Preserve encoding during DocumentState updates (#77354)

Fixes dotnet/sdk#46780

This issue was introduced by #74623, which changed
`Document.WithSyntaxRoot` to always use the encoding specified by the
new syntax root. APIs like SyntaxRewriter do not preserve the input
encoding during node rewrites, so `Formatter.Format` would overwrite the
document syntax root with a node that did not specify any encoding. The
fix _allows_ `WithSyntaxRoot` to specify a new encoding, but for cases
where the encoding of the new syntax root is not known, we fall back to
the original behavior of preserving the encoding from the previous root
and/or document text.

src/Workspaces/Core/Portable/Workspace/Solution/DocumentState.cs

@@ -497,6 +497,20 @@ internal DocumentState UpdateTree(SyntaxNode newRoot, PreservationMode mode)
 
         // use the encoding that we get from the new root
         var encoding = newRoot.SyntaxTree.Encoding;
+        if (encoding is null)
+        {
+            // The new tree doesn't specify an encoding. For these cases, continue to use the previous encoding of the
+            // document.
+            if (TryGetSyntaxTree(out var priorTree))
+            {
+                // this is most likely available since UpdateTree is normally called after modifying the existing tree.
+                encoding = priorTree.Encoding;
+            }
+            else if (TryGetText(out var priorText))
+            {
+                encoding = priorText.Encoding;
+            }
+        }
 
         var syntaxTreeFactory = LanguageServices.GetRequiredService<ISyntaxTreeFactoryService>();
 

src/Workspaces/MSBuild/Test/VisualStudioMSBuildWorkspaceTests.cs

@@ -2898,13 +2898,14 @@ class C { }";
             Assert.Equal(encoding.EncodingName, text.Encoding.EncodingName);
             Assert.Equal(fileContent, text.ToString());
 
-            // update root blindly again, after observing encoding, see that encoding is overridden to null
+            // update root blindly again, after observing encoding, see that encoding is preserved
+            // 🐉 Tools rely on encoding preservation; see https://github.com/dotnet/sdk/issues/46780
             var doc3 = document.WithSyntaxRoot(gen.CompilationUnit()); // empty CU
             var doc3text = await doc3.GetTextAsync();
-            Assert.Null(doc3text.Encoding);
+            Assert.Same(text.Encoding, doc3text.Encoding);
             var doc3tree = await doc3.GetSyntaxTreeAsync();
-            Assert.Null(doc3tree.Encoding);
-            Assert.Null(doc3tree.GetText().Encoding);
+            Assert.Same(text.Encoding, doc3tree.Encoding);
+            Assert.Same(text.Encoding, doc3tree.GetText().Encoding);
 
             // change doc to have no encoding, still succeeds at writing to disk with old encoding
             var root = await document.GetSyntaxRootAsync();