IntelliSense not working for torch.nn

[Salzy]

Environment data

• Language Server version: 0.4.114.0
• OS and version: Debian 10
• Python version (& distribution if applicable, e.g. Anaconda): 3.7.3 (Miniconda - CondaForge)
• PyTorch version: 1.3.0 and 1.3.1

Expected behaviour

Auto-completion through IntelliSense when typing (i.e. torch.nn. ).

Actual behaviour

Auto-completion does not work for torch.nn. However, it works for other modules like torch.nn.functional.

I already tried to delete the language server directory and restarting VSCode. In addition, I already switched to beta, but nothing changed.

There are already closed related issues:
Intellisense doesnt work for all Pytorch methods
torch.nn missing from torch module and from torch import completions

Logs

Analysis of models.GAN.gan (User) queued. Dependencies: torch, torch.nn
Completions in file:///home/p***/Generative-models/models/GAN/gan.py at (19, 24)
Analysis version 4594 of 1 entries has started.
Analysis of models.GAN.gan (User) on depth 0 completed in 0.48 ms.
Analysis complete: 1 modules in 17.34 ms.
Analysis version 4594 of 1 entries has been completed in 0.66 ms.
IntelliCode identified 3 likely completion items.

Code Snippet / Additional lnformation

from torch import nn

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.model = nn.

[MikhailArkhipov]

From the log it looks like torch is still being analyzed and so far only partial data is known. Do you see

Analysis complete: 1152 modules in 16709.68 ms.
Analysis version 2024 of 1111 entries has been completed in 4644.41 ms.
Forcing full garbage collection and heap compaction.

in the Output? Any Analyzing, X items remaining in the status bar?
Setting

  "python.analysis.logLevel": "Trace",

may help to see if analysis is still running.

[vict0rsch]

I have the same issue with torch.nn not being recognized but import torch.nn.functional as F suggesting the right info.

I made a gist with the output of the extension's output with logLevel: trace => https://gist.github.com/vict0rsch/99510e0e527cb3bef6dfad04265999c3

[Salzy]

"python.analysis.logLevel": "Trace" enabled gives me the following (in the attachments you will find the full trace):

Hover in file:///home/p***/Generative-models/models/GAN/gan.py at (66, 21)
Analysis of gan (User) queued. Dependencies: torch, torch.nn, torch.nn.functional, numpy
Analysis version 1615 of 1 entries has started.
Analysis of gan (User) on depth 0 completed in 9.51 ms.
Analysis complete: 1 modules in 12.75 ms.
Analysis version 1615 of 1 entries has been completed in 9.95 ms.
Analysis of gan (User) queued. Dependencies: torch, torch.nn, torch.nn.functional, numpy
Analysis version 1616 of 1 entries has started.
Analysis of gan (User) on depth 0 completed in 4.3 ms.
Analysis complete: 1 modules in 9.25 ms.
Analysis version 1616 of 1 entries has been completed in 4.82 ms.
Analysis of gan (User) queued. Dependencies: torch, torch.nn, torch.nn.functional, numpy
Analysis version 1617 of 1 entries has started.
Analysis of gan (User) on depth 0 completed in 5.56 ms.
Analysis complete: 1 modules in 8.05 ms.
Analysis version 1617 of 1 entries has been completed in 6.1 ms.
Completions in file:///home/p***/Generative-models/models/GAN/gan.py at (67, 9)

trace.txt

[vict0rsch]

@MikhailArkhipov any news on that?

[dasturge]

Having this issue as well!

[MikhailArkhipov]

Could you try latest in beta channel? So far I couldn't repro it.

"python.analysis.downloadChannel": "beta"

[dasturge]

@MikhailArkhipov As an update, I actually get autocomplete just fine in normal operation, it's while in a remote-ssh session that I don't have the autocomplete for torch.nn.
my settings.json on the remote:
{
"python.condaPath": "~/anaconda3/bin/conda",
"python.dataScience.allowUnauthorizedRemoteConnection": true,
"python.dataScience.autoPreviewNotebooksInInteractivePane": true,
"python.pythonPath": "/home/username/anaconda3/envs/torch/bin/python",
"autoDocstring.docstringFormat": "sphinx",
"python.analysis.downloadChannel": "beta"
}

I put the beta download channel on both remote and local User settings for vscode, I don't know if remote-ssh actually looks at the remote config though.

[vict0rsch]

config looks unrecognized and issue not solved after reload of window ; that being said, is there anything else we could check in the analysis parameters? like caching level or else

I checked a clean folder: new virtual env, only one file with import torch=> same issue

FYI: trace gist

[MikhailArkhipov]

The setting is not 'public' it is mostly for development purposes so yes, it is normally appear grayed out.
Might be a path watcher bug, PR #1841 is up.

[MikhailArkhipov]

#1841 is in 0.5.10+

[vict0rsch]

How can I check which version of the language server I'm running?

[jakebailey]

https://github.com/microsoft/python-language-server/wiki/Versioning-and-download-channels

[vict0rsch]

thanks ; I'll check now

[vict0rsch]

Still there. Issue not fixed. trace gist

[MikhailArkhipov]

Could you attach the log from output window please