Jupyter server and python interactive crashes on launch, trying to find temp.txt

[michaelberks]

Environment data

• VS Code version: 1.33.1
• Extension version (available under the Extensions sidebar): 2019.3.6558
• OS and version: Microsoft Windows 10.0
• Python version (& distribution if applicable, e.g. Anaconda): Anaconda 3.6.4
• Type of virtual environment used (N/A | venv | virtualenv | conda | ...): Tried both default base above, and new, clean Anaconda python 3.7.3 environment, same error in both
• Relevant/affected Python packages and their versions: jupyter 1.0.0

Expected behaviour

Jupyter server starts and runs. I have not used VS code for the last month. Previously I used jupyter interactive cell mode in .py files via Don Jayamanne's Jupyter extension, everything worked fine. On opening VS code this morning I allowed it to install update to version 1.33.1, and then noted the warning re the Jupyter extension being deprecated, so disabled that package. The new problem occurs trying to work interactively with the new python extension.

I tried re-enabling and running the old Jupyter extension, however that fails, with error related to previewHTML, which I assumed fell under the banner of the extension being no longer supported. The combination of the two means my usual workflow is now completely inoperable.

Actual behaviour

Anything that requires starting the jupyter server or python interactive window (eg. Clicking on 'Run cell' in .py file, selecting 'Show Python Interactive Window' or 'Run Selection/Line in Python Interactive Window') causes the following error notification to pop-up:

Activating Python 3.6.4 64-bit ('base': conda) to run Jupyter failed with Error: StdErr from ShellExec, Access is denied.
The system cannot find the file specified.
Could Not Find C:\Program Files\Microsoft VS Code\temp.txt
Access is denied.
The system cannot find the file specified.
Could Not Find C:\Program Files\Microsoft VS Code\temp.txt.

One thing I noticed: if I select a line and hit 'Shift+Enter' so it runs in the ipython terminal, in the bunch of commands that appear in the terminal window (presumably triggered by whatever internal VS python extension script is being) prior to the iPython terminal launching for the first time, the following are the first 4:

C:/Anaconda3/Scripts/activate
python C:\Anaconda3\etc\keras\load_config.py 1>temp.txt
set /p KERAS_BACKEND= 0<temp.txt
del temp.txt

AFAICT this just a means to set an environment variable with the Keras backend - but the fact it was trying to do this via writing to and from temp.txt made me wonder if something similar was being initiated in jupyter server launcher, which was failing bacuse the working directory for VS code (C:\Program Files\Microsoft VS Code) is write protected.

Steps to reproduce:

1. Open visual studio code
2. Create a file test_jupyter.py, with lines:

#%%
print('Hello')

3. Click 'Run cell'

OR

1. Open visual studio code
2. Open command pallette and select 'Python: Show Python Interactive window'

Logs

Output for Python in the Output panel (View→Output, change the drop-down the upper-right of the Output panel to Python)

Starting Jedi Python language engine.
##########Linting Output - pylint##########

--------------------------------------------------------------------

Your code has been rated at 10.00/10 (previous run: 10.00/10, +0.00)

Output from Console under the Developer Tools panel (toggle Developer Tools on under Help; turn on source maps to make any tracebacks be useful by running Enable source map support for extension debugging)

Python Extension: getActivatedEnvironmentVariables, Class name = b, Arg 1: undefined, Arg 2: {"architecture":3,"path":"C:\\Anaconda3\\python.exe","version":{"raw":"3.6.4-final","major":3,"minor":6,"patch":4,"prerelease":["final"],"build":[],"version":"3.6.4-final"},"sysPrefix":"C:\\Anaconda3","fileHash":"2706b186a4d0a2e33818a0350c54fd24995fa8855d66ae99f6e916bc0006323a77b017071861b61e86321f075287cbba7ea27ffc18157ee3f590b7e065c7428f","companyDisplayName":"Anaconda, Inc.","type":"Conda","envName":"base","envPath":"C:\\Anaconda3","displayName":"Python 3.6.4 64-bit ('base': conda)"}, Arg 3: true Error: StdErr from ShellExec, Access is denied.
The system cannot find the file specified.
Could Not Find C:\Program Files\Microsoft VS Code\temp.txt
Access is denied.
The system cannot find the file specified.
Could Not Find C:\Program Files\Microsoft VS Code\temp.txt

	at b.<anonymous> (C:\Users\momeemb2\.vscode\extensions\ms-python.python-2019.3.6558\out\client\extension.js:83:594490)
	at Generator.next (<anonymous>)
	at s (C:\Users\momeemb2\.vscode\extensions\ms-python.python-2019.3.6558\out\client\extension.js:83:592836)

[rchiodo]

We don't do the temp.txt output. That must be something in your conda activation scripts. Perhaps conda recently changed to doing this.

[rchiodo]

@DonJayamanne this is more of a general purpose failure. Running from the terminal fails as well.

[michaelberks]

Ok, just to confirm, it was the keras activation that was causing the issue, and specifically, these lines (2-5) in keras_activate.bat (for my anaconda this is stored in CONDA_ROOT\etc\conda\activate.d*):

python %CONDA_PREFIX%\etc\keras\load_config.py > temp.txt
set /p KERAS_BACKEND=<temp.txt
del temp.txt

This activation script gets called when the jupyter server tries to initialise anaconda, and fails if the application doesn't have write permissions in the working directory it's running from.

I've solved for now, commenting out the above lines and hardcoding KERAS_BACKEND=tensorflow (which will be overwritten to theano in the next block of the activation script if the tensorflow backend won't import).

Worth logging here in case any other users have similar problems.

[EDIT] The only thing I'm confused on is I'm 99% sure the keras activation scripts haven't changed since everything worked with the old VS code Jupyter extension, so it still suggests some change in how things get called with the new Python extension/VS code version. Did the old extension not call these scripts? Or did it do so from another working directory?

*Note this is symlinked to the keras package directory *CONDA_ROOT\pkgs\keras-X.Y.Z-py36_0\etc\conda\activate.d*

[rchiodo]

So this was in your conda activate script? Or in a keras specific script? Just wondering if we should log an issue on Keras or Anaconda.

[michaelberks]

I think Keras specific. Sorry, I don't really know enough about the underlying mechanics of all this to be sure.

But from what I can work out, when you conda install keras into an environment, it sym-links the keras_activate.bat script from the keras package into a folder etc\conda\activate.d that contains scripts that get called when conda activate is called on that environment.

So the keras specific script gets called by something that is run whenever conda activate is called - and thus when the VS code python extension tries to launch the jupyter server, it first tries to actiavte the current python environment, which triggers the call to conda activate, and hence the troublesome keras script.

I assume if you've never installed keras this wouldn't have been a problem - but I don't have a fully clean anaconda installation to test that.

[rchiodo]

Yeah sounds like a keras problem. At least somebody reported it there. I linked the bug from Keras that's mostly relevant.

[rchiodo]

Sorry never answered this question:

EDIT] The only thing I'm confused on is I'm 99% sure the keras activation scripts haven't changed since everything worked with the old VS code Jupyter extension, so it still suggests some change in how things get called with the new Python extension/VS code version. Did the old extension not call these scripts? Or did it do so from another working directory?

A couple of months ago we changed all of our stuff to call conda activate before running. This is necessary to get a number of python modules to load (like numpy)