied to output as data.
- Tests: Demo.cs compiles with no diagnostics; range 68..72 resolves to
  2 statements (LocalDeclarationStatement, ForStatement); AnalyzeDataFlow
  succeeds on the fixture's for-loop node; mid-statement range fails
  cleanly. 30/30 green.
This commit is contained in:
2026-09-12 15:28:14 +01:00
parent 406e8a97f6
commit b012d0aaa0
9 changed files with 526 additions and 1 deletions
+3 -1
View File
@@ -3,4 +3,6 @@ artifacts/
bin/
obj/
.dotnet-cli/
*.user
*.user
# NuGet restore scratch dir (DOTNET_CLI_HOME fallout)
.local/
+17
View File
@@ -13,6 +13,10 @@ Project("{2150E333-8FDC-42A3-9474-1A3956D46DE8}") = "tests", "tests", "{0AB3BF05
EndProject
Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "BeforeAfter.Tests", "tests\BeforeAfter.Tests\BeforeAfter.Tests.csproj", "{F5E0C182-CBC3-4440-AF8F-32887F9B589C}"
EndProject
Project("{2150E333-8FDC-42A3-9474-1A3956D46DE8}") = "tools", "tools", "{07C2787E-EAC7-C090-1BA3-A61EC2A24D84}"
EndProject
Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "ExtractMethod", "tools\ExtractMethod\ExtractMethod.csproj", "{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}"
EndProject
Global
GlobalSection(SolutionConfigurationPlatforms) = preSolution
Debug|Any CPU = Debug|Any CPU
@@ -59,6 +63,18 @@ Global
{F5E0C182-CBC3-4440-AF8F-32887F9B589C}.Release|x64.Build.0 = Release|Any CPU
{F5E0C182-CBC3-4440-AF8F-32887F9B589C}.Release|x86.ActiveCfg = Release|Any CPU
{F5E0C182-CBC3-4440-AF8F-32887F9B589C}.Release|x86.Build.0 = Release|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Debug|Any CPU.Build.0 = Debug|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Debug|x64.ActiveCfg = Debug|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Debug|x64.Build.0 = Debug|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Debug|x86.ActiveCfg = Debug|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Debug|x86.Build.0 = Debug|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Release|Any CPU.ActiveCfg = Release|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Release|Any CPU.Build.0 = Release|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Release|x64.ActiveCfg = Release|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Release|x64.Build.0 = Release|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Release|x86.ActiveCfg = Release|Any CPU
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C}.Release|x86.Build.0 = Release|Any CPU
EndGlobalSection
GlobalSection(SolutionProperties) = preSolution
HideSolutionNode = FALSE
@@ -67,5 +83,6 @@ Global
{2FF256B4-B085-449E-95AD-B9E6202DA94A} = {827E0CD3-B72D-47B6-A68D-7590B98EB39B}
{E65C0410-A863-44B0-88A4-6C74EF929419} = {827E0CD3-B72D-47B6-A68D-7590B98EB39B}
{F5E0C182-CBC3-4440-AF8F-32887F9B589C} = {0AB3BF05-4346-4AA6-1389-037BE0695223}
{3E8944E9-BB07-424B-B3A6-59FB072BAB3C} = {07C2787E-EAC7-C090-1BA3-A61EC2A24D84}
EndGlobalSection
EndGlobal
@@ -21,6 +21,19 @@
<ItemGroup>
<ProjectReference Include="..\..\src\Before\Before.csproj" />
<ProjectReference Include="..\..\src\After\After.csproj" />
<ProjectReference Include="..\..\tools\ExtractMethod\ExtractMethod.csproj" />
</ItemGroup>
<ItemGroup>
<!-- The ExtractMethod fixtures are DATA for the Roslyn tool, not source:
they are excluded from this project's compilation (the scratch
compilation gives the tool the exact checked-in text) and copied to
the output dir so tests can find them by path. -->
<Compile Remove="ExtractMethod\Fixtures\**\*.cs" />
<!-- Rebase the copy target to Fixtures/ in the output dir: the tool's
apphost (the Executable 'ExtractMethod') is copied here too, and a
folder named ExtractMethod/ would collide with that file. -->
<Content Include="ExtractMethod\Fixtures\**\*.cs" Link="Fixtures\%(RecursiveDir)%(Filename)%(Extension)" CopyToOutputDirectory="PreserveNewest" />
</ItemGroup>
</Project>
@@ -0,0 +1,102 @@
using ExtractMethod.Tooling;
using Microsoft.CodeAnalysis;
using Microsoft.CodeAnalysis.CSharp;
using Microsoft.CodeAnalysis.CSharp.Syntax;
namespace BeforeAfter.Tests.ExtractMethod;
/// <summary>
/// Tests for yak em 01: the scaffold (console tool) and the checked-in demo
/// fixture. The fixture is DATA (excluded from compilation, copied to the
/// output dir), and it is parsed with the tool's own <see cref="CompilationLoader"/>
/// so these tests exercise the exact loading path the CLI uses.
