net/url: URL allows malformed query round trip

[artyom]

What did you do?

package main

import (
	"fmt"
	"log"
	"net/url"
)

func main() {
	u, err := url.Parse("http://example.com/bad path/?bad query#bad fragment")
	if err != nil {
		log.Fatal(err)
	}
	fmt.Println(u.String())
}

https://play.golang.org/p/hdX1zpv3BN

What did you expect to see?

I expect either url.Parse return a non-nil error or URL.String method return fully escaped url representation — http://example.com/bad%20path/?bad%20query#bad%20fragment — with query being escaped the same way as path or fragment.

What did you see instead?

http://example.com/bad%20path/?bad query#bad%20fragment

For the reference, such url is rejected by net/http.Server: https://play.golang.org/p/2gujmbXZlu

Does this issue reproduce with the latest release (go1.9.2)?

Yes

System details

go version devel +9a13f8e11c Tue Nov 28 06:47:50 2017 +0000 darwin/amd64
GOARCH="amd64"
GOBIN=""
GOCACHE="/Users/artyom/Library/Caches/go-build"
GOEXE=""
GOHOSTARCH="amd64"
GOHOSTOS="darwin"
GOOS="darwin"
GOPATH="/tmp/go:/Users/artyom/go"
GORACE=""
GOROOT="/Users/artyom/Repositories/go"
GOTMPDIR=""
GOTOOLDIR="/Users/artyom/Repositories/go/pkg/tool/darwin_amd64"
GCCGO="gccgo"
CC="clang"
CXX="clang++"
CGO_ENABLED="1"
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -m64 -pthread -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -fdebug-prefix-map=/var/folders/lb/3rk8rqs53czgb4v35w_342xc0000gn/T/go-build624293827=/tmp/go-build -gno-record-gcc-switches -fno-common"
GOROOT/bin/go version: go version devel +9a13f8e11c Tue Nov 28 06:47:50 2017 +0000 darwin/amd64
GOROOT/bin/go tool compile -V: compile version devel +9a13f8e11c Tue Nov 28 06:47:50 2017 +0000
uname -v: Darwin Kernel Version 17.2.0: Fri Sep 29 18:27:05 PDT 2017; root:xnu-4570.20.62~3/RELEASE_X86_64
ProductName:	Mac OS X
ProductVersion:	10.13.1
BuildVersion:	17B48
lldb --version: lldb-900.0.57
  Swift-4.0

https://tools.ietf.org/html/rfc3986#section-3.4 states that query component should be defined as (appendix A):

  query       = *( pchar / "/" / "?" )
  pchar         = unreserved / pct-encoded / sub-delims / ":" / "@"
  unreserved    = ALPHA / DIGIT / "-" / "." / "_" / "~"
  pct-encoded   = "%" HEXDIG HEXDIG
  sub-delims    = "!" / "$" / "&" / "'" / "(" / ")"
                  / "*" / "+" / "," / ";" / "="

There's no whitespace character in this list. whatwg agrees on that:

A URL-query string must be zero or more URL units.

[...]

The URL units are URL code points and percent-encoded bytes.

[...]

The URL code points are ASCII alphanumeric, U+0021 (!), U+0024 ($), U+0026 (&), U+0027 ('), U+0028 LEFT PARENTHESIS, U+0029 RIGHT PARENTHESIS, U+002A (*), U+002B (+), U+002C (,), U+002D (-), U+002E (.), U+002F (/), U+003A (:), U+003B (;), U+003D (=), U+003F (?), U+0040 (@), U+005F (_), U+007E (~), and code points in the range U+00A0 to U+10FFFD, inclusive, excluding surrogates and noncharacters.

[namusyaka]

note:
This seems to be related to this line, and its behavior looks reasonable.

[namusyaka]

I have investigated the issue and confirmed that this comment is incorrect, or at least the current (*URL) String() implementation is wrong.
In fact, query components aren't escaped at all, not only in this example.
If the comment is positive, I think we should escape the query component appropriately.
/cc @tombergan

[fraenkel]

@namusyaka I think you are mistaken. The problem is with all forms of Parse which does no validation on the query or fragment pieces. The parse of the uri should have failed. Since it does not, the Fragment or RawQuery fields are incorrect and hence String() which is doing the right thing appears to provide a bad url.

All that has to happen is to apply validation of the query and fragment pieces in parse, and that should solve the issues. It might start breaking code that is sending around "bad" URLs.

[fraenkel]

Its not clear that this can be fixed given the behavior of shouldEscape() which states that most of the sub-delims should escape.

[namusyaka]

@fraenkel Yes, I knew it's more easy to understand the behavior.
However, if your opinion will reflect to the implementation, another cases such as http://example.com/foo bar, http://example.com/foo? bar and http://example.com/foo?bar# baz should be failed on parse, and then I'm worried that breaks backword-compatibility.

[gopherbot]

Change https://golang.org/cl/99135 mentions this issue: net/url: reject invalid query strings when parsing URLs.

[bradfitz]

Copying my comment from the CL:

I changed the commit message to accurately reflect what later versions of this CL did. It now says:

net/url: escape URL.RawQuery on Parse if it contains invalid characters

But while reviewing this, I'm worried it might change the behavior of:

https://golang.org/pkg/net/url/#URL.Query
which says:

"It silently discards malformed value pairs. To check errors use ParseQuery."

Before it silently discarded things, and with this CL it would start returning escaped versions instead. Do we care?

---

I think we probably should wait for Go 1.12 at this point. This needs a decision on what to do.

[bradfitz]

Reopening, as the fix is being reverted.

[gopherbot]

Change https://golang.org/cl/137716 mentions this issue: Revert "net/url: escape URL.RawQuery on Parse if it contains invalid characters"

[agnivade]

@bradfitz - Any thoughts on this ? Or do we want to push to 1.13 ?