// Copyright 2013 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include <errno.h>
#include "base/macros.h"
#include "testing/gtest/include/gtest/gtest.h"
#include "url/url_canon.h"
#include "url/url_canon_internal.h"
#include "url/url_canon_stdstring.h"
#include "url/url_parse.h"
#include "url/url_test_utils.h"
// Some implementations of base/basictypes.h may define ARRAYSIZE.
// If it's not defined, we define it to the ARRAYSIZE_UNSAFE macro
// which is in our version of basictypes.h.
#ifndef ARRAYSIZE
#define ARRAYSIZE ARRAYSIZE_UNSAFE
#endif
namespace url {
using test_utils::WStringToUTF16;
using test_utils::ConvertUTF8ToUTF16;
using test_utils::ConvertUTF16ToUTF8;
namespace {
struct ComponentCase {
const char* input;
const char* expected;
Component expected_component;
bool expected_success;
};
// ComponentCase but with dual 8-bit/16-bit input. Generally, the unit tests
// treat each input as optional, and will only try processing if non-NULL.
// The output is always 8-bit.
struct DualComponentCase {
const char* input8;
const wchar_t* input16;
const char* expected;
Component expected_component;
bool expected_success;
};
// Test cases for CanonicalizeIPAddress(). The inputs are identical to
// DualComponentCase, but the output has extra CanonHostInfo fields.
struct IPAddressCase {
const char* input8;
const wchar_t* input16;
const char* expected;
Component expected_component;
// CanonHostInfo fields, for verbose output.
CanonHostInfo::Family expected_family;
int expected_num_ipv4_components;
const char* expected_address_hex; // Two hex chars per IP address byte.
};
std::string BytesToHexString(unsigned char bytes[16], int length) {
EXPECT_TRUE(length == 0 || length == 4 || length == 16)
<< "Bad IP address length: " << length;
std::string result;
for (int i = 0; i < length; ++i) {
result.push_back(kHexCharLookup[(bytes[i] >> 4) & 0xf]);
result.push_back(kHexCharLookup[bytes[i] & 0xf]);
}
return result;
}
struct ReplaceCase {
const char* base;
const char* scheme;
const char* username;
const char* password;
const char* host;
const char* port;
const char* path;
const char* query;
const char* ref;
const char* expected;
};
// Magic string used in the replacements code that tells SetupReplComp to
// call the clear function.
const char kDeleteComp[] = "|";
// Sets up a replacement for a single component. This is given pointers to
// the set and clear function for the component being replaced, and will
// either set the component (if it exists) or clear it (if the replacement
// string matches kDeleteComp).
//
// This template is currently used only for the 8-bit case, and the strlen
// causes it to fail in other cases. It is left a template in case we have
// tests for wide replacements.
template<typename CHAR>
void SetupReplComp(
void (Replacements<CHAR>::*set)(const CHAR*, const Component&),
void (Replacements<CHAR>::*clear)(),
Replacements<CHAR>* rep,
const CHAR* str) {
if (str && str[0] == kDeleteComp[0]) {
(rep->*clear)();
} else if (str) {
(rep->*set)(str, Component(0, static_cast<int>(strlen(str))));
}
}
} // namespace
TEST(URLCanonTest, DoAppendUTF8) {
struct UTF8Case {
unsigned input;
const char* output;
} utf_cases[] = {
// Valid code points.
{0x24, "\x24"},
{0xA2, "\xC2\xA2"},
{0x20AC, "\xE2\x82\xAC"},
{0x24B62, "\xF0\xA4\xAD\xA2"},
{0x10FFFF, "\xF4\x8F\xBF\xBF"},
};
std::string out_str;
for (size_t i = 0; i < ARRAYSIZE(utf_cases); i++) {
out_str.clear();
StdStringCanonOutput output(&out_str);
AppendUTF8Value(utf_cases[i].input, &output);
output.Complete();
EXPECT_EQ(utf_cases[i].output, out_str);
}
}
#if defined(GTEST_HAS_DEATH_TEST)
// TODO(mattm): Can't run this in debug mode for now, since the DCHECK will
// cause the Chromium stacktrace dialog to appear and hang the test.
// See http://crbug.com/49580.
#if defined(NDEBUG) && !defined(DCHECK_ALWAYS_ON)
#define MAYBE_DoAppendUTF8Invalid DoAppendUTF8Invalid
#else
#define MAYBE_DoAppendUTF8Invalid DISABLED_DoAppendUTF8Invalid
#endif
TEST(URLCanonTest, MAYBE_DoAppendUTF8Invalid) {
std::string out_str;
StdStringCanonOutput output(&out_str);
// Invalid code point (too large).
ASSERT_DEBUG_DEATH({
AppendUTF8Value(0x110000, &output);
output.Complete();
EXPECT_EQ("", out_str);
}, "");
}
#endif // defined(GTEST_HAS_DEATH_TEST)
TEST(URLCanonTest, UTF) {
// Low-level test that we handle reading, canonicalization, and writing
// UTF-8/UTF-16 strings properly.
struct UTFCase {
const char* input8;
const wchar_t* input16;
bool expected_success;
const char* output;
} utf_cases[] = {
// Valid canonical input should get passed through & escaped.
{"\xe4\xbd\xa0\xe5\xa5\xbd", L"\x4f60\x597d", true, "%E4%BD%A0%E5%A5%BD"},
// Test a characer that takes > 16 bits (U+10300 = old italic letter A)
{"\xF0\x90\x8C\x80", L"\xd800\xdf00", true, "%F0%90%8C%80"},
// Non-shortest-form UTF-8 are invalid. The bad char should be replaced
// with the invalid character (EF BF DB in UTF-8).
{"\xf0\x84\xbd\xa0\xe5\xa5\xbd", NULL, false, "%EF%BF%BD%E5%A5%BD"},
// Invalid UTF-8 sequences should be marked as invalid (the first
// sequence is truncated).
{"\xe4\xa0\xe5\xa5\xbd", L"\xd800\x597d", false, "%EF%BF%BD%E5%A5%BD"},
// Character going off the end.
{"\xe4\xbd\xa0\xe5\xa5", L"\x4f60\xd800", false, "%E4%BD%A0%EF%BF%BD"},
// ...same with low surrogates with no high surrogate.
{"\xed\xb0\x80", L"\xdc00", false, "%EF%BF%BD"},
// Test a UTF-8 encoded surrogate value is marked as invalid.
// ED A0 80 = U+D800
{"\xed\xa0\x80", NULL, false, "%EF%BF%BD"},
};
std::string out_str;
for (size_t i = 0; i < ARRAYSIZE(utf_cases); i++) {
if (utf_cases[i].input8) {
out_str.clear();
StdStringCanonOutput output(&out_str);
int input_len = static_cast<int>(strlen(utf_cases[i].input8));
bool success = true;
for (int ch = 0; ch < input_len; ch++) {
success &= AppendUTF8EscapedChar(utf_cases[i].input8, &ch, input_len,
&output);
}
output.Complete();
EXPECT_EQ(utf_cases[i].expected_success, success);
EXPECT_EQ(std::string(utf_cases[i].output), out_str);
}
if (utf_cases[i].input16) {
out_str.clear();
StdStringCanonOutput output(&out_str);
base::string16 input_str(WStringToUTF16(utf_cases[i].input16));
int input_len = static_cast<int>(input_str.length());
bool success = true;
for (int ch = 0; ch < input_len; ch++) {
success &= AppendUTF8EscapedChar(input_str.c_str(), &ch, input_len,
&output);
}
output.Complete();
EXPECT_EQ(utf_cases[i].expected_success, success);
EXPECT_EQ(std::string(utf_cases[i].output), out_str);
}
if (utf_cases[i].input8 && utf_cases[i].input16 &&
utf_cases[i].expected_success) {
// Check that the UTF-8 and UTF-16 inputs are equivalent.
