Don't list all of the keywords of Python and suggest them when I type a left square bracket

[yvvt0379]

Environment data

• Language Server version: 2021.8.2
• OS and version: Windows 10 20H2
• Python version (& distribution if applicable, e.g. Anaconda): Python 3.9.6

Expected behaviour

Don't list all of the keywords of Python and suggest them when I type a left square bracket

Actual behaviour

Lists all of the keywords of Python and suggests them when I type a left square bracket.

Logs

[FG] completion at c:\Users\Admin\ywt\test.py:0:1 ...
[FG]   parsing: c:\Users\Admin\ywt\test.py (6ms)
[FG]   parsing: c:\Users\Admin\.vscode\extensions\ms-python.vscode-pylance-2021.8.2\dist\typeshed-fallback\stdlib\builtins.pyi [fs read 0ms] (56ms)
[FG]   binding: c:\Users\Admin\.vscode\extensions\ms-python.vscode-pylance-2021.8.2\dist\typeshed-fallback\stdlib\builtins.pyi (13ms)
[FG]   binding: c:\Users\Admin\ywt\test.py (0ms)
[FG] completion at c:\Users\Admin\ywt\test.py:0:1 [found 191 items] (78ms)
Background analysis message: setFileOpened
Background analysis message: markFilesDirty
[FG] parsing: c:\Users\Admin\.vscode\extensions\ms-python.vscode-pylance-2021.8.2\dist\typeshed-fallback\stdlib\typing.pyi [fs read 0ms] (15ms)
[FG] binding: c:\Users\Admin\.vscode\extensions\ms-python.vscode-pylance-2021.8.2\dist\typeshed-fallback\stdlib\typing.pyi (5ms)
[FG] parsing: c:\Users\Admin\.vscode\extensions\ms-python.vscode-pylance-2021.8.2\dist\typeshed-fallback\stdlib\_typeshed\__init__.pyi [fs read 1ms] (10ms)
[FG] binding: c:\Users\Admin\.vscode\extensions\ms-python.vscode-pylance-2021.8.2\dist\typeshed-fallback\stdlib\_typeshed\__init__.pyi (1ms)
[FG] parsing: c:\Users\Admin\.vscode\extensions\ms-python.vscode-pylance-2021.8.2\dist\typeshed-fallback\stdlib\typing_extensions.pyi [fs read 0ms] (2ms)
[FG] binding: c:\Users\Admin\.vscode\extensions\ms-python.vscode-pylance-2021.8.2\dist\typeshed-fallback\stdlib\typing_extensions.pyi (0ms)
Background analysis message: getDiagnosticsForRange
Background analysis message: getDiagnosticsForRange
Background analysis message: getSemanticTokens delta
[BG(1)] getSemanticTokens delta previousResultId:1629710631950 at c:\Users\Admin\ywt\test.py ...
[BG(1)]   parsing: c:\Users\Admin\ywt\test.py (0ms)
[BG(1)]   binding: c:\Users\Admin\ywt\test.py (1ms)
[BG(1)] getSemanticTokens delta previousResultId:1629710631950 at c:\Users\Admin\ywt\test.py (1ms)
Background analysis message: analyze
[BG(1)] analyzing: c:\Users\Admin\ywt\test.py ...
[BG(1)]   checking: c:\Users\Admin\ywt\test.py (5ms)
[BG(1)] analyzing: c:\Users\Admin\ywt\test.py (5ms)
Background analysis message: resumeAnalysis
Background analysis message: markFilesDirty
Background analysis message: analyze

Code Snippet / Additional information

XXX

[yvvt0379]

In addition, previously suggested items are also included in the suggestion list.

[jakebailey]

Pylance doesn't have contextual completions for keywords; we offer them up most everywhere, but with a very low priority compared to other items in the completion list.

Are you seeing otherwise? What do you mean by "previously suggested items are also included in the suggestion list"?

[jakebailey]

(See also #1557, which should be retitled to be more clear.)

[yvvt0379]

Are you seeing otherwise?

No, I haven't seen it elsewhere, I just see it when I type [(but not after a variable defined explicitly a dict).

What do you mean by "previously suggested items are also included in the suggestion list"?

Here's a screenshot. min, max and print are suggestions I previously accepted.

[erictraut]

This is what I'd expect to see in this case — at least given the current implementation. The names min, max and print are symbols that are used in the current scope, so they appear on the top of the list. As @jakebailey mentioned, keywords are currently suggested even in contexts where they are not appropriate, but they're always sorted low on the list.

Perhaps as a simple short-term improvement we could omit all keyword suggestions if you haven't started to type an identifier. Once one an alphabetical character has been typed, we can suggest only the keywords that match that filter. For example, if you were to place the insertion point between the square brackets, no keywords would be suggested until you type a character, say an n, and then the keyword not would be included in the suggestion list. @jakebailey, if you like that idea, it's a trivial change — just change this line to return false.

In the future, we could add further smarts to offer only the keywords that are syntactically appropriate at the current location, but that's a much larger change.

[jakebailey]

Tracing this out, I was surprised to learn that we actually place keywords above "normal" completions when making the sort text: https://github.com/microsoft/pyright/blob/fa4194fd8b0e6d9fb970468e50a974e0d4eeee5c/packages/pyright-internal/src/languageService/completionProvider.ts#L177-L181

Perhaps that could change? We already appear to place "likely keywords" at the top of the list, so this might work better than not offering up keywords until more than one character was typed.

[jakebailey]

I've also retitled #1557 to better reflect the idea of contextually offering up completions (but the above suggestion could happen sooner than a bigger change to how we do keywords).

[erictraut]

When I wrote the original completion code in pyright, I took a stab at the best sort order, but it wasn't informed by any user feedback. Based on the feedback in this thread and others, I think it's reasonable to demote keywords below "normal" and perhaps even "private" symbols — at least until we add code to provide contextual keyword filtering.

[jakebailey]

This may even be something we could run an experiment on (to see what people actually like).

[jakebailey]

Related is #1136, where @DanielRosenwasser suggested giving keywords and variables the same precedence. (I forgot that issue was there; found it while pruning old issues.)

[judej]

Moving this issue to discussion as an enhancement request for comments and upvotes.