ILProcessor should also update custom debug info #867
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
* ILProcessor should also update custom debug info When eidting IL with ILProcessor various pieces of debug information have references to the origin IL instructions. These references can be either resolved (point to Instruction instance), in which case the editting mostly works, or unresolved (store IL offset only) in which case they need to be resolved before the editting can occur (after the edit the original IL offsets are invalid and unresolvable). This is effectively a continuation of jbevain#687 which implemented this for local scopes. This change extends this to async method stepping info and state machine scopes. The change refactors the code to make it easier to reuse the same logic between the various debug infos being processed. Updated the existing tests from jbevain#687 to include async and state machine debug info (completely made up) and validate that it gets updated correctly. * PR Feedback Renamed some parameters/locals to better match the existing code style. * PR Feedback
|
Is there any chance this would fix Fody/Fody#1110 (VS not being able to debug an assembly written by Cecil)? |
|
I don't think that the same problem. This problem (wrong custom debug info) has been in Cecil since forever it seems. |
Native PDB is not supported on Linux and the test infra falls back to portable PDB automatically. Since the two PDB implementations read the custom debug info from a different place the test constructing the input needs to adapt to this difference as well.
jonpryor
pushed a commit
to dotnet/android
that referenced
this pull request
Jul 3, 2024
Context: #9043 Changes: jbevain/cecil@0.11.4...0.11.5 * jbevain/cecil@8c123e1: Bump to 0.11.5 * jbevain/cecil@870ce3e: Fix RVA field alignment (jbevain/cecil#888) * jbevain/cecil@4ad9c0f: Fix that method resolution would consider all function-pointers to be the same (jbevain/cecil#885) * jbevain/cecil@cc48622: Fix a StackOverflowException reading windows runtime assemblies. (jbevain/cecil#879) * jbevain/cecil@341fb14: Treat instance and static methods as different methods during resolution (jbevain/cecil#882) * jbevain/cecil@e052ab5: Address issue #873 (jbevain/cecil#874) * jbevain/cecil@92f32da: Add `MethodImplAttributes.AggressiveOptimization` (jbevain/cecil#855) * jbevain/cecil@c4cfe16: Fix a race condition between certain Has properties and their collection property. (jbevain/cecil#843) * jbevain/cecil@65a2912: Add more style configuration (jbevain/cecil#854) * jbevain/cecil@9eb00e4: ILProcessor should also update custom debug info (jbevain/cecil#867) * jbevain/cecil@7d36386: Fix corrupted debug header directory entry when writing multiple such entries. (jbevain/cecil#869) * jbevain/cecil@6f94613: InvariantCulture for operand to string conversion in Instruction.ToString() (jbevain/cecil#870) * jbevain/cecil@42b9ef1: Add support for generic attributes (jbevain/cecil#871) * jbevain/cecil@49b1c52: Add `Unmanaged` calling convention (jbevain/cecil#852) * jbevain/cecil@2c68927: Fix mixed module ReadSymbols() (jbevain/cecil#851) * jbevain/cecil@f7b64f7: Fix custom attribute with enum on generic type (jbevain/cecil#827) * jbevain/cecil@79b43e8: Fix deterministic MVID and add PdbChecksum (jbevain/cecil#810) * jbevain/cecil@8b593d5: Harden debug scope update logic (jbevain/cecil#824) * jbevain/cecil@a56b5bd: FieldRVA alignment (jbevain/cecil#817) * jbevain/cecil@75372c7: Switch to netcoreapp3.1 for tests (jbevain/cecil#823) * jbevain/cecil@5f69faa: Add support for generating the method and generic method comment signature with nested types (jbevain/cecil#801) * jbevain/cecil@a0a6ce4: Addressing issue #781 (jbevain/cecil#782) * jbevain/cecil@2f1077d: Update the version of Microsoft.NETFramework.ReferenceAssemblies.net40 (jbevain/cecil#787) * jbevain/cecil@ede17f9: Fix handling of empty string constants (jbevain/cecil#776) #9043 stops building