20x slowdown compared to 0.11.x

[alexdima]

• I have checked issues with bug label and found no duplicates

Expected Behavior

Typedoc works as before, and takes reasonable time to execute.

Actual Behavior

Typedoc is super super slow, going from 2.5s (v0.11.1) to 49s (v0.14.2).

Steps to reproduce the bug

• save as monaco.d.ts - monaco.d.ts.txt
• run in PowerShell:

 Measure-Command {..\typedoc-0.11.1\node_modules\.bin\typedoc --mode file --out out --includeDeclarations --entryPoint monaco monaco.d.ts}
TotalMilliseconds : 2499.9106

Measure-Command {..\typedoc-0.13.0\node_modules\.bin\typedoc --mode file --out out --includeDeclarations --entryPoint monaco monaco.d.ts}
TotalMilliseconds : 44939.4815

Measure-Command {..\typedoc-0.14.2\node_modules\.bin\typedoc --mode file --out out --includeDeclarations --entryPoint monaco monaco.d.ts}
TotalMilliseconds : 48930.0364

Environment

• Typedoc version: 0.14.2
• Node.js version: 0.15.1
• OS: Windows

[aciccarello]

Thanks for the report. I wonder if this is a windows focused issue.

Related to #960

[Gerrit0]

I think this must be a windows issue, I haven't run into it. Thanks for the reproduction! Time to pull out git bisect and see what went wrong.

[Gerrit0]

git bisect says d4b2f56 is the first bad commit, I haven't had time to look into what exactly the issue is.

[Gerrit0]

Finally getting back to this, and I still have absolutely no clue why updating to TS 3.x causes this major slowdown when compilation speed remains essentially constant.

The Chrome devtools simply can't handle TypeDoc when trying to parse the above file. I get a hard crash as soon as the profiling is complete, which is making this more difficult than it ought to be.

Even the built in profiler with --prof has issues, when trying to generate useful output from the log file with node --prof-process --preprocess -j .\isolate-000002C68C42A0C0-v8.log | flamebearer I get a ton of Code move event for unknown code: 0x1ba.. errors.

Despite this, I have managed to discover a few performance issues which let me reduce the runtime of the 20x slower code to 15x slower... which obviously isn't great. I'm still missing the core issue. Hopefully I'll be able to discover what exactly is happening with more investigation.

[Gerrit0]

Well, whatever the issue was... we're back to being pretty fast.

C:\Users\gtbir\Desktop\library-tests> measure-command { npx typedoc .\src\monaco.d.ts }
Rendering [========================================] 100%

...
TotalMilliseconds : 4750.1524