/// </summary>
public class DemoFixtureTests
{
private static readonly string FixturePath =
Path.Combine(AppContext.BaseDirectory, "Fixtures", "Demo.cs");
private static (SyntaxTree Tree, CSharpCompilation Compilation) LoadFixture()
{
Assert.True(File.Exists(FixturePath), $"fixture missing at {FixturePath}");
var tree = CompilationLoader.ParseFile(FixturePath);
var compilation = CompilationLoader.CreateCompilation(tree, "DemoFixture");
return (tree, compilation);
}
// ---------------------------------------------------------------------
// (a) The fixture is pristine under the scratch compilation: if this
// fails, the fixture (or the TRUSTED_PLATFORM_ASSEMBLIES reference
// loading) is broken and every later yak would analyze garbage.
// ---------------------------------------------------------------------
[Fact]
public void Demo_compiles_with_no_diagnostics()
{
var (_, compilation) = LoadFixture();
Assert.Empty(compilation.GetDiagnostics());
}
// ---------------------------------------------------------------------
// (b) A known line range snaps to the expected whole statements.
// Lines 68..72 of Demo.cs are `int total = 0;` followed by the whole
// for-loop (for keyword .. closing brace) inside Summarize — exactly
// the "multi-statement range incl. a for-loop" shape of the spec.
// ---------------------------------------------------------------------
[Fact]
public void Known_range_resolves_to_expected_statements()
{
var (tree, _) = LoadFixture();
var report = SelectionResolver.Resolve(tree, 68, 72);
Assert.True(report.Succeeded, report.Error);
Assert.Equal(2, report.Count);
Assert.Equal(
new[] { SyntaxKind.LocalDeclarationStatement, SyntaxKind.ForStatement },
report.Kinds);
Assert.Equal("Summarize", report.Method?.Identifier.ValueText);
}
// ---------------------------------------------------------------------
// (c) The semantic model's data-flow analysis works on the fixture's
// for-loop node — the foundation every classification in yak 02
// builds on. (AnalyzeDataFlow on the loop itself walks the bound loop
// body; Succeeded == false would mean the scratch compilation or the
// selected node cannot be bound.)
// ---------------------------------------------------------------------
[Fact]
public void Fixture_for_loop_supports_data_flow_analysis()
{
var (tree, compilation) = LoadFixture();
var forStatement = tree.GetRoot()
.DescendantNodes()
.OfType<ForStatementSyntax>()
.Single(n => EnclosingMethodName(n) == "Summarize");
var dataFlow = compilation.GetSemanticModel(tree).AnalyzeDataFlow(forStatement);
Assert.True(dataFlow.Succeeded);
}
// ---------------------------------------------------------------------
// Acceptance also demands "clean error otherwise": selecting only the
// for-loop header line (line 69) splits the statement and must fail with
// a readable reason instead of a crash or a silent wrong count.
// ---------------------------------------------------------------------
[Fact]
public void Range_ending_mid_statement_fails_cleanly()
{
var (tree, _) = LoadFixture();
var report = SelectionResolver.Resolve(tree, 69, 69);
Assert.False(report.Succeeded);
Assert.NotNull(report.Error);
Assert.Contains("no whole", report.Error);
}
private static string? EnclosingMethodName(SyntaxNode node) =>
node.AncestorsAndSelf().OfType<MethodDeclarationSyntax>().FirstOrDefault()?.Identifier.ValueText;
}
@@ -0,0 +1,95 @@
// Demo fixture for the extract-method micro-tool (tools/ExtractMethod).
//
// This file is DATA, not project source: the test project excludes it from
// compilation (see BeforeAfter.Tests.csproj) and runs Roslyn over its raw
// text, so the tool sees exactly what is checked in here. It uses explicit
// usings because the scratch compilation has no implicit usings.
//
// Bucket map (the classification yak 02 works from):
// bucket where example
// local, read-only (param) ScoreReads seed, score
// written + read-later (return) Summarize, Heaviest total, message, best
// scratch local (local) Summarize for-loop i
// enclosing-method param (param) ScoreReads, Summarize bonus, limit
// field access (extract-first) ScoreReads _seed (optional)
// property access (extract-first) Summarize Scale (optional)
// indexer access (extract-first) Heaviest widgets[i]
// method invocation (extract-first, flagged) Heaviest best.Bigger(...)
