URLs, CMake, and Networking

Alex Christensen, Apple

URLs

• guy in Netherlands who is trying to unify URL standard across all browsers
• implemented spec inside WebKit
• mostly a good idea
• currently passing 2/3 of tests
  • most browsers pass at least that amount (but of different tests)
• there are open issues (not perfect)
  • e.g. how to deal with unicode fragments
• the spec is changing
• implementation uses finite state machine to look at the string
  • zero allocations in one pass
  • certain characters (tab, newline, etc.) should be ignored
  • input can be any sequence of unicode
  • output *should* be ascii
    • %2E is sometimes a . (depending on the server)
  • ~1000 unit tests for a bunch of weird URLs
    • http://127.0.0.-/ should be invalid
  • modifies DOMURL and urls everywhere
    • interfaces with NSURL, etc.
  • ~25% faster than old parser
  • blog post coming soon :)
• unfortunately, there is always someone with “old” spec
  • should communicate to web authors
  • typically, authors don’t use weird edge cases
  • but, WebKit apps may implicitly assume some of these “old” points
    • e.g. not having the second / after scheme
  • as such, we cannot delete the old parser (for now)
• currently on by default, but only for Desktop and Mobile Safari
• these changes effect URL components as well
  • e.g. changes to query or path via JS

CMake

• works everywhere except on Mac using WebKit2
• slightly different directory structure with CMake
• >30% build time improvement compared to XCode
• we also have a bot for this
• build webkit --cmake
• preprocessing of IDLs must be done

Networking

• ResourceHandle class, designed for networking in 2001
  • contains a ResourceHandleClient that has information about the networking
    • functions have been added with Async for WebKit2, that doesn’t use the synchronous model of networking
    • will be changing the logic of all network code to remove synchronous methods
      • all code will now be asynchronous
  • will replace Network* instead of ResourceHandle* as the completion handler
    • designed to be more asynchronous