Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Enable support for an LSP client to open source generated files #75180

Merged
merged 12 commits into from
Oct 2, 2024

Conversation

dibarbet
Copy link
Member

@dibarbet dibarbet commented Sep 20, 2024

A new version of #68771

I was having a heck of a time with the conflicts, so I ended up just re-doing it (and copying changes over that I thought were relevant).

I believe I addressed the comments, however I still do have some questions.

Server side of dotnet/vscode-csharp#6426

Note - this does not yet implement support for refreshing the file on changes

@dibarbet dibarbet requested a review from a team as a code owner September 20, 2024 00:50
@dotnet-issue-labeler dotnet-issue-labeler bot added Area-IDE untriaged Issues and PRs which have not yet been triaged by a lead labels Sep 20, 2024
// Source generated documents cannot go through OnDocumentOpened/Closed.
// There is a separate OnSourceGeneratedDocumentOpened/Closed method, but there is no need
// for us to call it in LSP - it deals with mapping TextBuffers to text containers.
return;
Copy link
Member Author

@dibarbet dibarbet Sep 20, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@jasonmalinowski @CyrusNajmabadi for thoughts. I decided to not call it - otherwise I'd have to implement similar TryOnSourceGeneratedDocumentOpen/Closed like we have for regular documents. It didn't seem necessary as the open was only trying to associate things with text buffers.

Since these are readonly (and content managed by the server), there is nothing to update in the workspace either

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i think the comment coudlo be beefed up.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So I've forgotten, for regular documents, why are we opening/closing the documents? Is it just so other code can ask "what's open" from the workspace or are we still getting some other benefit?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I honestly can't remembner anything that would require the regular documents to actually be opened in the workspace. Generally LSP features should be checking if LSP thinks the document is opened.

And there isn't any source text container stuff to do either, sicne we have our own LSP fork management.

@@ -43,24 +43,31 @@ internal class LanguageInfoProvider : ILanguageInfoProvider
{ ".mts", s_typeScriptLanguageInformation },
};

public LanguageInformation GetLanguageInformation(string documentPath, string? lspLanguageId)
public LanguageInformation GetLanguageInformation(Uri uri, string? lspLanguageId)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is lspLanguageId the language selected in the VS Code status bar for a particular document? If so, why would we not want to honor that over the file extension, in the case that the user has chosen something different?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is the languageId the LSP client gives us (which I believe in VSCode will change based on the language selected in the status bar).

I suppose generally we should use the languageId instead (if we're able to map it), otherwise fall back to the file extension.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm going to defer this for now though - there is some small risk of regression, especially in VS if they give us the wrong languageId. I'd rather take that in a separate PR

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Agreed that in the case of a well-behaved LSP client, we should trust the language ID. Also agreed we should keep this as is for now and verify we have well-behaved clients before changing it. 😄

var assemblyVersion = Uri.EscapeDataString(identity.Generator.AssemblyVersion.ToString());
var typeName = Uri.EscapeDataString(identity.Generator.TypeName);

var uri = $"{Scheme}://{projectId}/{hintPathPortion}?{DocumentIdParam}={documentId}&{HintNameParam}={hintName}&{AssemblyNameParam}={assemblyName}&{AssemblyVersionParam}={assemblyVersion}&{TypeNameParam}={typeName}";
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

so, the part i trust is the projectId/docId. teh rest i think should all be opaque data we effectively don't munge with (specifically referring to hintpathportion).

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The project ID/document ID is still the core part, yes. But including everything else means we're able to round-trip the entire SourceGeneratedDocumentIdentity. This allows for at least one edge case to be better handled: if the user opens a source generated file, and then something means that file isn't being generated any more, we can still respond to an LSP request for this URI if the file is still open on the client, as if it's otherwise a regular SourceGeneratedDocument rather than it being a generic miscellaneous files file.

Copy link
Member

@jasonmalinowski jasonmalinowski left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Overall this still jives with what I expect this to look like. I feel like my old PR had some issues around opening/closing documents repeatedly and miscellaneous files breaking, but I trust you've tested that well.

My two open questions:

  1. Regarding the DateTime.Now stuff -- what's that being used for?
  2. For the code we have that's doing document open/closed on mutating workspaces -- is that just so code can ask what documents are open or did we have something more happening there?

var assemblyVersion = Uri.EscapeDataString(identity.Generator.AssemblyVersion.ToString());
var typeName = Uri.EscapeDataString(identity.Generator.TypeName);

var uri = $"{Scheme}://{projectId}/{hintPathPortion}?{DocumentIdParam}={documentId}&{HintNameParam}={hintName}&{AssemblyNameParam}={assemblyName}&{AssemblyVersionParam}={assemblyVersion}&{TypeNameParam}={typeName}";
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The project ID/document ID is still the core part, yes. But including everything else means we're able to round-trip the entire SourceGeneratedDocumentIdentity. This allows for at least one edge case to be better handled: if the user opens a source generated file, and then something means that file isn't being generated any more, we can still respond to an LSP request for this URI if the file is still open on the client, as if it's otherwise a regular SourceGeneratedDocument rather than it being a generic miscellaneous files file.