// multi-statement range incl. for Summarize total + for-loop
//
// Tests pin EXACT line numbers of the statements they select. Keep the
// formatting stable; when a line must change, update DemoFixtureTests too.
using System.Collections.Generic;
namespace BeforeAfter.Tests.ExtractMethod.Fixtures;
/// <summary>
/// Small first-class citizen of the extract-method tool: every method below
/// shows one or more of the classification "buckets" the tool must report.
/// </summary>
public class Demo
{
// Own-class state: reading it from a selection is "extract-first optional"
// (a same-class extracted method can still see these fields). Initialized
// here so the fixture compiles with zero diagnostics (CS0649 otherwise).
private int _seed = 5;
// Own-class property: like a field, accessible from a same-class method,
// so hoisting it is optional in the report.
public int Scale { get; set; } = 1;
/// <summary>Small local type so the scratch compilation binds.</summary>
public sealed record Widget(int Weight, int Count, string Label)
{
/// <summary>Instance method used to exercise the invocation bucket.</summary>
public Widget Bigger(Widget other) =>
Weight >= other.Weight ? this : other;
}
/// <summary>
/// Bucket: enclosing-method parameter read (bonus =&gt; param), local
/// read-only (seed, score =&gt; param), own-class field access
/// (_seed =&gt; extract-first, optional). Ends with a return expression.
/// </summary>
public int ScoreReads(int bonus)
{
int seed = _seed;
int score = seed * 2;
return score + bonus;
}
/// <summary>
/// Bucket: written + read-later locals (total, message =&gt; return),
/// scratch local (i =&gt; local), own-class property access (Scale =&gt;
/// extract-first optional), multi-statement range incl. a for-loop.
/// </summary>
public string Summarize(int limit)
{
int total = 0;
for (int i = 0; i < limit; i++)
{
total += i;
}
int scaled = total * Scale;
string message = "sum=" + scaled;
return message;
}
/// <summary>
/// Bucket: indexer access (widgets[i] on List&lt;T&gt; =&gt; extract-first),
/// method invocation (Bigger =&gt; extract-first, flagged because hoisting
/// changes how often it is evaluated), local written + read-later
/// (best =&gt; return).
/// </summary>
public Widget Heaviest(List<Widget> widgets, int count)
{
Widget best = widgets[0];
for (int i = 1; i < count; i++)
{
best = best.Bigger(widgets[i]);
}
return best;
}
}
+20
View File
@@ -0,0 +1,20 @@
<Project Sdk="Microsoft.NET.Sdk">
<PropertyGroup>
<OutputType>Exe</OutputType>
<TargetFramework>net10.0</TargetFramework>
<ImplicitUsings>enable</ImplicitUsings>
<Nullable>enable</Nullable>
<RootNamespace>ExtractMethod</RootNamespace>
</PropertyGroup>
<ItemGroup>
<!--
Roslyn IS the tool: we parse C# source, build a scratch compilation and
ask the semantic model for data flow. Pinned to the latest stable on
nuget.org (5.9.0, Roslyn's versioning now tracks the VS release train).
-->
<PackageReference Include="Microsoft.CodeAnalysis.CSharp" Version="5.9.0" />
</ItemGroup>
</Project>
+51
View File
@@ -0,0 +1,51 @@
using ExtractMethod.Tooling;
// CLIs are boring on purpose: argument parsing and printing live here, all
// Roslyn logic lives in Tooling/ so the tests can drive it directly.