API-34 and makes API-35 stable, and in attempting to do so encounters this error when running all of the in-tree Windows smoke tests on CI: D:\a\_work\1\s\bin\Release\dotnet\packs\Microsoft.Android.Sdk.Windows\35.0.0-ci.pr.gh9043.6\tools\Xamarin.Android.Bindings.JavaDependencyVerification.targets(22,5): error MSB4062: The "Xamarin.Android.Tasks.GetMicrosoftNuGetPackagesMap" task could not be loaded from the assembly D:\a\_work\1\s\bin\Release\dotnet\packs\Microsoft.Android.Sdk.Windows\35.0.0-ci.pr.gh9043.6\tools\Xamarin.Android.Build.Tasks.dll. Could not load file or assembly 'Mono.Cecil, Version=0.11.4.0, Culture=neutral, PublicKeyToken=50cebf1cceb9d05e'. The system cannot find the file specified. Confirm that the <UsingTask> declaration is correct, that the assembly and all its dependencies are available, and that the task contains a public class that implements Microsoft.Build.Framework.ITask. [D:\a\_work\1\a\TestRelease\07-02_16.40.15\temp\DotNetBuildandroid-armFalseFalseFalse\UnnamedProject.csproj] This appears to be caused by the fact that previously we: 1. Build `Microsoft.Android.Sdk.ILLink.csproj`, which pulls in [Microsoft.NET.ILLink/9.0.0-preview.6.24319.11][0], which pulls in [Microsoft.DotNet.Cecil/0.11.4-alpha.24313.1][1], which contains `Mono.Cecil.dll` versioned as 0.11.5.0. (Yes, it's "odd" that `Microsoft.DotNet.Cecil/0.11.4*` would contain a Cecil versioned as 0.11.5, but that's what it has!) 2. Build the rest of dotnet/android with the [Mono.Cecil/0.11.4][2] package. 3. The Mono.Cecil/0.11.4 package "wins" and ends up in the output directory and the sdk pack. So long as the `ILLink*`-related assemblies don't use any Mono.Cecil/0.11.5 APIs, this works, however it is risky because we don't know exactly what API `ILLink*` uses. #9043 appears to change the build order such that Mono.Cecil/0.11.5 now "wins" and is in the output directory. This causes the above MSB4062 assembly load error. We could try to fix the ordering and make Mono.Cecil/0.11.4 "win", but this leaves us vulnerable to missing some API that `ILLink*` needs. As such, it's better that we update the rest of dotnet/android to use the `0.11.5` version of `Mono.Cecil`. This update breaks the `LinkerTests.FixAbstractMethodsStep_Explicit()` unit test. Specifically, this logic now returns `null` instead of being able to be resolved: new MethodReference (iface_method.Name, void_type, iface) .Resolve (); It feels like this makes sense: a method name and return type doesn't seem like it would be enough to resolve, as parameters are not considered. The fix is simply to use the existing `MethodDefinition` as the `MethodReference`, there is no reason to create a new one. This change fixes the test. [0]: https://dev.azure.com/dnceng/public/_artifacts/feed/dotnet9-transport/NuGet/Microsoft.NET.ILLink/overview/9.0.0-preview.7.24328.10 [1]: https://dev.azure.com/dnceng/public/_artifacts/feed/dotnet9-transport/NuGet/Microsoft.DotNet.Cecil/overview/0.11.4-alpha.24313.1 [2]: https://www.nuget.org/packages/Mono.Cecil/0.11.4
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When editing IL with ILProcessor various pieces of debug information have references to the origin IL instructions. These references can be either resolved (point to Instruction instance), in which case the editing mostly works, or unresolved (store IL offset only) in which case they need to be resolved before the editing can occur (after the edit the original IL offsets are invalid and unresolvable).
This is effectively a continuation of #687 which implemented this for local scopes. This change extends this to async method stepping info and state machine scopes.
The change refactors the code to make it easier to reuse the same logic between the various debug infos being processed.
Updated the existing tests from #687 to include async and state machine debug info (completely made up) and validate that it gets updated correctly.
This is an upstream port of dotnet#34