querystring, url: repeated '&' in a paramsString will be skipped #10967

watilde · 2017-01-23T21:01:40Z

Updates:

skip the parsing process if a pair of a key and a value is empty
add test cases for querystring and URLSearchParams

Fixes: #10454

/cc @nodejs/url

Checklist

make -j4 test (UNIX), or vcbuild test (Windows) passes
tests and/or benchmarks are included
commit message follows commit guidelines

Affected core subsystem(s)

querystring, url, url-whatwg, test

TimothyGu · 2017-01-23T21:44:11Z

I believe the consensus was to freeze the querystring module, and put all new work onto URLSearchParams.

@jasnell, can you weigh in on this?

mscdex · 2017-01-23T22:44:32Z

@TimothyGu I don't recall that. Perhaps you're thinking about the url module instead?

TimothyGu · 2017-01-24T02:44:31Z

Regardless of any previous planning, I am not sure this PR (or any PRs that change the behavior of this module) is the way to go.

A change in the behavior of this module is almost definitively semver-major. It may also cause ripple effects in the entire ecosystem, though I am unsure how much of an effect this might bring.
querystring module has applications more than simply parsing and serializing URL query strings. The fact that sep, eq, and decodeURIComponent are modifiable means that other usage, including those that depend on the current behavior, is possible. I cannot find it right now, but I have even read a Node.js "tip" somewhere that encourages using this module for performant generic serialized format.
Even if the goal to make querystring WHATWG-compliant is accepted, there are other issues whose fixes may or must be semver-major as well.
1. Serialization of spaces as + rather than %20, left paren as ) rather than %27, etc.: url: Improve WHATWG URLSearchParams spec compliance #9484 (comment).
2. This module returns an object/hashmap, which is fundamentally incompatible with the list format WHATWG URL Standard mandates. For example, qs.stringify(qs.parse('a=1&b=2&a=3')) returns 'a=1&a=3&b=2' instead of 'a=1&b=2&a=3'.
Proposal to eventually move URLSearchParams parsing to native code.

jasnell · 2017-01-24T02:47:30Z

I wouldn't describe it as consensus. In my opinion significant changes to the the existing url module should be avoided, particularly semver-major or semver-minor. If it's a bug fix then I'm good with it. I'd rather keep things pretty steady state but this change makes sense. It's also worth noting that the URLSearchParams implementation currently still makes use of the querystring module.

joyeecheung · 2017-01-24T06:03:21Z

I think this looks more like a bugfix, albeit a breaking one...I doubt anyone would rely on the behavior of parsing && and getting an empty-ish object instead of treating it as a bug (although it's harmless for common usage AFAICT) ?

In fact you can't even reserialize to the original string with this bug. No error would be thrown if you only use the Node.js built-in module, but it could be a problem when it ends up in the hand of other modules in other languages.

> qs.parse('a=1&&b=1')
{ a: '1', '': '', b: '1' }
> qs.stringify(qs.parse('a=1&&b=1'))
'a=1&=&b=1'

FWIW, in Python:

>>> parse_qs('a=1&&b=2')
{'a': ['1'], 'b': ['2']}

PHP:

php > $foo = array();
php > parse_str('a=1&&b=2', $foo);
php > var_dump($foo);
array(2) {
  ["a"]=>
  string(1) "1"
  ["b"]=>
  string(1) "2"
}

Ruby:

2.3.1 :003 > CGI.parse("a=1&&b=2")
 => {"a"=>["1"], "b"=>["2"]}

joyeecheung · 2017-01-24T06:32:05Z

Regarding @TimothyGu 's concerns I think 1 might not be that significant seeing what other languages behaves. 2 would be more possible, but I can't think of a reason for relying on an empty key (if it's just for custom searialization), though some people could rely on an empty value, maybe changing || to && on L281 would be better?
As for 3&4, I think it doesn't hurt to fix the querystring module, whether URLSearchParams is gonna get reimplemented or not. It even makes more sense to land a fix in querystring if URLSearchParams is gonna be fixed separately.

watilde · 2017-01-24T08:52:37Z

I've tested the above case @joyeecheung mentioned, and it seems to work fine in this branch

$ ./out/Release/node
> var qs = require('querystring')
undefined
> qs.parse('a=1&&b=1')
{ a: '1', b: '1' }
> qs.stringify(qs.parse('a=1&&b=1'))
'a=1&b=1'

About changing || to &&, we have cases that the key or the value can be empty like the following. I tried updating it once, but some of the test cases threw some errors.

> qs.parse('=a&b')
{ '': 'a', b: '' }

joyeecheung · 2017-01-24T10:16:55Z

lib/querystring.js

-          else
-            obj[key] = [curValue, value];
+
+        if (key.length > 0 || value.length > 0) {


Maybe just key.length > 0? Empty values are not that uncommon (e.g. qs.parse('b=&x=1'))

Just took a look at the spec and this bit of code, I think the current way is the best :D. The only one that needs to be ignored is &&, otherwise it doesn't match what the spec needs.

