[simonrozsival]

Context: #11212
Context: #13505
Context: https://github.com/dotnet/msbuild/blob/main/src/Tasks/Unzip.cs

The _ExtractAar target used the custom ExtractJarsFromAar task to extract *.jar and annotations.zip entries from @(LibraryProjectZip) and @(_AndroidReferenceLibraryProjectZip) into $(IntermediateOutputPath)library_project_jars, library_project_annotations and their reference_* counterparts. That is one more custom task to host, test and eventually make multithreaded, while MSBuild's <Unzip/> already supports Include patterns.

~~ Why not just <Unzip SkipUnchangedFiles="true"/>? ~~

The incrementality of the extracted files matters. _ExportJarToXml uses @(InputJar) and @(ReferenceJar) (the extracted jars) as Inputs. ExtractJarsFromAar rewrote a file only when its content changed, and it deleted files that no longer came from any archive.

<Unzip/> sets every extracted file's mtime to the zip entry's timestamp. Its SkipUnchangedFiles check compares only entry timestamp and size. Gradle/AGP-produced AARs typically use fixed timestamps (1980-02-01), so:

• an updated classes.jar would be written with a 1980 mtime, and _ExportJarToXml would treat itself as up to date, leaving bindings stale;
• an updated classes.jar of the same size would not be extracted at all.

~~ Approach ~~

_ExtractAar now:

1. Extracts into $(IntermediateOutputPath)extract_aar_staging\, using three batched <Unzip/> calls per library:
   • non-.aar inputs: Include="*.jar"
   • .aar inputs: Include="classes.jar;libs/*.jar", which mirrors Files.ShouldSkipEntryInAar()
   • all inputs: Include="annotations.zip;*/annotations.zip"
2. <Touch/>es the staged files. CopyIfChanged would otherwise skip a file whose destination is newer and has the same length.
3. Copies them into the final directories with the existing CopyIfChanged task. It compares content, so unchanged files keep their timestamps.
4. <Delete/>s stale files: everything under the four output directories minus the expected set, computed with an Exclude transform over %(RecursiveDir).
5. Removes the staging directory.

This keeps the previous incremental semantics, and the target's Inputs/Outputs are unchanged. javac and D8 are unaffected: neither reads these directories (their jars come from ResolveLibraryProjectImports). The only consumers are class-parse and the generator's AnnotationsZipFiles. The added cost is one temporary disk write of the matched entries, and only when _ExtractAar runs.

<Unzip/> pattern notes: * also matches /, and patterns are separator-normalized only when Include contains ;. That is why every pattern containing a path separator is part of a ;-list.

~~ Behavior differences ~~

• Path-traversal entries are rejected by <Unzip/>'s own zip-slip check instead of Files.GetArchiveExtractionPath(). Because <Unzip/> filters before it validates, a traversal .jar entry in an .aar that doesn't match classes.jar/libs/* is now ignored instead of failing the build. Nothing is written outside the destination in either case.
• classes.jar matching is now case-insensitive.

ExtractJarsFromAar, its <UsingTask/> and ExtractJarsFromAarTests are removed. Files.ShouldSkipEntryInAar() and Files.GetArchiveExtractionPath() remain, because ResolveLibraryProjectImports uses them.

~~ Testing ~~

Not built or tested locally, to save disk space in this batch; validation is left to CI. Relevant coverage is in BindingBuildTest: the annotations.zip support test and the lint.jar exclusion check, plus the binding incremental build tests.

[Copilot]

🟡 Changes recommended

The final copy step omits descendant-link validation, allowing archive files to be written through symlinks or junctions outside the intended extraction directories.

5 open issues

• Validate final destinations against symlink and junction traversal · New
• Normalize ZIP entry separators before extracting on Unix · New
• Disable SkipUnchangedFiles to preserve later-archive-wins behavior · New
• Preserve platform-aware path comparison when finding stale files · New
• Add regression coverage for staging, copying, and stale-file cleanup · New

What changed in this PR

Replaces the custom AAR extraction task with MSBuild’s built-in Unzip, reducing custom task maintenance in the Android binding build pipeline.

Changes:

• Adds staged extraction for bound and reference JARs and annotation archives.
• Uses timestamp refresh, content-aware copying, and stale-file deletion to preserve incremental behavior.
• Removes ExtractJarsFromAar, its task registration, and its dedicated tests.

File	Description
src/​Xamarin.Android.Build.Tasks/​Tests/​Xamarin.Android.Build.Tests/​Tasks/​ExtractJarsFromAarTests.cs	Removes the custom extractor’s tests.
src/​Xamarin.Android.Build.Tasks/​Tasks/​ExtractJarsFromAar.cs	Removes the custom extraction implementation.
src/​Xamarin.Android.Build.Tasks/​Microsoft.Android.Sdk/​targets/​Microsoft.Android.Sdk.AndroidLibraries.targets	Replaces the task invocation with staged extraction, copying, and cleanup.

_{Review effort: Balanced}