Dang it, how do I turn it back into a draft?

Codecov Report

Merging #9694 (611e327) into main (261afd9) will decrease coverage by 0%.
The diff coverage is 50%.

❗ Current head 611e327 differs from pull request most recent head 01920b3. Consider uploading reports for the commit 01920b3 to get more accurate results

@@          Coverage Diff          @@
##           main   #9694    +/-   ##
=====================================
- Coverage    63%     63%    -1%     
=====================================
  Files       204     204            
  Lines      9840    9843     +3     
  Branches   1556    1557     +1     
=====================================
+ Hits       6242    6243     +1     
+ Misses     3335    3100   -235     
- Partials    263     500   +237     

Not sure if you are still in Draft. But seems pretty peachy to me. NotebookControllerManager feels a good bit cleaner as the big plus.

Sorry there is one small downside that I forgot to mention.

Previously we used to only validate the preferred kernel when searching through the cache. Now we validate the entire cache before returning the preferred kernel.

So imagine you have 20 kernels in your cache, We will now make sure the file for all 20 intepreters exists before returning the preferred one. We used to only check for file existence on the first one.

I don't think a STAT call on 20 files is going to take very long, but it is a consequence of moving the caching around.

I think I could make it so the cache is not validated during the preferred search though. It's all internal to the kernel finder anyway.

I think the algorithm would have to be tweaked though. Like pass the validation into the findPreferredKernel function and instead of it returning the top of the list, it validates the top until it finds one that's valid.

But it seems like premature optimization to me.

I wrote some code to test that hypothesis:

var glob = require("glob");
var fsExtra = require("fs-extra");

async function run() {
// options is optional
var files = await new Promise((resolve, reject) => {
        glob('./src/**/*.ts', (ex, res) => {
            if (ex) {
                reject(ex);
            } else {
                resolve(res);
            }
        });
    });
    let start = new Date().getTime();
    await Promise.all(files.map(async (f) => {
      await fsExtra.pathExists(f);
    }));
    let end = new Date().getTime();
    console.log(`Total time for ${files.length} files is ${end-start}`);
}

run();

That prints out this:

Total time for 771 files is 15

So 15 ms for 771 files. So even if the user has a ridiculous number of interpreters, it's still super fast to check for existence.