// UTF-16 -> UTF-8
std::string input8_str(utf_cases[i].input8);
base::string16 input16_str(WStringToUTF16(utf_cases[i].input16));
EXPECT_EQ(input8_str, ConvertUTF16ToUTF8(input16_str));
// UTF-8 -> UTF-16
EXPECT_EQ(input16_str, ConvertUTF8ToUTF16(input8_str));
}
}
}
TEST(URLCanonTest, Scheme) {
// Here, we're mostly testing that unusual characters are handled properly.
// The canonicalizer doesn't do any parsing or whitespace detection. It will
// also do its best on error, and will escape funny sequences (these won't be
// valid schemes and it will return error).
//
// Note that the canonicalizer will append a colon to the output to separate
// out the rest of the URL, which is not present in the input. We check,
// however, that the output range includes everything but the colon.
ComponentCase scheme_cases[] = {
{"http", "http:", Component(0, 4), true},
{"HTTP", "http:", Component(0, 4), true},
{" HTTP ", "%20http%20:", Component(0, 10), false},
{"htt: ", "htt%3A%20:", Component(0, 9), false},
{"\xe4\xbd\xa0\xe5\xa5\xbdhttp", "%E4%BD%A0%E5%A5%BDhttp:", Component(0, 22), false},
// Don't re-escape something already escaped. Note that it will
// "canonicalize" the 'A' to 'a', but that's OK.
{"ht%3Atp", "ht%3atp:", Component(0, 7), false},
};
std::string out_str;
for (size_t i = 0; i < arraysize(scheme_cases); i++) {
int url_len = static_cast<int>(strlen(scheme_cases[i].input));
Component in_comp(0, url_len);
Component out_comp;
out_str.clear();
StdStringCanonOutput output1(&out_str);
bool success = CanonicalizeScheme(scheme_cases[i].input, in_comp, &output1,
&out_comp);
output1.Complete();
EXPECT_EQ(scheme_cases[i].expected_success, success);
EXPECT_EQ(std::string(scheme_cases[i].expected), out_str);
EXPECT_EQ(scheme_cases[i].expected_component.begin, out_comp.begin);
EXPECT_EQ(scheme_cases[i].expected_component.len, out_comp.len);
// Now try the wide version
out_str.clear();
StdStringCanonOutput output2(&out_str);
base::string16 wide_input(ConvertUTF8ToUTF16(scheme_cases[i].input));
in_comp.len = static_cast<int>(wide_input.length());
success = CanonicalizeScheme(wide_input.c_str(), in_comp, &output2,
&out_comp);
output2.Complete();
EXPECT_EQ(scheme_cases[i].expected_success, success);
EXPECT_EQ(std::string(scheme_cases[i].expected), out_str);
EXPECT_EQ(scheme_cases[i].expected_component.begin, out_comp.begin);
EXPECT_EQ(scheme_cases[i].expected_component.len, out_comp.len);
}
// Test the case where the scheme is declared nonexistant, it should be
// converted into an empty scheme.
Component out_comp;
out_str.clear();
StdStringCanonOutput output(&out_str);
EXPECT_TRUE(CanonicalizeScheme("", Component(0, -1), &output, &out_comp));
output.Complete();
EXPECT_EQ(std::string(":"), out_str);
EXPECT_EQ(0, out_comp.begin);
EXPECT_EQ(0, out_comp.len);
}
TEST(URLCanonTest, Host) {
IPAddressCase host_cases[] = {
// Basic canonicalization, uppercase should be converted to lowercase.
{"GoOgLe.CoM", L"GoOgLe.CoM", "google.com", Component(0, 10), CanonHostInfo::NEUTRAL, -1, ""},
// Spaces and some other characters should be escaped.
{"Goo%20 goo%7C|.com", L"Goo%20 goo%7C|.com", "goo%20%20goo%7C%7C.com", Component(0, 22), CanonHostInfo::NEUTRAL, -1, ""},
// Exciting different types of spaces!
{NULL, L"GOO\x00a0\x3000goo.com", "goo%20%20goo.com", Component(0, 16), CanonHostInfo::NEUTRAL, -1, ""},
// Other types of space (no-break, zero-width, zero-width-no-break) are
// name-prepped away to nothing.
{NULL, L"GOO\x200b\x2060\xfeffgoo.com", "googoo.com", Component(0, 10), CanonHostInfo::NEUTRAL, -1, ""},
// Ideographic full stop (full-width period for Chinese, etc.) should be
// treated as a dot.
{NULL, L"www.foo\x3002" L"bar.com", "www.foo.bar.com", Component(0, 15), CanonHostInfo::NEUTRAL, -1, ""},
// Invalid unicode characters should fail...
// ...In wide input, ICU will barf and we'll end up with the input as
// escaped UTF-8 (the invalid character should be replaced with the
// replacement character).
{"\xef\xb7\x90zyx.com", L"\xfdd0zyx.com", "%EF%BF%BDzyx.com", Component(0, 16), CanonHostInfo::BROKEN, -1, ""},
// ...This is the same as previous but with with escaped.
{"%ef%b7%90zyx.com", L"%ef%b7%90zyx.com", "%EF%BF%BDzyx.com", Component(0, 16), CanonHostInfo::BROKEN, -1, ""},
// Test name prepping, fullwidth input should be converted to ASCII and NOT
// IDN-ized. This is "Go" in fullwidth UTF-8/UTF-16.
{"\xef\xbc\xa7\xef\xbd\x8f.com", L"\xff27\xff4f.com", "go.com", Component(0, 6), CanonHostInfo::NEUTRAL, -1, ""},
// Test that fullwidth escaped values are properly name-prepped,
// then converted or rejected.
// ...%41 in fullwidth = 'A' (also as escaped UTF-8 input)
{"\xef\xbc\x85\xef\xbc\x94\xef\xbc\x91.com", L"\xff05\xff14\xff11.com", "a.com", Component(0, 5), CanonHostInfo::NEUTRAL, -1, ""},
{"%ef%bc%85%ef%bc%94%ef%bc%91.com", L"%ef%bc%85%ef%bc%94%ef%bc%91.com", "a.com", Component(0, 5), CanonHostInfo::NEUTRAL, -1, ""},
// ...%00 in fullwidth should fail (also as escaped UTF-8 input)
{"\xef\xbc\x85\xef\xbc\x90\xef\xbc\x90.com", L"\xff05\xff10\xff10.com", "%00.com", Component(0, 7), CanonHostInfo::BROKEN, -1, ""},
{"%ef%bc%85%ef%bc%90%ef%bc%90.com", L"%ef%bc%85%ef%bc%90%ef%bc%90.com", "%00.com", Component(0, 7), CanonHostInfo::BROKEN, -1, ""},
// Basic IDN support, UTF-8 and UTF-16 input should be converted to IDN
{"\xe4\xbd\xa0\xe5\xa5\xbd\xe4\xbd\xa0\xe5\xa5\xbd", L"\x4f60\x597d\x4f60\x597d", "xn--6qqa088eba", Component(0, 14), CanonHostInfo::NEUTRAL, -1, ""},
// See http://unicode.org/cldr/utility/idna.jsp for other
// examples/experiments and http://goo.gl/7yG11o
// for the full list of characters handled differently by
// IDNA 2003, UTS 46 (http://unicode.org/reports/tr46/ ) and IDNA 2008.
// 4 Deviation characters are mapped/ignored in UTS 46 transitional
// mechansm. UTS 46, table 4 row (g).
// Sharp-s is mapped to 'ss' in UTS 46 and IDNA 2003.
// Otherwise, it'd be "xn--fuball-cta.de".
{"fu\xc3\x9f" "ball.de", L"fu\x00df" L"ball.de", "fussball.de",
Component(0, 11), CanonHostInfo::NEUTRAL, -1, ""},
// Final-sigma (U+03C3) is mapped to regular sigma (U+03C2).
// Otherwise, it'd be "xn--wxaijb9b".
