url.parse now encodes components

[mojodna]

This behavior changed between Node-0.10 and 0.12/iojs and does not appear to be documented anywhere.

io.js v2.3.3:

> url.parse("http://example.com/{z}/{x}/{y}.png#foo{}")
{ protocol: 'http:',
  slashes: true,
  auth: null,
  host: 'example.com',
  port: null,
  hostname: 'example.com',
  hash: '#foo%7B%7D',
  search: null,
  query: null,
  pathname: '/%7Bz%7D/%7Bx%7D/%7By%7D.png',
  path: '/%7Bz%7D/%7Bx%7D/%7By%7D.png',
  href: 'http://example.com/%7Bz%7D/%7Bx%7D/%7By%7D.png#foo%7B%7D' }

node.js v0.10.39:

> url.parse("http://example.com/{z}/{x}/{y}.png#foo{}")
{ protocol: 'http:',
  slashes: true,
  auth: null,
  host: 'example.com',
  port: null,
  hostname: 'example.com',
  hash: '#foo{}',
  search: null,
  query: null,
  pathname: '/{z}/{x}/{y}.png',
  path: '/{z}/{x}/{y}.png',
  href: 'http://example.com/{z}/{x}/{y}.png#foo{}' }

This also means that URIs no longer roundtrip:

iojs v2.3.3

> url.format(url.parse("http://example.com/{z}/{x}/{y}.png#foo{}"))
'http://example.com/%7Bz%7D/%7Bx%7D/%7By%7D.png#foo%7B%7D'

node.js v0.10.39:

> url.format(url.parse("http://example.com/{z}/{x}/{y}.png#foo{}"))
'http://example.com/{z}/{x}/{y}.png#foo{}'

[rvagg]

This should go on to https://github.com/joyent/node/wiki/API-changes-between-v0.10-and-v0.12 but does anyone have a git ref that can explain the context around this?

[Fishrock123]

@mojodna any idea if this difference is also between 0.12 and 0.10?

[mojodna]

@Fishrock123 it is: nodejs/node-v0.x-archive#25636

[brendanashworth]

I believe we have our lucky winner/s: 17a379e, 881ef7c

found with guessing / version bisecting with v0.11.0 and v0.11.1

[Fishrock123]

Interesting. See nodejs/node-v0.x-archive#5474 and linked issues.

[brendanashworth]

@Fishrock123 hmm, seems like they put two patches up (38149bb, 17a379e) to fix the same bug and reverted the one in http (7124387), leaving url. Full backwards compatibilty?.. nahh ¯\_(ツ)_/¯

[mojodna]

The inability to cleanly round-trip is my biggest complaint about the regression, as it leads to code like this (plus, I'm probably forgetting to decode some components):

if (semver.satisfies(process.version, ">=0.11.0")) {
  // Node 0.12 changes the behavior of url.parse such that components are
  // url-encoded
  uri.hash = decodeURIComponent(uri.hash);
  uri.pathname = decodeURIComponent(uri.pathname);
  uri.path = decodeURIComponent(uri.path);
  uri.href = decodeURIComponent(uri.href);
}

(from mojodna/tilelive-http@a74871c#diff-168726dbe96b3ce427e7fedce31bb0bcR90)

[Fishrock123]

Crap we should probably actually fix this for 4.0.0.. hmmm

[Fishrock123]

cc @bnoordhuis since he reverted the http patch.

[rvagg]

@Fishrock123 put it on the 4.0.0 milestone if you feel strongly enough about it and think it needs to be done

[Fishrock123]

something needs to be done, I'm not sure if I should revert it or document it..

[mojodna]

My preference (for whatever it's worth) would be to revert it, solely on the basis of round-tripping not working.