Thanks for following up the code and the spec hardly!

joyeecheung · 2017-01-24T12:06:25Z

CI: https://ci.nodejs.org/job/node-test-pull-request/6023/

jasnell · 2017-01-24T16:07:06Z

I agree that it's a bugfix. Parsing && as meaning an empty key makes exceedingly little sense. I'm definitely +1 on getting this fixed.

jasnell

Almost there

jasnell · 2017-01-24T16:24:53Z

test/parallel/test-querystring.js

@@ -51,6 +51,8 @@ const qsTestCases = [
     __defineGetter__: 'baz' }],
  // See: https://github.com/joyent/node/issues/3058
  ['foo&bar=baz', 'foo=&bar=baz', { foo: '', bar: 'baz' }],
+  ['&&foo=bar&&', 'foo=bar', { foo: 'bar' }],
+  ['&&&&', '', {}],


It would be helpful to also test the following cases:

`a=b&c&d=e` `a=b&c=&d=e` `a=b&=c&d=e` `a=b&=&c=d`

That's right! Adding the cases that the key or the value is empty sounds useful for revealing the current behaviour. I've added them :)

jasnell · 2017-01-24T18:53:57Z

New CI since the test was updated: https://ci.nodejs.org/job/node-test-pull-request/6032/

watilde · 2017-01-25T08:55:50Z

Some of the nice performance improvements in querystring were added! Resolved conflicts.

watilde · 2017-01-25T09:02:59Z

CI: https://ci.nodejs.org/job/node-test-pull-request/6042/

TimothyGu · 2017-01-25T18:48:49Z

test/parallel/test-querystring.js

+  ['a=b&c&d=e', 'a=b&c=&d=e', { a: 'b', c: '', d: 'e' }],
+  ['a=b&c=&d=e', 'a=b&c=&d=e', { a: 'b', c: '', d: 'e' }],
+  ['a=b&=c&d=e', 'a=b&=c&d=e', { a: 'b', '': 'c', d: 'e' }],
+  ['a=b&=&c=d', 'a=b&c=d', { a: 'b', c: 'd' }],


This should be parsed as { '': '', a: 'b', c: 'd' }. Try new URLSearchParams('&=&').get('').

Wow I didn't know that, thanks! Then I will try to support the case.

Ok, I've updated! The logic was changed to simple one :)

watilde · 2017-02-01T20:39:47Z

I made improvements in performance again, and now it seems to be better than my last update.

Move one condition statement to a better line
make lint: ok
make test: ok
benchmark: https://gist.github.com/watilde/2401ff39ee3a5113da33d627409e3e38

mscdex · 2017-02-01T20:48:45Z

@watilde That benchmark output looks truncated (the benchmark/compare.js script by default performs more runs than that). Having the output piped to RScript benchmark/compare.R will yield more useful information as @joyeecheung pointed out.

watilde · 2017-02-02T07:04:45Z

@mscdex That's right indeed! Then I just tried it, and here is the result:

Https://gist.github.com/watilde/3f3ab84e3deb44287bee274081b5f64d

joyeecheung · 2017-02-02T08:03:44Z

CI: https://ci.nodejs.org/job/node-test-pull-request/6161/

joaocgreis · 2017-02-02T18:07:52Z

Aborted CI, only arm-fanned was pending because of nodejs/build#611

jasnell · 2017-02-02T20:27:21Z

New CI: https://ci.nodejs.org/job/node-test-pull-request/6172/

* update state machine in parse * repeated sep should be adjusted * `&=&=` should be `{ '': [ '', '' ] }` * add test cases for querystring and URLSearchParams Fixes: #10454 PR-URL: #10967 Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Timothy Gu <timothygu99@gmail.com>

jasnell · 2017-02-02T21:15:39Z

Landed in 4e259b2

TimothyGu · 2017-02-02T21:19:08Z

Should this be marked as semver-major?

mscdex · 2017-02-02T21:19:10Z

I was hoping to look at this this weekend before this landed...

jasnell · 2017-02-02T21:20:39Z

oy.. my apologies @mscdex... if you indicated that in the thread and I missed it, I do apologize.

jasnell · 2017-02-02T21:21:39Z

if this ends up not being acceptable to you @mscdex, then we should be able to do a quick revert

watilde · 2017-02-02T21:23:56Z

Isn't this patch update since it's just bug-fix?

mscdex · 2017-02-02T21:27:11Z

@watilde It might be a bug fix for the whatwg url module, but it's arguably not for the non-whatwg url module since there is no real spec that it adheres to. If this change were made to whatwg url's own querystring implementation that would be different and semver-major could just be applied to the non-whatwg url changes.

@jasnell #10967 (comment) I probably won't get time until this weekend to review this, so I don't think a 'quick revert' will happen unless else everyone agrees.

joyeecheung · 2017-02-02T23:00:39Z

I think we can wail until @mscdex 's review to decide then. Personally I think it's fine to accept this if it doesn't introduce any new behavior other than fixing #10454 because the old behavior looks buggy to me (see #10967 (comment)).

mscdex · 2017-02-02T23:09:47Z

@joyeecheung Calling it a 'bug fix' is debatable. One could also argue that stringify() should not include include key/value pairs that are both empty. I'd personally rather be on the safe side and land it in v8.0.0.

joyeecheung · 2017-02-02T23:17:42Z

@mscdex Sorry for being vague, I am fine with marking this semver-major.

