This is a repro for https://groups.google.com/forum/#!topic/v8-users/KddMkLLHh2w

• clone repository
• npm install .
• node setup.js
  • this will download Electron and unpack it twice, under ~/tmp/electron-original and ~/tmp/electron-snapshot
  • we have two installations of Electron because we will mess with (change) one of it, we will change its snapshot_blob.bin.
• node prepare.js
  • this will create ~/tmp/bundle.js (from the VSCode compilation output in ~/out)
  • ~/tmp/bundle.js will be loaded through mksnapshot and the snapshot_blob.bin will be placed in the expected location inside ~/tmp/electron-snapshot

• both executions use --no-lazy
• ./run1.sh
  • this executes Electron from ~/tmp/electron-original.
  • this will create an Electron browser window from index1.html.
  • it will load via vm the script ~/tmp/bundle.js.
  • on first execution, it will create ~/tmp/bundle-cached-data, so please run at least twice to get good time measurements.
  • on second and subsequent executions, it will use ~/tmp/bundle-cached-data.
  • finally, once ~/tmp/bundle.js is loaded, it will invoke the benchmark.
• ./run2.sh
  • this executes Electron from ~/tmp/electron-snapshot.
  • this will create an Electron browser window from index2.html.
  • because we have created a snapshot from ~/tmp/bundle.js, it does not need to load anything
  • it will directly invoke the benchmark

(Measurements done on a MacBook Pro, 2016, 3.3GHz Intel Core i7) The first number is the total time, of which, the second number is invoking 'vs/workbench/electron-browser/main'.startup(configuration) in ms. Each test is rerun by killing (ctrl+c) the executable from the terminal, i.e. not by reloading the window:

> ./run1.sh
1747, 359 # excluded
767, 366
723, 331
796, 347
799, 353
830, 373
802, 344
793, 357
789, 351
797, 355
795, 352
> ./run2.sh
746, 703 # excluded
458, 433
479, 455
475, 451
458, 437
481, 458
474, 451
457, 435
475, 452
482, 456
501, 478

• Average:
  • First column: ./run1.sh: 789.1, 352.9
  • Second column: ./run2.sh: 474.0, 450.6
  • invoking the startup code is 27% slower when using a startup snapshot

  

If you are curious to see what is actually executing, uncomment the <link> tag on line number 6 in index1.html or index2.html (the CSS is commented out to not affect measurements):