{"\xcf\x83\xcf\x8c\xce\xbb\xce\xbf\xcf\x82", L"\x3c3\x3cc\x3bb\x3bf\x3c2",
"xn--wxaikc6b", Component(0, 12),
CanonHostInfo::NEUTRAL, -1, ""},
// ZWNJ (U+200C) and ZWJ (U+200D) are mapped away in UTS 46 transitional
// handling as well as in IDNA 2003.
{"a\xe2\x80\x8c" "b\xe2\x80\x8d" "c", L"a\x200c" L"b\x200d" L"c", "abc",
Component(0, 3), CanonHostInfo::NEUTRAL, -1, ""},
// ZWJ between Devanagari characters is still mapped away in UTS 46
// transitional handling. IDNA 2008 would give xn--11bo0mv54g.
{"\xe0\xa4\x95\xe0\xa5\x8d\xe2\x80\x8d\xe0\xa4\x9c",
L"\x915\x94d\x200d\x91c", "xn--11bo0m",
Component(0, 10), CanonHostInfo::NEUTRAL, -1, ""},
// Fullwidth exclamation mark is disallowed. UTS 46, table 4, row (b)
// However, we do allow this at the moment because we don't use
// STD3 rules and canonicalize full-width ASCII to ASCII.
{"wow\xef\xbc\x81", L"wow\xff01", "wow%21",
Component(0, 6), CanonHostInfo::NEUTRAL, -1, ""},
// U+2132 (turned capital F) is disallowed. UTS 46, table 4, row (c)
// Allowed in IDNA 2003, but the mapping changed after Unicode 3.2
{"\xe2\x84\xb2oo", L"\x2132oo", "%E2%84%B2oo",
Component(0, 11), CanonHostInfo::BROKEN, -1, ""},
// U+2F868 (CJK Comp) is disallowed. UTS 46, table 4, row (d)
// Allowed in IDNA 2003, but the mapping changed after Unicode 3.2
{"\xf0\xaf\xa1\xa8\xe5\xa7\xbb.cn", L"\xd87e\xdc68\x59fb.cn",
"%F0%AF%A1%A8%E5%A7%BB.cn",
Component(0, 24), CanonHostInfo::BROKEN, -1, ""},
// Maps uppercase letters to lower case letters. UTS 46 table 4 row (e)
{"M\xc3\x9cNCHEN", L"M\xdcNCHEN", "xn--mnchen-3ya",
Component(0, 14), CanonHostInfo::NEUTRAL, -1, ""},
// Symbol/punctuations are allowed in IDNA 2003/UTS46.
// Not allowed in IDNA 2008. UTS 46 table 4 row (f).
{"\xe2\x99\xa5ny.us", L"\x2665ny.us", "xn--ny-s0x.us",
Component(0, 13), CanonHostInfo::NEUTRAL, -1, ""},
// U+11013 is new in Unicode 6.0 and is allowed. UTS 46 table 4, row (h)
// We used to allow it because we passed through unassigned code points.
{"\xf0\x91\x80\x93.com", L"\xd804\xdc13.com", "xn--n00d.com",
Component(0, 12), CanonHostInfo::NEUTRAL, -1, ""},
// U+0602 is disallowed in UTS46/IDNA 2008. UTS 46 table 4, row(i)
// Used to be allowed in INDA 2003.
{"\xd8\x82.eg", L"\x602.eg", "%D8%82.eg",
Component(0, 9), CanonHostInfo::BROKEN, -1, ""},
// U+20B7 is new in Unicode 5.2 (not a part of IDNA 2003 based
// on Unicode 3.2). We did allow it in the past because we let unassigned
// code point pass. We continue to allow it even though it's a
// "punctuation and symbol" blocked in IDNA 2008.
// UTS 46 table 4, row (j)
{"\xe2\x82\xb7.com", L"\x20b7.com", "xn--wzg.com",
Component(0, 11), CanonHostInfo::NEUTRAL, -1, ""},
// Maps uppercase letters to lower case letters.
// In IDNA 2003, it's allowed without case-folding
// ( xn--bc-7cb.com ) because it's not defined in Unicode 3.2
// (added in Unicode 4.1). UTS 46 table 4 row (k)
{"bc\xc8\xba.com", L"bc\x23a.com", "xn--bc-is1a.com",
Component(0, 15), CanonHostInfo::NEUTRAL, -1, ""},
// BiDi check test
// "Divehi" in Divehi (Thaana script) ends with BidiClass=NSM.
// Disallowed in IDNA 2003 but now allowed in UTS 46/IDNA 2008.
{"\xde\x8b\xde\xa8\xde\x88\xde\xac\xde\x80\xde\xa8",
L"\x78b\x7a8\x788\x7ac\x780\x7a8", "xn--hqbpi0jcw",
Component(0, 13), CanonHostInfo::NEUTRAL, -1, ""},
// Disallowed in both IDNA 2003 and 2008 with BiDi check.
// Labels starting with a RTL character cannot end with a LTR character.
{"\xd8\xac\xd8\xa7\xd8\xb1xyz", L"\x62c\x627\x631xyz",
"%D8%AC%D8%A7%D8%B1xyz", Component(0, 21),
CanonHostInfo::BROKEN, -1, ""},
// Labels starting with a RTL character can end with BC=EN (European
// number). Disallowed in IDNA 2003 but now allowed.
{"\xd8\xac\xd8\xa7\xd8\xb1" "2", L"\x62c\x627\x631" L"2",
"xn--2-ymcov", Component(0, 11),
CanonHostInfo::NEUTRAL, -1, ""},
// Labels starting with a RTL character cannot have "L" characters
// even if it ends with an BC=EN. Disallowed in both IDNA 2003/2008.
{"\xd8\xac\xd8\xa7\xd8\xb1xy2", L"\x62c\x627\x631xy2",
"%D8%AC%D8%A7%D8%B1xy2", Component(0, 21),
CanonHostInfo::BROKEN, -1, ""},
// Labels starting with a RTL character can end with BC=AN (Arabic number)
// Disallowed in IDNA 2003, but now allowed.
{"\xd8\xac\xd8\xa7\xd8\xb1\xd9\xa2", L"\x62c\x627\x631\x662",
"xn--mgbjq0r", Component(0, 11),
CanonHostInfo::NEUTRAL, -1, ""},
// Labels starting with a RTL character cannot have "L" characters
// even if it ends with an BC=AN (Arabic number).
// Disallowed in both IDNA 2003/2008.
{"\xd8\xac\xd8\xa7\xd8\xb1xy\xd9\xa2", L"\x62c\x627\x631xy\x662",
"%D8%AC%D8%A7%D8%B1xy%D9%A2", Component(0, 26),
CanonHostInfo::BROKEN, -1, ""},
// Labels starting with a RTL character cannot mix BC=EN and BC=AN
{"\xd8\xac\xd8\xa7\xd8\xb1xy2\xd9\xa2", L"\x62c\x627\x631xy2\x662",
"%D8%AC%D8%A7%D8%B1xy2%D9%A2", Component(0, 27),
CanonHostInfo::BROKEN, -1, ""},
// As of Unicode 6.2, U+20CF is not assigned. We do not allow it.
{"\xe2\x83\x8f.com", L"\x20cf.com", "%E2%83%8F.com",
Component(0, 13), CanonHostInfo::BROKEN, -1, ""},
// U+0080 is not allowed.
{"\xc2\x80.com", L"\x80.com", "%C2%80.com",
Component(0, 10), CanonHostInfo::BROKEN, -1, ""},
// Mixed UTF-8 and escaped UTF-8 (narrow case) and UTF-16 and escaped
// Mixed UTF-8 and escaped UTF-8 (narrow case) and UTF-16 and escaped
// UTF-8 (wide case). The output should be equivalent to the true wide
// character input above).
{"%E4%BD%A0%E5%A5%BD\xe4\xbd\xa0\xe5\xa5\xbd",
L"%E4%BD%A0%E5%A5%BD\x4f60\x597d", "xn--6qqa088eba",
Component(0, 14), CanonHostInfo::NEUTRAL, -1, ""},
// Invalid escaped characters should fail and the percents should be
// escaped.