string usage = "usage: ExtractMethod <file.cs> <startLine> <endLine> (1-based, inclusive)";
if (args.Length != 3)
{
Console.Error.WriteLine(usage);
return 2;
}
string file = Path.GetFullPath(args[0]);
if (!File.Exists(file))
{
Console.Error.WriteLine($"error: file not found: {file}");
Console.Error.WriteLine(usage);
return 2;
}
if (!int.TryParse(args[1], out int startLine) || !int.TryParse(args[2], out int endLine))
{
Console.Error.WriteLine($"error: line numbers must be integers");
Console.Error.WriteLine(usage);
return 2;
}
// 1. parse the file
var tree = CompilationLoader.ParseFile(file);
// 2. build the scratch compilation (refs from TRUSTED_PLATFORM_ASSEMBLIES)
var compilation = CompilationLoader.CreateCompilation(tree, Path.GetFileNameWithoutExtension(file));
if (compilation.GetDiagnostics().Any(d => d.Severity == Microsoft.CodeAnalysis.DiagnosticSeverity.Error))
{
Console.Error.WriteLine("warning: the file does not compile cleanly under a plain Roslyn compilation; reporting syntax-level resolution only");
}
// 3. snap the range to whole statements, report cleanly otherwise
var report = SelectionResolver.Resolve(tree, startLine, endLine);
if (!report.Succeeded)
{
Console.Error.WriteLine($"error: {report.Error}");
return SelectionResolver.ExitError;
}
Console.WriteLine(
$"{report.Count} statement(s) selected, lines {report.StartLine}..{report.EndLine} " +
$"in {report.Method?.Identifier.ValueText}(): " +
string.Join(", ", report.Kinds));
return 0;
@@ -0,0 +1,57 @@
using Microsoft.CodeAnalysis;
using Microsoft.CodeAnalysis.CSharp;
namespace ExtractMethod.Tooling;
/// <summary>
/// Turns a plain .cs file into a Roslyn <see cref="CSharpCompilation"/> that a
/// semantic model can be queried against.
///
/// Shared by the CLI (Program.cs) and the tests so both exercise the same
/// loading path — the tests should not re-implement this, or they would be
/// testing their own assumptions instead of the tool.
/// </summary>
public static class CompilationLoader
{
/// <summary>
/// Parse a file into a <see cref="SyntaxTree"/> using the latest language
/// version. Nothing here is project-specific: no .sln, no .csproj, no
/// generated files — the tool is a micro-tool that runs on a single file.
/// </summary>
public static SyntaxTree ParseFile(string path)
{
var text = File.ReadAllText(Path.GetFullPath(path));
return CSharpSyntaxTree.ParseText(
text,
path: path,
options: CSharpParseOptions.Default.WithLanguageVersion(LanguageVersion.Latest));
}
/// <summary>
/// Build a scratch compilation over <paramref name="tree"/>.
///
/// WHY <c>TRUSTED_PLATFORM_ASSEMBLIES</c>: it is the runtime-resolved list
/// of the actual BCL assemblies this process (net10.0) runs on, e.g.
/// System.Private.CoreLib, System.Runtime, System.Console. Referencing
/// them means an arbitrary fixture can bind List&lt;T&gt;, string, and the
/// rest of the BCL without any NuGet/EF/SQLite haul. (A single
/// typeof(object) reference is NOT enough on modern .NET — that only pulls
/// in the core library, and Console/GC/enums etc. fail to bind.)
/// </summary>
public static CSharpCompilation CreateCompilation(SyntaxTree tree, string assemblyName)
{
var refs = ((string)AppContext.GetData("TRUSTED_PLATFORM_ASSEMBLIES")!)
.Split(Path.PathSeparator)
.Select(p => (MetadataReference)MetadataReference.CreateFromFile(p));
return CSharpCompilation.Create(
assemblyName,
new[] { tree },
refs,
new CSharpCompilationOptions(
OutputKind.DynamicallyLinkedLibrary,
// Match the repo's net10.0 projects so the fixture's nullable
// annotations compile without warnings.
nullableContextOptions: NullableContextOptions.Enable));
}
}
@@ -0,0 +1,168 @@
using Microsoft.CodeAnalysis;
using Microsoft.CodeAnalysis.CSharp;
using Microsoft.CodeAnalysis.CSharp.Syntax;
using Microsoft.CodeAnalysis.Text;
namespace ExtractMethod.Tooling;
/// <summary>A line range resolved to the whole statements it covers.</summary>
public sealed record SelectionReport
{
public required bool Succeeded { get; init; }
/// <summary>Human-readable reason when <see cref="Succeeded"/> is false.</summary>
public string? Error { get; init; }
/// <summary>
/// The selected whole statements, in source order, all siblings in one
/// statement list (v1: the enclosing method's body block).
/// </summary>
public required IReadOnlyList<StatementSyntax> Statements { get; init; }
/// <summary>The method that owns the selected statements.</summary>
public required MethodDeclarationSyntax? Method { get; init; }
public int StartLine { get; init; }
public int EndLine { get; init; }
public int Count => Statements.Count;
/// <summary>Statement kinds (e.g. LocalDeclarationStatement, ForStatement).</summary>
public IEnumerable<SyntaxKind> Kinds => Statements.Select(s => s.Kind());
}
/// <summary>
/// Snap a 1-based, inclusive line range (IDE-selection style) to a contiguous
/// set of WHOLE statements — never split a statement (parent spec decision #1).
