[plugin] fetching tasks is slow for big multi-root workspaces

[akosyakov]

validateSchema should be computed once and cached

but I am not sure that we need it at all, does VS Code validate task configs?

We removed such logic already for preferences, since it is slow and VS Code does not care about.

cc @elaihau @RomanNikitenko

[akosyakov]

I was able to reproduce it with a bit of customizations JLab project:

• check out https://github.com/jupyterlab/jupyterlab
• create .vscode/tasks.json with the following content in the project root

{
  // See https://go.microsoft.com/fwlink/?LinkId=733558
  // for the documentation about the tasks.json format
  "version": "2.0.0",
  "tasks": [
    {
      "type": "npm",
      "script": "build",
      "problemMatcher": [
        "$tsc"
      ]
    }
  ]
}

• create a multi-root workspace from this JSON (replace file:///workspace with your absolute path) :

{
   "folders": [
      {
         "path": "file:///workspace/jupyterlab"
      },
      {
         "path": "file:///workspace/jupyterlab/dev_mode"
      },
      {
         "path": "file:///workspace/jupyterlab/examples"
      },
      {
         "path": "file:///workspace/jupyterlab/packages"
      },
      {
         "path": "file:///workspace/jupyterlab/packages/services/examples/node"
      },
      {
         "path": "file:///workspace/jupyterlab/packages/services/examples/browser"
      },
      {
         "path": "file:///workspace/jupyterlab/packages/services/examples/typescript-browser-with-output"
      },
      {
         "path": "file:///workspace/jupyterlab/buildutils"
      },
      {
         "path": "file:///workspace/jupyterlab/buildutils/template"
      },
      {
         "path": "file:///workspace/jupyterlab/buildutils/buildutils/test-template"
      },
      {
         "path": "file:///workspace/jupyterlab/tests"
      },
      {
         "path": "file:///workspace/jupyterlab/testutils"
      },
      {
         "path": "file:///workspace/jupyterlab/jupyterlab/tests/mock_packages/extension"
      }
   ],
   "settings": {}
}

• start Theia and pass the path to the multi-root workspace
• open README.md as an editor and for preview
• reload the page
• try to navigate in the explorer, till npm scripts appear
• notice how sluggish it is and how long does it take for npm scripts to appear

The profiling snapshot: https://drive.google.com/file/d/1pyCUMaGm-7scjJ-KMqHmfm2gINs6BdVx/view?usp=sharing

There are 2 hot spots:

• isTaskConfigValid (it only happens if you have tasks.json)

- comparing tasks in `$fetchTasks` from the plugin system:

[akosyakov]

isTaskConfigValid is handled in #7676

[akosyakov]

@elaihau Do you know how to test whether we need to call compareTasks in $fetchTasks? I wonder do we need it at all or we simply should return all tasks? How did you validate VS Code extensions' expectations?

[akosyakov]

VS Code does not remove duplicate, it returns all tasks:

I does not look like $fetchTasks is aligned with VS Code extensions' expectations. That's what Theia gives:

Tested with https://github.com/akosyakov/vscode-fetch-tasks-test and adding following tasks.json under .vscode in this repo:

{
  // See https://go.microsoft.com/fwlink/?LinkId=733558
  // for the documentation about the tasks.json format
  "version": "2.0.0",
  "tasks": [
    {
      "label": "build",
      "type": "shell",
      "command": "msbuild",
      "args": [
        // Ask msbuild to generate full paths for file names.
        "/property:GenerateFullPaths=true",
        "/t:build",
        // Do not generate summary otherwise it leads to duplicate errors in Problems panel
        "/consoleloggerparameters:NoSummary"
      ],
      "group": "build",
      "presentation": {
        // Reveal the output only if unrecognized errors occur.
        "reveal": "silent"
      },
      // Use the standard MS compiler pattern to detect errors, warnings and infos
      "problemMatcher": "$msCompile"
    },
    {
      "label": "build",
      "type": "shell",
      "command": "msbuild",
      "args": [
        // Ask msbuild to generate full paths for file names.
        "/property:GenerateFullPaths=true",
        "/t:build",
        // Do not generate summary otherwise it leads to duplicate errors in Problems panel
        "/consoleloggerparameters:NoSummary"
      ],
      "group": "build",
      "presentation": {
        // Reveal the output only if unrecognized errors occur.
        "reveal": "silent"
      },
      // Use the standard MS compiler pattern to detect errors, warnings and infos
      "problemMatcher": "$msCompile"
    }
  ]
}

[elaihau]

@elaihau Do you know how to test whether we need to call compareTasks in $fetchTasks? I wonder do we need it at all or we simply should return all tasks? How did you validate VS Code extensions' expectations?

If we don't want to

• take care of identical tasks, or
• filter out the invalid task configs,
  we probably can make $fetchTasks() a lot faster

I took a look at the $fetchTasks() function in tasks-main.ts and noticed that it does not take the cutomized detected tasks into consideration.
Say, you have npm: build defined in package.json, and customize it in tasks.json this way:

{
  "type": "npm",
  "script": "build",
  "label": "Build the project"
}

The current $fetchTasks() simply returns the npm: build - the label you added to the task (i.e., "Build the project") is not taken care of. Is it something we need to fix? @akosyakov

[akosyakov]

If we don't want to take care of identical tasks, or filter out the invalid task configs, we probably can make $fetchTasks() a lot faster

I've simplified it in aba5392 Could you have a look there? But it only to make it faster it does not make sure that it aligned with VS Code extensions' expectations.

I took a look at the $fetchTasks() function in tasks-main.ts and noticed that it does not take the cutomized detected tasks into consideration.
Say, you have npm: build defined in package.json, and customize it in tasks.json this way:

I've opened #7681 for VS Code compatibility. Ideally results should be aligned as much as possible. But you can also try to investigate whether it's worth an effort, i.e. whether it can affect some extensions, users and how many. Let's have a look at it separate from performance optimisations.

[elaihau]

I've simplified it in aba5392 Could you have a look there? But it only to make it faster it does not make sure that it aligned with VS Code extensions' expectations.

@akosyakov
Sorry for the late response.
I believe we missed the part that avoids duplications.

I tried it in vscode with an extension.
If a detected task is configured in tasks.json, fetchTasks() in vscode returns one task.
In the current Theia, it returns both.

However, in vscode, same tasks, or tasks with same labels, can be contributed by different providers multiple times.

Do you want me to create an issue for it?

[akosyakov]

@elaihau It should be covered by doing #7681. That we have broken tests if we miss something again.