GeezerVison

Ryan Cavanaugh (@RyanCavanaugh) Vladimir Matveev (@vladima) Mohamed Hegazy (@mhegazy) comments?

Adding cross-reference to #7758

What's the perf impact of parsing the extra comments? Any?

I ran the following sample for the perf bench mark:

The Setup:

In an empty folder, do

npm install ember-cli

(I chose ember-cli because it is big enough to see the bind/parsing time difference)
then create an empty jsconfig.json with the following content:

{
    "exclude": []
}

The bench mark script is:

Write-Output "TSC runs:"
0..10 | % {tsc -p ".\jsconfig.json" --noEmit --diagnostics | Select-String "Bind Time|Parse Time"}

Write-Output "PR runs:"
0..10 | % { node --max-old-space-size=2000 "C:\Users\lizhe\Documents\github\TypeScript\built\local\tsc.js" -p ".\jsconfig.json" --noEmit --diagnostics | Select-String "Bind Time|Parse Time"}

The Result:

So it turns out that my previous benchmark wasn't very accurate at all, for several reasons:

1. I was running the tests on a laptop on battery mode, and there are browsers / other background programs running at the same time
2. The tsc.js for the PR was generated via calling jake local, which by default is in "debug mode", things like const enums are preserved for debugging purpose. I should instead use jake release local to generate the tsc.js.
3. I used the command line tsc for the baseline, which may have node different parameters with what I used for the PR tests.
4. On the PR I didn't pull from the latest master, so other changes may have impacted the results as well.

I rerun the benchmarks with these factors fixed. The new environment is:
a) a dedicated Ubuntu machine with no graphic interface (ssh)
b) a fresh installed windows 10 machine that is:

• not domain-joined
• no active tasks in task scheduler
• desktop with high-performance power management options
• as few background apps as possible (no onedrive, browser etc.)

c) use node for both cases, the new script is:

Param($count)

Write-Output "PR runs:"
0..($count - 1) | % { node --max-old-space-size=2000 "C:\Users\lan\Documents\github\TypeScript\built\local\tsc.js" -p ".\jsconfig.json" --noEmit --diagnostics | Select-String "Bind Time"}

Write-Output "Master runs:"
0..($count - 1) | % { node --max-old-space-size=2000 "C:\Program Files\nodejs\node_modules\typescript\lib\tsc.js" -p ".\jsconfig.json" --noEmit --diagnostics | Select-String "Bind Time" }

d) use the latest nightly and merge the PR with master locally. Checked the diff of two tsc.js files to make sure that the only differences are indeed caused by the PR.

And the benchmark result shows that the PR is not slower in binding time:

a) on the Ubuntu machine:

b) on the Windows machine:

👍 especially if it's somehow faster 😉

I would add a navigate to test for typedefs

👍

#8849 touches the same areas, can you merge this in after that ones goes in.