Refs: #137104

• anyOS @TylerLeonhardt
• anyOS @sbatten (web)
• anyOS @lramos15 (desktop)

Complexity: 3

Create Issue

Improved language detection for small inputs is (finally) out! Context:

• If the existing model can detect the language, it is used
• If not, we try a new model which detects with biasing from recently opened files, files in the workspace, and historically opened files.

Please test:

• files are able to be detected with less text than before, and ambiguities can be influenced by workspace contents and recently opened. For example:

# hello world
- this 
- is

can be MD or YAML, so if you have recently opened MD and/or have md files in your workspace, md should be picked, otherwise YAML. Similarly:

class Foo {
   private foo: Foo
   constructor() {
      this.foo = ''
   }
}

can be CS or TS, and again your recent files should influence this.

• working with very large untitled files doesn't have poor performance
• can go back to old way of detection by disabling workbench.editor.historyBasedLanguageDetection