* update state machine in parse * repeated sep should be adjusted * `&=&=` should be `{ '': [ '', '' ] }` * add test cases for querystring and URLSearchParams Fixes: nodejs#10454 PR-URL: nodejs#10967 Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Timothy Gu <timothygu99@gmail.com>

nodejs-github-bot added querystring Issues and PRs related to the built-in querystring module. dont-land-on-v7.x labels Jan 23, 2017

watilde mentioned this pull request Jan 23, 2017

'querystring.parse' works a inconsistent way with URLSearchParams(chrome) and python's urlparse.parse_qs #10454

Closed

watilde force-pushed the feature/fixes-querystring-parse branch from 7d22fc5 to 0701371 Compare January 23, 2017 21:13

nodejs-github-bot added the dont-land-on-v7.x label Jan 23, 2017

watilde changed the title ~~url: repeated '&' in a paramsString will be skipped~~ querystring, url: repeated '&' in a paramsString will be skipped Jan 23, 2017

mscdex removed the dont-land-on-v7.x label Jan 23, 2017

joyeecheung added url Issues and PRs related to the legacy built-in url module. whatwg-url Issues and PRs related to the WHATWG URL implementation. labels Jan 24, 2017

joyeecheung reviewed Jan 24, 2017

View reviewed changes

jasnell requested changes Jan 24, 2017

View reviewed changes

nodejs-github-bot added the dont-land-on-v7.x label Jan 24, 2017

jasnell approved these changes Jan 24, 2017

View reviewed changes

watilde force-pushed the feature/fixes-querystring-parse branch from 216afd3 to 840bb2f Compare January 25, 2017 08:53

nodejs-github-bot added the dont-land-on-v7.x label Jan 25, 2017

TimothyGu suggested changes Jan 25, 2017

View reviewed changes

watilde force-pushed the feature/fixes-querystring-parse branch from bfe41b6 to 7b6c578 Compare January 26, 2017 01:13

nodejs-github-bot added the dont-land-on-v7.x label Jan 26, 2017

watilde force-pushed the feature/fixes-querystring-parse branch from 666b945 to f81201a Compare February 1, 2017 20:39

TimothyGu approved these changes Feb 2, 2017

View reviewed changes

jasnell closed this Feb 2, 2017

mscdex added the semver-major PRs that contain breaking changes and should be released in the next major version. label Feb 2, 2017

watilde deleted the feature/fixes-querystring-parse branch February 2, 2017 21:21

watilde restored the feature/fixes-querystring-parse branch February 2, 2017 21:23

This was referenced Feb 6, 2017

querystring: fix state machine in querystring.parse #11171

Closed

querystring: fix empty pairs handling #11234

Closed

jasnell mentioned this pull request Apr 4, 2017

8.0.0 Release Proposal #12220

Closed

watilde deleted the feature/fixes-querystring-parse branch April 11, 2017 18:58

querystring, url: repeated '&' in a paramsString will be skipped #10967

querystring, url: repeated '&' in a paramsString will be skipped #10967

Conversation

watilde commented Jan 23, 2017 • edited Loading

Checklist

Affected core subsystem(s)

TimothyGu commented Jan 23, 2017

mscdex commented Jan 23, 2017

TimothyGu commented Jan 24, 2017

jasnell commented Jan 24, 2017

joyeecheung commented Jan 24, 2017 • edited Loading

joyeecheung commented Jan 24, 2017

watilde commented Jan 24, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

joyeecheung commented Jan 24, 2017

jasnell commented Jan 24, 2017

jasnell left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jasnell commented Jan 24, 2017

watilde commented Jan 25, 2017

watilde commented Jan 25, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

watilde commented Feb 1, 2017

mscdex commented Feb 1, 2017 • edited Loading

watilde commented Feb 2, 2017 • edited Loading

joyeecheung commented Feb 2, 2017

joaocgreis commented Feb 2, 2017

jasnell commented Feb 2, 2017

jasnell commented Feb 2, 2017

TimothyGu commented Feb 2, 2017

mscdex commented Feb 2, 2017 • edited Loading

jasnell commented Feb 2, 2017

jasnell commented Feb 2, 2017

watilde commented Feb 2, 2017

mscdex commented Feb 2, 2017 • edited Loading

joyeecheung commented Feb 2, 2017

mscdex commented Feb 2, 2017

joyeecheung commented Feb 2, 2017

watilde commented Jan 23, 2017 •

edited

Loading

joyeecheung commented Jan 24, 2017 •

edited

Loading

mscdex commented Feb 1, 2017 •

edited

Loading

watilde commented Feb 2, 2017 •

edited

Loading

mscdex commented Feb 2, 2017 •

edited

Loading

mscdex commented Feb 2, 2017 •

edited

Loading