Autocompletion partially works with numpy

[ubaldot]

I made some research and I found this that led me here.

However, I am not sure if the issue is on jedi or if it is on the lsp client that I am using.

The autocompletion works it other python modules, but with numpy.

To reproduce:

import numpy as np

a = np.array([1,2,3]) # when I type 'np.' then I get a list of all possible completions when I hit '.'
a. # Here, I don't get anything when I hit '.'

An excerpt of the lsp debug messages follows:

...
11/23/25 17:52:17: Sent {"method":"textDocument/completion","params":{"context":{"triggerCharacter":"","triggerKind":1},"textDocument":{"uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"position":{"character":1,"line":4}}}
11/23/25 17:52:17: Received {"id":15,"jsonrpc":"2.0","result":{"isIncomplete":false,"items":[{"label":"a","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aa","kind":6,"insertText":"a"},{"label":"abs(x)","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aabs","kind":3,"insertText":"abs"},{"label":"all(iterable)","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aall","kind":3,"insertText":"all"},{"label":"any(iterable)","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aany","kind":3,"insertText":"any"},{"label":"ascii(obj)","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aascii","kind":3,"insertText":"ascii"},{"label":"assert","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aassert","kind":14,"insertText":"assert"},{"label":"async","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aasync","kind":14,"insertText":"async"},{"label":"await","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aawait","kind":14,"insertText":"await"},{"label":"ArithmeticError","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aArithmeticError","kind":7,"insertText":"ArithmeticError"},{"label":"AssertionError","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aAssertionError","kind":7,"insertText":"AssertionError"},{"label":"AttributeError","data":{"doc_uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"sortText":"aAttributeError","kind":7,"insertText":"AttributeError"}]}}
11/23/25 17:52:17: Received {"method":"textDocument/publishDiagnostics","jsonrpc":"2.0","params":{"uri":"file:///Users/ubaldot/Documents/test_lsp.py","diagnostics":[{"source":"pyflakes","message":"'manim as mn' imported but unused","severity":2,"range":{"end":{"character":19,"line":1},"start":{"character":0,"line":1}}}],"version":78}}
11/23/25 17:52:18: Sent {"method":"textDocument/completion","params":{"context":{"triggerCharacter":".","triggerKind":2},"textDocument":{"uri":"file:///Users/ubaldot/Documents/test_lsp.py"},"position":{"character":2,"line":4}}}
11/23/25 17:52:18: Received {"id":16,"jsonrpc":"2.0","result":{"isIncomplete":false,"items":[]}}
11/23/25 17:52:18: Received {"method":"textDocument/publishDiagnostics","jsonrpc":"2.0","params":{"uri":"file:///Users/ubaldot/Documents/test_lsp.py","diagnostics":[{"source":"pyflakes","message":"invalid syntax","severity":1,"range":{"end":{"character":6,"line":4},"start":{"character":3,"line":4}}}],"version":79}}
11/23/25 17:52:45: Received {"method":"textDocument/publishDiagnostics","jsonrpc":"2.0","params":{"uri":"file:///Users/ubaldot/Documents/test_lsp.py","diagnostics":[{"source":"pyflakes","message":"invalid syntax","severity":1,"range":{"end":{"character":6,"line":3},"start":{"character":3,"line":3}}}],"version":80}}
...

As you see, there is a suspicious line that suggests that the lsp server does not return anything.

11/23/25 17:52:18: Received {"id":16,"jsonrpc":"2.0","result":{"isIncomplete":false,"items":[]}}

I am using numpy 2.3.1, python 3.12.11 pylsp v1.13.0, jedi 0.19.2

[davidhalter]

I personally would recommend you to try Zuban. I think it is already better than Jedi in a lot of ways (and worse in some). I plan on improving completions significantly over the next month. Most of my energy these days flows into making Zuban better, so I would recommend that (and I'm generally much faster working on issues there).

[ubaldot]

Oh, that sounds nice! I will certainly give it a shot. Btw, would that be possible to distribute it also as a conda-forge package?
Also, I hope that the LSP config stay as simple as in here without the need of going through 1000000000 parameters explained in the configuration section of other well-knowns LPS:s.

[davidhalter]

Sorry I don't want to maintain other distribution channels. This includes stuff like Debian packages, conda-forge and whatever else.

Also, I hope that the LSP config stay as simple as in here without the need of going through 1000000000 parameters explained in the configuration section of other well-knowns LPS:s.

No worries, I hate that as well. I'm a VIM user and I like to have things just working without config. I intentionally wrote Jedi and Zuban in a way that configuration should not be necessary.