///
/// v1 rule: the selection is matched against the top-level statement list of
/// the enclosing method's body block. A range nested inside a block whose
/// siblings live deeper (e.g. the body of a for-loop) is a clean error: such
/// selections would not survive extraction anyway, so the tool tells the user
/// to widen the selection to the whole enclosing statement (the loop).
/// </summary>
public static class SelectionResolver
{
/// <summary>Line range of a missing selection (error path).</summary>
private static readonly SelectionReport Failure = new()
{
Succeeded = false,
Statements = Array.Empty<StatementSyntax>(),
Method = null,
};
/// <summary>Common ExitCode for a clean resolution error (see also 2 = usage).</summary>
public const int ExitError = 1;
public static SelectionReport Resolve(SyntaxTree tree, int startLine, int endLine)
{
var text = tree.GetText();
// ---- validate and normalize the requested line range ----
if (startLine < 1 || startLine > text.Lines.Count)
{
return Fail($"startLine {startLine} is out of range (file has {text.Lines.Count} lines)");
}
if (endLine < startLine || endLine > text.Lines.Count)
{
return Fail($"endLine {endLine} is out of range (file has {text.Lines.Count} lines)");
}
// Snap the endpoints past whitespace-only lines: an IDE "select lines"
// range often trails a blank line; blank lines are not statements, so
// skipping them is snapping, not splitting.
while (endLine > startLine && IsBlank(text.Lines[endLine - 1])) endLine--;
while (startLine < endLine && IsBlank(text.Lines[startLine - 1])) startLine++;
// Positions of the first / last non-whitespace character of the
// bounding lines. Trailing content (e.g. a `// comment` after the last
// statement) intentionally KEEPS the end position past the statement,
// which the boundary check below reports as "range ends mid-statement".
int selStart = FirstNonWhitespacePosition(text.Lines[startLine - 1]);
int selEnd = LastNonWhitespacePosition(text.Lines[endLine - 1]);
// ---- find the enclosing method ----
// Anchor on the first token of the selection: at statement start the
// token belongs to the statement we want (FindNode at exact boundaries
// is ambiguous, FindToken is not).
var firstToken = tree.GetRoot().FindToken(selStart);
var enclosingStatement = firstToken.Parent?.AncestorsAndSelf()
.OfType<StatementSyntax>()
.FirstOrDefault();
var method = enclosingStatement?.AncestorsAndSelf()
.OfType<MethodDeclarationSyntax>()
.FirstOrDefault();
if (method is null)
{
return Fail($"{startLine} does not sit inside a method (local functions and expression bodies are not supported in v1)");
}
// The whole selection must live inside ONE method body.
if (method.Body is null || selEnd > method.Body.FullSpan.End)
{
return Fail($"range {startLine}..{endLine} crosses a method boundary (selection must stay inside one method body)");
}
// ---- v1: only the method body's own statement list is selectable ----
if (enclosingStatement is not null && enclosingStatement.Parent != method.Body)
{
return Fail(
$"range {startLine}..{endLine} is nested inside \"{enclosingStatement.Kind()}\" — " +
"select the whole enclosing statement (e.g. the entire for-loop) instead");
}
// ---- collect the whole statements covered by the range ----
var body = method.Body;
var selected = body.Statements
.Where(s => selStart <= s.SpanStart && s.Span.End <= selEnd)
.ToList();
if (selected.Count == 0)
{
return Fail($"range {startLine}..{endLine} covers no whole top-level statement (mid-statement or empty range)");
}
// Boundary checks: the selection must START at a statement's first
// token and END right after a statement's last token. Siblings never
// overlap, so equal boundaries also imply the selected run is
// contiguous (any statement between first and last lies inside the
// range and is therefore in `selected`).
if (selected[0].SpanStart != selStart)
{
return Fail($"range {startLine}..{endLine} starts inside a statement — snap to whole statements");
}
if (selected[^1].Span.End != selEnd)
{
return Fail($"range {startLine}..{endLine} ends inside a statement (or picks up trailing text on the last line) — snap to whole statements");
}
return new SelectionReport
{
Succeeded = true,
Statements = selected,
Method = method,
StartLine = startLine,
EndLine = endLine,
};
}
private static SelectionReport Fail(string reason) => Failure with { Error = reason };
private static bool IsBlank(TextLine line) => string.IsNullOrWhiteSpace(line.ToString());
private static int FirstNonWhitespacePosition(TextLine line)
{
var s = line.ToString();
return line.Start + (s.Length - s.TrimStart().Length);
}
private static int LastNonWhitespacePosition(TextLine line)
{
var s = line.ToString();
return line.End - (s.Length - s.TrimEnd().Length);
}
}