Add regression coverage for metadata newline preservation

[VolPlita]

Fixes #4083

Context

This change strengthens regression coverage around newline handling in project file parsing.

Historically, switching from XmlTextReader to XmlReader.Create in read-only load paths caused newline/whitespace normalization changes that broke multiline metadata and command scenarios. This PR adds coverage and clarifies why the current reader choice is intentional for compatibility.

Changes Made

• Added regression tests in src/Build.UnitTests/Construction/XmlReaderWithoutLocation_Tests.cs to verify multiline metadata with CRLF is preserved:
  • when loading from disk
  • when loading through read-only project loading (loadProjectsReadOnly: true)
• Refactored duplicated test XML setup into a shared helper method.
• Updated the compatibility comment in src/Build/Xml/XmlReaderExtension.cs to clearly document why XmlTextReader is used and linked related issues (#4210, #4213, #4083, #6232, #6669).

Testing

• Ran XmlReaderWithoutLocation_Tests and validated the new regression tests pass.
• Ran existing multiline Exec regression tests to ensure related behavior remains covered.

Notes

This PR does not change runtime parsing behavior; it improves documentation and regression protection for existing compatibility behavior.

[Copilot]

Pull request overview

This PR adds regression tests to ensure MSBuild’s XML loading path continues to preserve multi-line item metadata values (notably CRLF sequences), and updates an in-code compatibility comment explaining why MSBuild continues using XmlTextReader instead of XmlReader.Create to avoid whitespace/newline normalization behavior changes.

Changes:

• Expanded the compatibility comment in XmlReaderExtension to document newline/whitespace preservation rationale and link related history.
• Added Windows-only regression tests that load a project from disk (including “read-only” load) and assert CRLF preservation in item metadata.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/Build/Xml/XmlReaderExtension.cs	Updates compatibility commentary around XML reader choice to explain newline/whitespace behavior impact.
src/Build.UnitTests/Construction/XmlReaderWithoutLocation_Tests.cs	Adds regression tests validating CRLF preservation for item metadata when loading projects from disk and in read-only mode.

[VolPlita]

@dotnet-policy-service agree company="Microsoft"

[github-actions]

Caution

agentic threat detected
Threat detection flagged this output in warn mode. Manual review is REQUIRED before any follow-up automation.

Details

The threat detection results could not be parsed.

Review the workflow run logs for details.

Code Review — PR #14261

✅ 21/24 dimensions clean.

Dimension	Severity	Finding
Cross-Platform Correctness	MODERATE	[WindowsOnlyFact] is unneeded — XmlTextReader (Normalization=false) preserves \r\n on Linux too; use [Fact]
Design / Test Placement	MODERATE	These tests don't use the XmlReaderNoIXmlLineInfo wrapper that defines the class's scope; better placed in ItemEvaluation_Tests.cs or a new XmlReaderExtension_Tests.cs
Code Simplification	NIT	Split([Environment.NewLine], ...).Length.ShouldBe(5) is redundant — an exact ShouldBe(...) already guarantees the count

---

Design concern: test placement

XmlReaderWithoutLocation_Tests was built around one scenario: loading a project from an XmlReader that doesn't implement IXmlLineInfo. Neither new test uses the XmlReaderNoIXmlLineInfo wrapper, and loadProjectsReadOnly: true is orthogonal to "without location". Better homes would be src/Build.UnitTests/Evaluation/ItemEvaluation_Tests.cs (where metadata value semantics tests already live) or a new XmlReaderExtension_Tests.cs that explicitly targets the XmlReaderExtension behavior documented in the updated comment.

---

The comment update in XmlReaderExtension.cs is a clear improvement — it replaces the vague "Ignore loadAsReadOnly for now" placeholder with an accurate rationale and five cross-referenced issues. The tests themselves are correct and will catch real regressions; the findings above are structural improvements only.

Generated by Expert Code Review (on open) for #14261 · 1.8K AIC · ⊞ 30.3K

[OvesN]

Good expansion of our test coverage. I only have a few minor comments.