{"%zz%66%a", L"%zz%66%a", "%25zzf%25a", Component(0, 10),
CanonHostInfo::BROKEN, -1, ""},
// If we get an invalid character that has been escaped.
{"%25", L"%25", "%25", Component(0, 3),
CanonHostInfo::BROKEN, -1, ""},
{"hello%00", L"hello%00", "hello%00", Component(0, 8),
CanonHostInfo::BROKEN, -1, ""},
// Escaped numbers should be treated like IP addresses if they are.
{"%30%78%63%30%2e%30%32%35%30.01", L"%30%78%63%30%2e%30%32%35%30.01",
"192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 3,
"C0A80001"},
{"%30%78%63%30%2e%30%32%35%30.01%2e", L"%30%78%63%30%2e%30%32%35%30.01%2e",
"192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 3,
"C0A80001"},
// Invalid escaping should trigger the regular host error handling.
{"%3g%78%63%30%2e%30%32%35%30%2E.01", L"%3g%78%63%30%2e%30%32%35%30%2E.01", "%253gxc0.0250..01", Component(0, 17), CanonHostInfo::BROKEN, -1, ""},
// Something that isn't exactly an IP should get treated as a host and
// spaces escaped.
{"192.168.0.1 hello", L"192.168.0.1 hello", "192.168.0.1%20hello", Component(0, 19), CanonHostInfo::NEUTRAL, -1, ""},
// Fullwidth and escaped UTF-8 fullwidth should still be treated as IP.
// These are "0Xc0.0250.01" in fullwidth.
{"\xef\xbc\x90%Ef%bc\xb8%ef%Bd%83\xef\xbc\x90%EF%BC%8E\xef\xbc\x90\xef\xbc\x92\xef\xbc\x95\xef\xbc\x90\xef\xbc%8E\xef\xbc\x90\xef\xbc\x91", L"\xff10\xff38\xff43\xff10\xff0e\xff10\xff12\xff15\xff10\xff0e\xff10\xff11", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 3, "C0A80001"},
// Broken IP addresses get marked as such.
{"192.168.0.257", L"192.168.0.257", "192.168.0.257", Component(0, 13), CanonHostInfo::BROKEN, -1, ""},
{"[google.com]", L"[google.com]", "[google.com]", Component(0, 12), CanonHostInfo::BROKEN, -1, ""},
// Cyrillic letter followed by '(' should return punycode for '(' escaped
// before punycode string was created. I.e.
// if '(' is escaped after punycode is created we would get xn--%28-8tb
// (incorrect).
{"\xd1\x82(", L"\x0442(", "xn--%28-7ed", Component(0, 11),
CanonHostInfo::NEUTRAL, -1, ""},
// Address with all hexidecimal characters with leading number of 1<<32
// or greater and should return NEUTRAL rather than BROKEN if not all
// components are numbers.
{"12345678912345.de", L"12345678912345.de", "12345678912345.de", Component(0, 17), CanonHostInfo::NEUTRAL, -1, ""},
{"1.12345678912345.de", L"1.12345678912345.de", "1.12345678912345.de", Component(0, 19), CanonHostInfo::NEUTRAL, -1, ""},
{"12345678912345.12345678912345.de", L"12345678912345.12345678912345.de", "12345678912345.12345678912345.de", Component(0, 32), CanonHostInfo::NEUTRAL, -1, ""},
{"1.2.0xB3A73CE5B59.de", L"1.2.0xB3A73CE5B59.de", "1.2.0xb3a73ce5b59.de", Component(0, 20), CanonHostInfo::NEUTRAL, -1, ""},
{"12345678912345.0xde", L"12345678912345.0xde", "12345678912345.0xde", Component(0, 19), CanonHostInfo::BROKEN, -1, ""},
};
// CanonicalizeHost() non-verbose.
std::string out_str;
for (size_t i = 0; i < arraysize(host_cases); i++) {
// Narrow version.
if (host_cases[i].input8) {
int host_len = static_cast<int>(strlen(host_cases[i].input8));
Component in_comp(0, host_len);
Component out_comp;
out_str.clear();
StdStringCanonOutput output(&out_str);
bool success = CanonicalizeHost(host_cases[i].input8, in_comp, &output,
&out_comp);
output.Complete();
EXPECT_EQ(host_cases[i].expected_family != CanonHostInfo::BROKEN,
success) << "for input: " << host_cases[i].input8;
EXPECT_EQ(std::string(host_cases[i].expected), out_str) <<
"for input: " << host_cases[i].input8;
EXPECT_EQ(host_cases[i].expected_component.begin, out_comp.begin) <<
"for input: " << host_cases[i].input8;
EXPECT_EQ(host_cases[i].expected_component.len, out_comp.len) <<
"for input: " << host_cases[i].input8;
}
// Wide version.
if (host_cases[i].input16) {
base::string16 input16(WStringToUTF16(host_cases[i].input16));
int host_len = static_cast<int>(input16.length());
Component in_comp(0, host_len);
Component out_comp;
out_str.clear();
StdStringCanonOutput output(&out_str);
bool success = CanonicalizeHost(input16.c_str(), in_comp, &output,
&out_comp);
output.Complete();
EXPECT_EQ(host_cases[i].expected_family != CanonHostInfo::BROKEN,
success);
EXPECT_EQ(std::string(host_cases[i].expected), out_str);
EXPECT_EQ(host_cases[i].expected_component.begin, out_comp.begin);
EXPECT_EQ(host_cases[i].expected_component.len, out_comp.len);
}
}
// CanonicalizeHostVerbose()
for (size_t i = 0; i < arraysize(host_cases); i++) {
// Narrow version.
if (host_cases[i].input8) {
int host_len = static_cast<int>(strlen(host_cases[i].input8));
Component in_comp(0, host_len);
out_str.clear();
StdStringCanonOutput output(&out_str);
CanonHostInfo host_info;
CanonicalizeHostVerbose(host_cases[i].input8, in_comp, &output,
&host_info);
output.Complete();
EXPECT_EQ(host_cases[i].expected_family, host_info.family);
EXPECT_EQ(std::string(host_cases[i].expected), out_str);
EXPECT_EQ(host_cases[i].expected_component.begin,
host_info.out_host.begin);
EXPECT_EQ(host_cases[i].expected_component.len, host_info.out_host.len);
EXPECT_EQ(std::string(host_cases[i].expected_address_hex),
BytesToHexString(host_info.address, host_info.AddressLength()));
if (host_cases[i].expected_family == CanonHostInfo::IPV4) {
EXPECT_EQ(host_cases[i].expected_num_ipv4_components,
host_info.num_ipv4_components);
}
}
// Wide version.
if (host_cases[i].input16) {
base::string16 input16(WStringToUTF16(host_cases[i].input16));
int host_len = static_cast<int>(input16.length());
Component in_comp(0, host_len);
out_str.clear();
StdStringCanonOutput output(&out_str);
CanonHostInfo host_info;
CanonicalizeHostVerbose(input16.c_str(), in_comp, &output, &host_info);
output.Complete();
EXPECT_EQ(host_cases[i].expected_family, host_info.family);
EXPECT_EQ(std::string(host_cases[i].expected), out_str);
EXPECT_EQ(host_cases[i].expected_component.begin,
host_info.out_host.begin);
EXPECT_EQ(host_cases[i].expected_component.len, host_info.out_host.len);
EXPECT_EQ(std::string(host_cases[i].expected_address_hex),
BytesToHexString(host_info.address, host_info.AddressLength()));
if (host_cases[i].expected_family == CanonHostInfo::IPV4) {
EXPECT_EQ(host_cases[i].expected_num_ipv4_components,
host_info.num_ipv4_components);
}
}
}
}
TEST(URLCanonTest, IPv4) {
IPAddressCase cases[] = {
// Empty is not an IP address.