public static async ValueTask<Document?> GetDocumentAsync(this Solution solution, TextDocumentIdentifier documentIdentifier, CancellationToken cancellationToken)
{
var textDocument = await solution.GetTextDocumentAsync(documentIdentifier, cancellationToken).ConfigureAwait(false);
Contract.ThrowIfTrue(textDocument is not null && textDocument is not Document, $"{textDocument!.Id} is not a Document");
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The semantics here feel a bit funky -- so if the document doesn't exist we'll return null, but if it does exist but it's an additional file we'll throw? Seems like we might end up with a bunch of accidental exceptions that way?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In my view, the accidental exceptions would be more catching unintentional bugs - someone was asking for a document - but they got an additional document (or other kind of document) instead - which likely means they should be calling a different API.

If we returned null, then the code might 'work' but not be doing what they want.

// Source generated documents cannot go through OnDocumentOpened/Closed.
// There is a separate OnSourceGeneratedDocumentOpened/Closed method, but there is no need
// for us to call it in LSP - it deals with mapping TextBuffers to text containers.
return;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So I've forgotten, for regular documents, why are we opening/closing the documents? Is it just so other code can ask "what's open" from the workspace or are we still getting some other benefit?

return false;
}

var newState = existingState.WithText(text);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I forget, how expensive is this? If the state is different, can we actually use that later rather than doing it a second time?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It doesn't seem to expensive (compares checksums, then copies over to a new Sg state. As far as I can tell any actual work is deferred - https://source.dot.net/#Microsoft.CodeAnalysis.Workspaces/Workspace/Solution/SourceGeneratedDocumentState.cs,c9182b28095a2b1c,references

if you'd prefer I can expose a version of WithFrozenSourceGeneratedDocuments that takes in the sg states.

Copy link
Member Author

@dibarbet dibarbet Sep 30, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looked into this a little bit. Definitely possible to do with some small refactoring to WithFrozenSourceGeneratedDocuments to allow us to pass an existing state. However I don't think it will make much of a difference for the following reasons

  1. If the state is the same, the only work we do is calculating the checksums, which we would do regardless
  2. If the state is different, there still isn't much work done - we create a SourceGeneratedDocumentState, but all the parsing and everything else is deferred in lazy's.
  3. We create a max of one new state here if the text is different since this returns early as soon as we detect one difference.

Think it is fine to leave it like this, but let me know what you think

_trackedDocuments.Keys.Where(static uri => uri.Scheme == SourceGeneratedDocumentUri.Scheme)
.Select(uri => (identity: SourceGeneratedDocumentUri.DeserializeIdentity(workspaceCurrentSolution, uri), _trackedDocuments[uri].Text))
.Where(tuple => tuple.identity.HasValue)
.SelectAsArray(tuple => (tuple.identity!.Value, DateTime.Now, tuple.Text));
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@CyrusNajmabadi What is the time used for here, generally? Something complicated or in the VS for Windows case just updating the gold bar so the user knows when it was last updated?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Think its just the gold bar. Right now I just picked a time because we need to pass it to WithFrozenSourceGeneratedDocuments

@@ -64,9 +64,9 @@ public Task<IReadOnlyList<SearchResultPreviewPanelBase>> GetPreviewPanelsAsync(S
return null;

Uri? absoluteUri;
if (document.IsSourceGeneratedDocument)
if (document.SourceGeneratedDocumentIdentity is not null)
Copy link
Member

@jasonmalinowski jasonmalinowski Sep 28, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we had to pass through the identity just so we can create a URI, why not just have the document object just hold onto the URI in the first place? Then we can delete this handling here and the try/catch in favor of a document.GetURI() when the object is created?

Copy link
Member Author

@dibarbet dibarbet Sep 30, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

the code creating the NavigableDocument is in features, which doesn't have any of the URI stuff from the LSP layer (see next comment)

{
absoluteUri = ProtocolConversions.CreateUriFromSourceGeneratedFilePath(filePath);
absoluteUri = SourceGeneratedDocumentUri.Create(document.SourceGeneratedDocumentIdentity.Value);
}
else
{
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

And to be clear if we just pass a URI directly then we can delete this else block too. This was added as a part of #72442 which was a fine tactical fix but shouldn't be needed going forward, I'd imagine.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Going to leave this as-is - the navigable document is created at the features layer which doesn't include any of the URI stuff we have at the protocol layer. So I believe the else is still needed here since we won't be using the GetUri extension on document.

I don't see a problem with moving some of the URI stuff down to features, but I think that is best left to a separate PR (a number of helpers would need to move along with it)

@dibarbet
Copy link
Member Author

dibarbet commented Oct 2, 2024

Going to merge this - addressed Cyrus' comments, and Jason won't be available for a bit. Can take additional feedback later.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Area-IDE untriaged Issues and PRs which have not yet been triaged by a lead VSCode
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants