Update Roslyn to 4.4.0 1.22369.1

[JoeRobich]

The changes are mostly around how Global Options in Roslyn is changing. We need to provide fallbacks which derive from OmniSharpOptions for LineFormatting options that are normally found in the .editorconfig.

[333fred]

One small suggestion, otherwise LGTM

[filipw]

Not sure what is going on here, or whether there is some regression in Roslyn, but we have a bunch of System.OutOfMemoryException

[JoeRobich]

Thanks for helping fix up the scripting test @filipw! Going to keep working on it.

[333fred]

@JoeRobich selectively running tests, it appears that import completion is taking significant amounts of memory. By far the worst offender is ImportCompletionResolvesOnSubsequentQueries, though all show more memory usage than I would expect. It's particularly bad when they're the first tests that run, and looking at heap dumps there are huge numbers of PENamedTypeSymbols in memory. Additionally, when I insert Task.Delays between each theory leg, it becomes very obvious that something is holding onto memory, between theory runs, as memory increases by about the same amount each leg without going back down.

[333fred]

Looking at VMMAP of the process after it's run all 4 of the ImportCompletionResolvesOnSubsequentQueries tests, I see the following:

Most of the memory being used in these tests is not in the native heap. I'm not sure what to make of this, as I have no idea what could be causing such massive native allocations in our framework.

[333fred]

@filipw why did you revert this? Just to see if it would fix the OOM issue?

[filipw]

it failed many tests with System.InvalidOperationException : MoveToImmutable can only be performed when Count equals Capacity.

[filipw]

Additionally, when I insert Task.Delays between each theory leg, it becomes very obvious that something is holding onto memory, between theory runs, as memory increases by about the same amount each leg without going back down.

One thing is we inject SharedOmniSharpHostFixture which is a shared test host - it does not get recreated between the tests, only is attempted to be cleared and reused. Could be that.

[333fred]

One thing is we inject SharedOmniSharpHostFixture which is a shared test host - it does not get recreated between the tests, only is attempted to be cleared and reused. Could be that.

The import completion tests are not using the shared fixture (my latest commit adds a new fixture to share hosts between them).

[JoeRobich]

So, I made the assumption that the stack trace that is failing is also the stack that is allocating. I choose to look at the TestHelper GetReferences and by caching the MetaData References memory stay much, much lower.

                             [xUnit.net 00:03:06.06]       System.OutOfMemoryException : Insufficient memory to continue the execution of the program.
2022-07-21T06:44:52.1290096Z [xUnit.net 00:03:06.07]       Stack Trace:
2022-07-21T06:44:52.1291826Z [xUnit.net 00:03:06.07]            at System.Runtime.InteropServices.Marshal.AllocHGlobal(IntPtr cb)
2022-07-21T06:44:52.1292828Z [xUnit.net 00:03:06.07]            at System.Reflection.Internal.StreamMemoryBlockProvider.ReadMemoryBlockNoLock(Stream stream, Int64 start, Int32 size)
2022-07-21T06:44:52.1293843Z [xUnit.net 00:03:06.07]            at System.Reflection.PortableExecutable.PEReader..ctor(Stream peStream, PEStreamOptions options, Int32 size)
2022-07-21T06:44:52.1294777Z [xUnit.net 00:03:06.07]            at Microsoft.CodeAnalysis.ModuleMetadata.CreateFromStream(Stream peStream, PEStreamOptions options)
2022-07-21T06:44:52.1313146Z [xUnit.net 00:03:06.07]         D:\a\omnisharp-roslyn\omnisharp-roslyn\tests\TestUtility\TestHelpers.cs(121,0): at TestUtility.TestHelpers.<>c.<GetReferences>b__3_2(String l) <-- This might be problematic
2022-07-21T06:44:52.1313810Z [xUnit.net 00:03:06.07]            at System.Linq.Enumerable.SelectEnumerableIterator`2.MoveNext()
2022-07-21T06:44:52.1314299Z [xUnit.net 00:03:06.07]            at System.Linq.Enumerable.UnionIterator`1.MoveNext()
2022-07-21T06:44:52.1314808Z [xUnit.net 00:03:06.07]            at System.Collections.Generic.LargeArrayBuilder`1.AddRange(IEnumerable`1 items)
2022-07-21T06:44:52.1319398Z [xUnit.net 00:03:06.07]            at System.Collections.Generic.EnumerableHelpers.ToArray[T](IEnumerable`1 source)
2022-07-21T06:44:52.1320135Z [xUnit.net 00:03:06.07]            at System.Collections.Immutable.ImmutableArray.CreateRange[T](IEnumerable`1 items)
2022-07-21T06:44:52.1320745Z [xUnit.net 00:03:06.07]            at Roslyn.Utilities.EnumerableExtensions.ToBoxedImmutableArray[T](IEnumerable`1 items)
2022-07-21T06:44:52.1322078Z [xUnit.net 00:03:06.07]            at Microsoft.CodeAnalysis.ProjectInfo.Create(ProjectId id, VersionStamp version, String name, String assemblyName, String language, String filePath, String outputFilePath, CompilationOptions compilationOptions, ParseOptions parseOptions, IEnumerable`1 documents, IEnumerable`1 projectReferences, IEnumerable`1 metadataReferences, IEnumerable`1 analyzerReferences, IEnumerable`1 additionalDocuments, Boolean isSubmission, Type hostObjectType, String outputRefFilePath)

[JoeRobich]

Now to work out why the global line formatting options aren't applying to Cake services..

2022-07-22T01:09:43.1369369Z [xUnit.net 00:00:23.05]     OmniSharp.Cake.Tests.CodeActionsV2Facts.Can_extract_method [FAIL]
2022-07-22T01:09:43.1802021Z [xUnit.net 00:00:23.05]       Assert.Equal() Failure
2022-07-22T01:09:43.1805254Z [xUnit.net 00:00:23.05]       Expected: LinePositionSpanTextChange { StartLine=4, StartColumn=8, EndLine=4, EndColumn=8, NewText=NewMethod();\n    }\n\n    private static void NewMethod()\n    {\n         }
2022-07-22T01:09:43.2251706Z [xUnit.net 00:00:23.05]       Actual:   LinePositionSpanTextChange { StartLine=4, StartColumn=8, EndLine=6, EndColumn=0, NewText=NewMethod();\n    }\r\n\r\n    private static void NewMethod()\r\n    {\r\n        Console.Write("should be using System;");\r\n    }\r\n }

[filipw]

😅👏