Taking over 45 minutes to analyze project.

[bob-white]

VSCode version: 1.28.2
Python extension version: 2018.9.2
LanguageServer version: 0.1.52
Python 3.5
I have the following settings enabled.

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

After startup in the output window I see all the files in the current workspace directory being parsed and queued up for analysis (1510 files).
I see one single file get returned with:

Analysis complete for file://...blah blah blah

The AnalysisLog.txt file is largely filled with ImportNotFound lines about 3200 out of the 4000 lines in the file currently.

Is there any way to figure out why this is happening? Just earlier this week everything was working, granted it still took around 5-10 minutes to completely analyze the workspace, but at least it finished.

After about an hour and 15 minutes it finished analysis at version -1, and kicked off version 0.
I have yet to see a single file complete at version 0.

And after another hour and 15 minutes it finally stopped processing everything.

[jakebailey]

@bob-white Is the codebase you're testing available somewhere so we can try to reproduce this?

[bob-white]

Codebase isn't publicly available. Trying to see what the possibility is of getting something you guys can repro with though.

I was able to create a debug build of master and attach to that, and it appears I get stuck in DGG.Analyzer the first file has a queue size in the low 20s, and the second time I hit the function the queue has a size of 30k+.

I did discover that if I create a bunch of stubs using mypy's stubgen utility, after renaming them to .py the analysis seems to follow the same pattern of process one file, and then ALL the files. With approximately the same queue sizes in DGG.Analyzer.
Except it handles it in 45 seconds instead of an hour and fifteen.

If there is anything else I could test / inspect to maybe help narrow down the cause on my end just let me know.

[MikhailArkhipov]

Unfortunately, this is not unusual with 1510 files. We are planning to address this by several changes that are filed in #29. However, this is not coming immediately since improvements require significant changes.

[bob-white]

So I am kind of confused, I don't see any changes related to #29.
But otherwise this sounds like I should just go back to using Jedi and hope that a future update maybe fixes things?
Should I keep an eye on this issue or just watch #29 so that I can test stuff out if / when improvements start to land?

[jakebailey]

We use ZenHub, and #29 is an "epic" which links to 7 other issues, but isn't visible without the extension. I've just tagged them with a "performance" label, which should be easier to look at and check.

I would personally say to use what works for you. There are some changes on the horizon which will improve performance for non-library code, but they haven't been finished quite yet.

[bob-white]

Cool. Can't wait to give it a shot. Also thanks for tagging those, makes a lot more sense now.

[jeffery9]

have the same issue.. you can try to analysis https://github.com/odoo/odoo, it never finished the code analysis, jedi do this very quickly.

[MikhailArkhipov]

#546