{"", L"", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
{".", L".", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
// Regular IP addresses in different bases.
{"192.168.0.1", L"192.168.0.1", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 4, "C0A80001"},
{"0300.0250.00.01", L"0300.0250.00.01", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 4, "C0A80001"},
{"0xC0.0Xa8.0x0.0x1", L"0xC0.0Xa8.0x0.0x1", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 4, "C0A80001"},
// Non-IP addresses due to invalid characters.
{"192.168.9.com", L"192.168.9.com", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
// Invalid characters for the base should be rejected.
{"19a.168.0.1", L"19a.168.0.1", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
{"0308.0250.00.01", L"0308.0250.00.01", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
{"0xCG.0xA8.0x0.0x1", L"0xCG.0xA8.0x0.0x1", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
// If there are not enough components, the last one should fill them out.
{"192", L"192", "0.0.0.192", Component(0, 9), CanonHostInfo::IPV4, 1, "000000C0"},
{"0xC0a80001", L"0xC0a80001", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 1, "C0A80001"},
{"030052000001", L"030052000001", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 1, "C0A80001"},
{"000030052000001", L"000030052000001", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 1, "C0A80001"},
{"192.168", L"192.168", "192.0.0.168", Component(0, 11), CanonHostInfo::IPV4, 2, "C00000A8"},
{"192.0x00A80001", L"192.0x000A80001", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 2, "C0A80001"},
{"0xc0.052000001", L"0xc0.052000001", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 2, "C0A80001"},
{"192.168.1", L"192.168.1", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 3, "C0A80001"},
// Too many components means not an IP address.
{"192.168.0.0.1", L"192.168.0.0.1", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
// We allow a single trailing dot.
{"192.168.0.1.", L"192.168.0.1.", "192.168.0.1", Component(0, 11), CanonHostInfo::IPV4, 4, "C0A80001"},
{"192.168.0.1. hello", L"192.168.0.1. hello", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
{"192.168.0.1..", L"192.168.0.1..", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
// Two dots in a row means not an IP address.
{"192.168..1", L"192.168..1", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
// Any numerical overflow should be marked as BROKEN.
{"0x100.0", L"0x100.0", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"0x100.0.0", L"0x100.0.0", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"0x100.0.0.0", L"0x100.0.0.0", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"0.0x100.0.0", L"0.0x100.0.0", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"0.0.0x100.0", L"0.0.0x100.0", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"0.0.0.0x100", L"0.0.0.0x100", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"0.0.0x10000", L"0.0.0x10000", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"0.0x1000000", L"0.0x1000000", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"0x100000000", L"0x100000000", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Repeat the previous tests, minus 1, to verify boundaries.
{"0xFF.0", L"0xFF.0", "255.0.0.0", Component(0, 9), CanonHostInfo::IPV4, 2, "FF000000"},
{"0xFF.0.0", L"0xFF.0.0", "255.0.0.0", Component(0, 9), CanonHostInfo::IPV4, 3, "FF000000"},
{"0xFF.0.0.0", L"0xFF.0.0.0", "255.0.0.0", Component(0, 9), CanonHostInfo::IPV4, 4, "FF000000"},
{"0.0xFF.0.0", L"0.0xFF.0.0", "0.255.0.0", Component(0, 9), CanonHostInfo::IPV4, 4, "00FF0000"},
{"0.0.0xFF.0", L"0.0.0xFF.0", "0.0.255.0", Component(0, 9), CanonHostInfo::IPV4, 4, "0000FF00"},
{"0.0.0.0xFF", L"0.0.0.0xFF", "0.0.0.255", Component(0, 9), CanonHostInfo::IPV4, 4, "000000FF"},
{"0.0.0xFFFF", L"0.0.0xFFFF", "0.0.255.255", Component(0, 11), CanonHostInfo::IPV4, 3, "0000FFFF"},
{"0.0xFFFFFF", L"0.0xFFFFFF", "0.255.255.255", Component(0, 13), CanonHostInfo::IPV4, 2, "00FFFFFF"},
{"0xFFFFFFFF", L"0xFFFFFFFF", "255.255.255.255", Component(0, 15), CanonHostInfo::IPV4, 1, "FFFFFFFF"},
// Old trunctations tests. They're all "BROKEN" now.
{"276.256.0xf1a2.077777", L"276.256.0xf1a2.077777", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"192.168.0.257", L"192.168.0.257", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"192.168.0xa20001", L"192.168.0xa20001", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"192.015052000001", L"192.015052000001", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"0X12C0a80001", L"0X12C0a80001", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"276.1.2", L"276.1.2", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Spaces should be rejected.
{"192.168.0.1 hello", L"192.168.0.1 hello", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
// Very large numbers.
{"0000000000000300.0x00000000000000fF.00000000000000001", L"0000000000000300.0x00000000000000fF.00000000000000001", "192.255.0.1", Component(0, 11), CanonHostInfo::IPV4, 3, "C0FF0001"},
{"0000000000000300.0xffffffffFFFFFFFF.3022415481470977", L"0000000000000300.0xffffffffFFFFFFFF.3022415481470977", "", Component(0, 11), CanonHostInfo::BROKEN, -1, ""},
// A number has no length limit, but long numbers can still overflow.
{"00000000000000000001", L"00000000000000000001", "0.0.0.1", Component(0, 7), CanonHostInfo::IPV4, 1, "00000001"},
{"0000000000000000100000000000000001", L"0000000000000000100000000000000001", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// If a long component is non-numeric, it's a hostname, *not* a broken IP.
{"0.0.0.000000000000000000z", L"0.0.0.000000000000000000z", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
{"0.0.0.100000000000000000z", L"0.0.0.100000000000000000z", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
// Truncation of all zeros should still result in 0.
{"0.00.0x.0x0", L"0.00.0x.0x0", "0.0.0.0", Component(0, 7), CanonHostInfo::IPV4, 4, "00000000"},
};
for (size_t i = 0; i < arraysize(cases); i++) {
// 8-bit version.
Component component(0, static_cast<int>(strlen(cases[i].input8)));
std::string out_str1;
StdStringCanonOutput output1(&out_str1);
CanonHostInfo host_info;
CanonicalizeIPAddress(cases[i].input8, component, &output1, &host_info);
output1.Complete();
EXPECT_EQ(cases[i].expected_family, host_info.family);
EXPECT_EQ(std::string(cases[i].expected_address_hex),
BytesToHexString(host_info.address, host_info.AddressLength()));
if (host_info.family == CanonHostInfo::IPV4) {
EXPECT_STREQ(cases[i].expected, out_str1.c_str());
EXPECT_EQ(cases[i].expected_component.begin, host_info.out_host.begin);
EXPECT_EQ(cases[i].expected_component.len, host_info.out_host.len);
EXPECT_EQ(cases[i].expected_num_ipv4_components,
host_info.num_ipv4_components);
}
// 16-bit version.
base::string16 input16(WStringToUTF16(cases[i].input16));
component = Component(0, static_cast<int>(input16.length()));
std::string out_str2;
StdStringCanonOutput output2(&out_str2);
CanonicalizeIPAddress(input16.c_str(), component, &output2, &host_info);
output2.Complete();
EXPECT_EQ(cases[i].expected_family, host_info.family);
EXPECT_EQ(std::string(cases[i].expected_address_hex),
BytesToHexString(host_info.address, host_info.AddressLength()));
if (host_info.family == CanonHostInfo::IPV4) {
EXPECT_STREQ(cases[i].expected, out_str2.c_str());
EXPECT_EQ(cases[i].expected_component.begin, host_info.out_host.begin);
EXPECT_EQ(cases[i].expected_component.len, host_info.out_host.len);
EXPECT_EQ(cases[i].expected_num_ipv4_components,
host_info.num_ipv4_components);
}
}
}
TEST(URLCanonTest, IPv6) {
IPAddressCase cases[] = {
// Empty is not an IP address.
{"", L"", "", Component(), CanonHostInfo::NEUTRAL, -1, ""},
// Non-IPs with [:] characters are marked BROKEN.
{":", L":", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[", L"[", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[:", L"[:", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"]", L"]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{":]", L":]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[]", L"[]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[:]", L"[:]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Regular IP address is invalid without bounding '[' and ']'.
{"2001:db8::1", L"2001:db8::1", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[2001:db8::1", L"[2001:db8::1", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"2001:db8::1]", L"2001:db8::1]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Regular IP addresses.
{"[::]", L"[::]", "[::]", Component(0,4), CanonHostInfo::IPV6, -1, "00000000000000000000000000000000"},
{"[::1]", L"[::1]", "[::1]", Component(0,5), CanonHostInfo::IPV6, -1, "00000000000000000000000000000001"},
{"[1::]", L"[1::]", "[1::]", Component(0,5), CanonHostInfo::IPV6, -1, "00010000000000000000000000000000"},
// Leading zeros should be stripped.
{"[000:01:02:003:004:5:6:007]", L"[000:01:02:003:004:5:6:007]", "[0:1:2:3:4:5:6:7]", Component(0,17), CanonHostInfo::IPV6, -1, "00000001000200030004000500060007"},
// Upper case letters should be lowercased.
{"[A:b:c:DE:fF:0:1:aC]", L"[A:b:c:DE:fF:0:1:aC]", "[a:b:c:de:ff:0:1:ac]", Component(0,20), CanonHostInfo::IPV6, -1, "000A000B000C00DE00FF0000000100AC"},
// The same address can be written with different contractions, but should
// get canonicalized to the same thing.
{"[1:0:0:2::3:0]", L"[1:0:0:2::3:0]", "[1::2:0:0:3:0]", Component(0,14), CanonHostInfo::IPV6, -1, "00010000000000020000000000030000"},
{"[1::2:0:0:3:0]", L"[1::2:0:0:3:0]", "[1::2:0:0:3:0]", Component(0,14), CanonHostInfo::IPV6, -1, "00010000000000020000000000030000"},
// Addresses with embedded IPv4.
{"[::192.168.0.1]", L"[::192.168.0.1]", "[::c0a8:1]", Component(0,10), CanonHostInfo::IPV6, -1, "000000000000000000000000C0A80001"},
{"[::ffff:192.168.0.1]", L"[::ffff:192.168.0.1]", "[::ffff:c0a8:1]", Component(0,15), CanonHostInfo::IPV6, -1, "00000000000000000000FFFFC0A80001"},
{"[::eeee:192.168.0.1]", L"[::eeee:192.168.0.1]", "[::eeee:c0a8:1]", Component(0, 15), CanonHostInfo::IPV6, -1, "00000000000000000000EEEEC0A80001"},
{"[2001::192.168.0.1]", L"[2001::192.168.0.1]", "[2001::c0a8:1]", Component(0, 14), CanonHostInfo::IPV6, -1, "200100000000000000000000C0A80001"},
{"[1:2:192.168.0.1:5:6]", L"[1:2:192.168.0.1:5:6]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// IPv4 with last component missing.
{"[::ffff:192.1.2]", L"[::ffff:192.1.2]", "[::ffff:c001:2]", Component(0,15), CanonHostInfo::IPV6, -1, "00000000000000000000FFFFC0010002"},
// IPv4 using hex.
// TODO(eroman): Should this format be disallowed?
{"[::ffff:0xC0.0Xa8.0x0.0x1]", L"[::ffff:0xC0.0Xa8.0x0.0x1]", "[::ffff:c0a8:1]", Component(0,15), CanonHostInfo::IPV6, -1, "00000000000000000000FFFFC0A80001"},
// There may be zeros surrounding the "::" contraction.
{"[0:0::0:0:8]", L"[0:0::0:0:8]", "[::8]", Component(0,5), CanonHostInfo::IPV6, -1, "00000000000000000000000000000008"},
{"[2001:db8::1]", L"[2001:db8::1]", "[2001:db8::1]", Component(0,13), CanonHostInfo::IPV6, -1, "20010DB8000000000000000000000001"},
// Can only have one "::" contraction in an IPv6 string literal.
{"[2001::db8::1]", L"[2001::db8::1]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// No more than 2 consecutive ':'s.
{"[2001:db8:::1]", L"[2001:db8:::1]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[:::]", L"[:::]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Non-IP addresses due to invalid characters.
{"[2001::.com]", L"[2001::.com]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// If there are not enough components, the last one should fill them out.
// ... omitted at this time ...
// Too many components means not an IP address. Similarly with too few if using IPv4 compat or mapped addresses.
{"[::192.168.0.0.1]", L"[::192.168.0.0.1]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[::ffff:192.168.0.0.1]", L"[::ffff:192.168.0.0.1]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[1:2:3:4:5:6:7:8:9]", L"[1:2:3:4:5:6:7:8:9]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Too many bits (even though 8 comonents, the last one holds 32 bits).
{"[0:0:0:0:0:0:0:192.168.0.1]", L"[0:0:0:0:0:0:0:192.168.0.1]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Too many bits specified -- the contraction would have to be zero-length
// to not exceed 128 bits.
{"[1:2:3:4:5:6::192.168.0.1]", L"[1:2:3:4:5:6::192.168.0.1]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// The contraction is for 16 bits of zero.
{"[1:2:3:4:5:6::8]", L"[1:2:3:4:5:6::8]", "[1:2:3:4:5:6:0:8]", Component(0,17), CanonHostInfo::IPV6, -1, "00010002000300040005000600000008"},
// Cannot have a trailing colon.
{"[1:2:3:4:5:6:7:8:]", L"[1:2:3:4:5:6:7:8:]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[1:2:3:4:5:6:192.168.0.1:]", L"[1:2:3:4:5:6:192.168.0.1:]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Cannot have negative numbers.
{"[-1:2:3:4:5:6:7:8]", L"[-1:2:3:4:5:6:7:8]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Scope ID -- the URL may contain an optional ["%" <scope_id>] section.
// The scope_id should be included in the canonicalized URL, and is an
// unsigned decimal number.
// Invalid because no ID was given after the percent.
// Don't allow scope-id
{"[1::%1]", L"[1::%1]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[1::%eth0]", L"[1::%eth0]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[1::%]", L"[1::%]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[%]", L"[%]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[::%:]", L"[::%:]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Don't allow leading or trailing colons.
{"[:0:0::0:0:8]", L"[:0:0::0:0:8]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[0:0::0:0:8:]", L"[0:0::0:0:8:]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
{"[:0:0::0:0:8:]", L"[:0:0::0:0:8:]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// We allow a single trailing dot.
// ... omitted at this time ...
// Two dots in a row means not an IP address.
{"[::192.168..1]", L"[::192.168..1]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
// Any non-first components get truncated to one byte.
// ... omitted at this time ...
// Spaces should be rejected.
{"[::1 hello]", L"[::1 hello]", "", Component(), CanonHostInfo::BROKEN, -1, ""},
};
for (size_t i = 0; i < arraysize(cases); i++) {
// 8-bit version.
Component component(0, static_cast<int>(strlen(cases[i].input8)));
std::string out_str1;
StdStringCanonOutput output1(&out_str1);
CanonHostInfo host_info;
CanonicalizeIPAddress(cases[i].input8, component, &output1, &host_info);
output1.Complete();
EXPECT_EQ(cases[i].expected_family, host_info.family);
EXPECT_EQ(std::string(cases[i].expected_address_hex),
BytesToHexString(host_info.address, host_info.AddressLength())) << "iter " << i << " host " << cases[i].input8;
if (host_info.family == CanonHostInfo::IPV6) {
EXPECT_STREQ(cases[i].expected, out_str1.c_str());
EXPECT_EQ(cases[i].expected_component.begin,
host_info.out_host.begin);
EXPECT_EQ(cases[i].expected_component.len, host_info.out_host.len);
}
// 16-bit version.
base::string16 input16(WStringToUTF16(cases[i].input16));
component = Component(0, static_cast<int>(input16.length()));
std::string out_str2;
StdStringCanonOutput output2(&out_str2);
CanonicalizeIPAddress(input16.c_str(), component, &output2, &host_info);
output2.Complete();
EXPECT_EQ(cases[i].expected_family, host_info.family);
EXPECT_EQ(std::string(cases[i].expected_address_hex),
BytesToHexString(host_info.address, host_info.AddressLength()));
if (host_info.family == CanonHostInfo::IPV6) {
EXPECT_STREQ(cases[i].expected, out_str2.c_str());
EXPECT_EQ(cases[i].expected_component.begin, host_info.out_host.begin);
EXPECT_EQ(cases[i].expected_component.len, host_info.out_host.len);
}
}
}
TEST(URLCanonTest, IPEmpty) {
std::string out_str1;
StdStringCanonOutput output1(&out_str1);
CanonHostInfo host_info;
// This tests tests.
const char spec[] = "192.168.0.1";
CanonicalizeIPAddress(spec, Component(), &output1, &host_info);
EXPECT_FALSE(host_info.IsIPAddress());
CanonicalizeIPAddress(spec, Component(0, 0), &output1, &host_info);
EXPECT_FALSE(host_info.IsIPAddress());
}
TEST(URLCanonTest, UserInfo) {
// Note that the canonicalizer should escape and treat empty components as
// not being there.
// We actually parse a full input URL so we can get the initial components.
struct UserComponentCase {
const char* input;
const char* expected;
Component expected_username;
Component expected_password;
bool expected_success;
} user_info_cases[] = {
{"http://user:pass@host.com/", "user:pass@", Component(0, 4), Component(5, 4), true},
{"http://@host.com/", "", Component(0, -1), Component(0, -1), true},
{"http://:@host.com/", "", Component(0, -1), Component(0, -1), true},
{"http://foo:@host.com/", "foo@", Component(0, 3), Component(0, -1), true},
{"http://:foo@host.com/", ":foo@", Component(0, 0), Component(1, 3), true},
{"http://^ :$\t@host.com/", "%5E%20:$%09@", Component(0, 6), Component(7, 4), true},
{"http://user:pass@/", "user:pass@", Component(0, 4), Component(5, 4), true},
{"http://%2540:bar@domain.com/", "%2540:bar@", Component(0, 5), Component(6, 3), true },
// IE7 compatability: old versions allowed backslashes in usernames, but
// IE7 does not. We disallow it as well.
{"ftp://me\\mydomain:pass@foo.com/", "", Component(0, -1), Component(0, -1), true},
};
for (size_t i = 0; i < ARRAYSIZE(user_info_cases); i++) {
int url_len = static_cast<int>(strlen(user_info_cases[i].input));
Parsed parsed;
ParseStandardURL(user_info_cases[i].input, url_len, &parsed);
Component out_user, out_pass;
std::string out_str;
StdStringCanonOutput output1(&out_str);
bool success = CanonicalizeUserInfo(user_info_cases[i].input,
parsed.username,
user_info_cases[i].input,
parsed.password,
&output1,
&out_user,
&out_pass);
output1.Complete();
EXPECT_EQ(user_info_cases[i].expected_success, success);
EXPECT_EQ(std::string(user_info_cases[i].expected), out_str);
EXPECT_EQ(user_info_cases[i].expected_username.begin, out_user.begin);
EXPECT_EQ(user_info_cases[i].expected_username.len, out_user.len);
EXPECT_EQ(user_info_cases[i].expected_password.begin, out_pass.begin);
EXPECT_EQ(user_info_cases[i].expected_password.len, out_pass.len);
// Now try the wide version
out_str.clear();
StdStringCanonOutput output2(&out_str);
base::string16 wide_input(ConvertUTF8ToUTF16(user_info_cases[i].input));
success = CanonicalizeUserInfo(wide_input.c_str(),
parsed.username,
wide_input.c_str(),
parsed.password,
&output2,
&out_user,
&out_pass);
output2.Complete();
EXPECT_EQ(user_info_cases[i].expected_success, success);
EXPECT_EQ(std::string(user_info_cases[i].expected), out_str);
EXPECT_EQ(user_info_cases[i].expected_username.begin, out_user.begin);
EXPECT_EQ(user_info_cases[i].expected_username.len, out_user.len);
EXPECT_EQ(user_info_cases[i].expected_password.begin, out_pass.begin);
EXPECT_EQ(user_info_cases[i].expected_password.len, out_pass.len);
}
}
TEST(URLCanonTest, Port) {
// We only need to test that the number gets properly put into the output
// buffer. The parser unit tests will test scanning the number correctly.
//
// Note that the CanonicalizePort will always prepend a colon to the output
// to separate it from the colon that it assumes preceeds it.
struct PortCase {
const char* input;
int default_port;
const char* expected;
Component expected_component;
bool expected_success;
} port_cases[] = {
// Invalid input should be copied w/ failure.
{"as df", 80, ":as%20df", Component(1, 7), false},
{"-2", 80, ":-2", Component(1, 2), false},
// Default port should be omitted.
{"80", 80, "", Component(0, -1), true},
{"8080", 80, ":8080", Component(1, 4), true},
// PORT_UNSPECIFIED should mean always keep the port.
{"80", PORT_UNSPECIFIED, ":80", Component(1, 2), true},
};
for (size_t i = 0; i < ARRAYSIZE(port_cases); i++) {
int url_len = static_cast<int>(strlen(port_cases[i].input));
Component in_comp(0, url_len);
Component out_comp;
std::string out_str;
StdStringCanonOutput output1(&out_str);
bool success = CanonicalizePort(port_cases[i].input,
in_comp,
port_cases[i].default_port,
&output1,
&out_comp);
output1.Complete();
EXPECT_EQ(port_cases[i].expected_success, success);
EXPECT_EQ(std::string(port_cases[i].expected), out_str);
EXPECT_EQ(port_cases[i].expected_component.begin, out_comp.begin);
EXPECT_EQ(port_cases[i].expected_component.len, out_comp.len);
// Now try the wide version
out_str.clear();
StdStringCanonOutput output2(&out_str);
base::string16 wide_input(ConvertUTF8ToUTF16(port_cases[i].input));
success = CanonicalizePort(wide_input.c_str(),
in_comp,
port_cases[i].default_port,
&output2,
&out_comp);
output2.Complete();
EXPECT_EQ(port_cases[i].expected_success, success);
EXPECT_EQ(std::string(port_cases[i].expected), out_str);
EXPECT_EQ(port_cases[i].expected_component.begin, out_comp.begin);
EXPECT_EQ(port_cases[i].expected_component.len, out_comp.len);
}
}
TEST(URLCanonTest, Path) {
DualComponentCase path_cases[] = {
// ----- path collapsing tests -----
{"/././foo", L"/././foo", "/foo", Component(0, 4), true},
{"/./.foo", L"/./.foo", "/.foo", Component(0, 5), true},
{"/foo/.", L"/foo/.", "/foo/", Component(0, 5), true},
{"/foo/./", L"/foo/./", "/foo/", Component(0, 5), true},
// double dots followed by a slash or the end of the string count
{"/foo/bar/..", L"/foo/bar/..", "/foo/", Component(0, 5), true},
{"/foo/bar/../", L"/foo/bar/../", "/foo/", Component(0, 5), true},
// don't count double dots when they aren't followed by a slash
{"/foo/..bar", L"/foo/..bar", "/foo/..bar", Component(0, 10), true},
// some in the middle
{"/foo/bar/../ton", L"/foo/bar/../ton", "/foo/ton", Component(0, 8), true},
{"/foo/bar/../ton/../../a", L"/foo/bar/../ton/../../a", "/a", Component(0, 2), true},
// we should not be able to go above the root
{"/foo/../../..", L"/foo/../../..", "/", Component(0, 1), true},
{"/foo/../../../ton", L"/foo/../../../ton", "/ton", Component(0, 4), true},
// escaped dots should be unescaped and treated the same as dots
{"/foo/%2e", L"/foo/%2e", "/foo/", Component(0, 5), true},
{"/foo/%2e%2", L"/foo/%2e%2", "/foo/.%2", Component(0, 8), true},
{"/foo/%2e./%2e%2e/.%2e/%2e.bar", L"/foo/%2e./%2e%2e/.%2e/%2e.bar", "/..bar", Component(0, 6), true},
// Multiple slashes in a row should be preserved and treated like empty
// directory names.
{"////../..", L"////../..", "//", Component(0, 2), true},
// ----- escaping tests -----
{"/foo", L"/foo", "/foo", Component(0, 4), true},
// Valid escape sequence
{"/%20foo", L"/%20foo", "/%20foo", Component(0, 7), true},
// Invalid escape sequence we should pass through unchanged.
{"/foo%", L"/foo%", "/foo%", Component(0, 5), true},
{"/foo%2", L"/foo%2", "/foo%2", Component(0, 6), true},
// Invalid escape sequence: bad characters should be treated the same as
// the sourrounding text, not as escaped (in this case, UTF-8).
{"/foo%2zbar", L"/foo%2zbar", "/foo%2zbar", Component(0, 10), true},
{"/foo%2\xc2\xa9zbar", NULL, "/foo%2%C2%A9zbar", Component(0, 16), true},
{NULL, L"/foo%2\xc2\xa9zbar", "/foo%2%C3%82%C2%A9zbar", Component(0, 22), true},
// Regular characters that are escaped should be unescaped
{"/foo%41%7a", L"/foo%41%7a", "/fooAz", Component(0, 6), true},
// Funny characters that are unescaped should be escaped
{"/foo\x09\x91%91", NULL, "/foo%09%91%91", Component(0, 13), true},
{NULL, L"/foo\x09\x91%91", "/foo%09%C2%91%91", Component(0, 16), true},
// Invalid characters that are escaped should cause a failure.
{"/foo%00%51", L"/foo%00%51", "/foo%00Q", Component(0, 8), false},
// Some characters should be passed through unchanged regardless of esc.
{"/(%28:%3A%29)", L"/(%28:%3A%29)", "/(%28:%3A%29)", Component(0, 13), true},
// Characters that are properly escaped should not have the case changed
// of hex letters.
{"/%3A%3a%3C%3c", L"/%3A%3a%3C%3c", "/%3A%3a%3C%3c", Component(0, 13), true},
// Funny characters that are unescaped should be escaped
{"/foo\tbar", L"/foo\tbar", "/foo%09bar", Component(0, 10), true},
// Backslashes should get converted to forward slashes
{"\\foo\\bar", L"\\foo\\bar", "/foo/bar", Component(0, 8), true},
// Hashes found in paths (possibly only when the caller explicitly sets
// the path on an already-parsed URL) should be escaped.
{"/foo#bar", L"/foo#bar", "/foo%23bar", Component(0, 10), true},
// %7f should be allowed and %3D should not be unescaped (these were wrong
// in a previous version).
{"/%7Ffp3%3Eju%3Dduvgw%3Dd", L"/%7Ffp3%3Eju%3Dduvgw%3Dd", "/%7Ffp3%3Eju%3Dduvgw%3Dd", Component(0, 24), true},
// @ should be passed through unchanged (escaped or unescaped).
{"/@asdf%40", L"/@asdf%40", "/@asdf%40", Component(0, 9), true},
// ----- encoding tests -----
// Basic conversions
{"/\xe4\xbd\xa0\xe5\xa5\xbd\xe4\xbd\xa0\xe5\xa5\xbd", L"/\x4f60\x597d\x4f60\x597d", "/%E4%BD%A0%E5%A5%BD%E4%BD%A0%E5%A5%BD", Component(0, 37), true},
// Invalid unicode characters should fail. We only do validation on
// UTF-16 input, so this doesn't happen on 8-bit.
{"/\xef\xb7\x90zyx", NULL, "/%EF%B7%90zyx", Component(0, 13), true},
{NULL, L"/\xfdd0zyx", "/%EF%BF%BDzyx", Component(0, 13), false},
};
for (size_t i = 0; i < arraysize(path_cases); i++) {
if (path_cases[i].input8) {
int len = static_cast<int>(strlen(path_cases[i].input8));
Component in_comp(0, len);
Component out_comp;
std::string out_str;
StdStringCanonOutput output(&out_str);
bool success =
CanonicalizePath(path_cases[i].input8, in_comp, &output, &out_comp);
output.Complete();
EXPECT_EQ(path_cases[i].expected_success, success);
EXPECT_EQ(path_cases[i].expected_component.begin, out_comp.begin);
EXPECT_EQ(path_cases[i].expected_component.len, out_comp.len);
EXPECT_EQ(path_cases[i].expected, out_str);
}
if (path_cases[i].input16) {
base::string16 input16(WStringToUTF16(path_cases[i].input16));
int len = static_cast<int>(input16.length());
Component in_comp(0, len);
Component out_comp;
std::string out_str;
StdStringCanonOutput output(&out_str);
bool success =
CanonicalizePath(input16.c_str(), in_comp, &output, &out_comp);
output.Complete();
EXPECT_EQ(path_cases[i].expected_success, success);
EXPECT_EQ(path_cases[i].expected_component.begin, out_comp.begin);
EXPECT_EQ(path_cases[i].expected_component.len, out_comp.len);
EXPECT_EQ(path_cases[i].expected, out_str);
}
}
// Manual test: embedded NULLs should be escaped and the URL should be marked
// as invalid.
const char path_with_null[] = "/ab\0c";
Component in_comp(0, 5);
Component out_comp;
std::string out_str;
StdStringCanonOutput output(&out_str);
bool success = CanonicalizePath(path_with_null, in_comp, &output, &out_comp);
output.Complete();
EXPECT_FALSE(success);
EXPECT_EQ("/ab%00c", out_str);
}
TEST(URLCanonTest, Query) {
struct QueryCase {
const char* input8;
const wchar_t* input16;
const char* expected;
} query_cases[] = {
// Regular ASCII case.
{"foo=bar", L"foo=bar", "?foo=bar"},
// Allow question marks in the query without escaping
{"as?df", L"as?df", "?as?df"},
// Always escape '#' since it would mark the ref.
{"as#df", L"as#df", "?as%23df"},
// Escape some questionable 8-bit characters, but never unescape.
{"\x02hello\x7f bye", L"\x02hello\x7f bye", "?%02hello%7F%20bye"},
{"%40%41123", L"%40%41123", "?%40%41123"},
// Chinese input/output
{"q=\xe4\xbd\xa0\xe5\xa5\xbd", L"q=\x4f60\x597d", "?q=%E4%BD%A0%E5%A5%BD"},
// Invalid UTF-8/16 input should be replaced with invalid characters.
{"q=\xed\xed", L"q=\xd800\xd800", "?q=%EF%BF%BD%EF%BF%BD"},
// Don't allow < or > because sometimes they are used for XSS if the
// URL is echoed in content. Firefox does this, IE doesn't.
{"q=<asdf>", L"q=<asdf>", "?q=%3Casdf%3E"},
// Escape double quotemarks in the query.
{"q=\"asdf\"", L"q=\"asdf\"", "?q=%22asdf%22"},
};
for (size_t i = 0; i < ARRAYSIZE(query_cases); i++) {
Component out_comp;
if (query_cases[i].input8) {
int len = static_cast<int>(strlen(query_cases[i].input8));
Component in_comp(0, len);
std::string out_str;
StdStringCanonOutput output(&out_str);
CanonicalizeQuery(query_cases[i].input8, in_comp, NULL, &output,
&out_comp);
output.Complete();
EXPECT_EQ(query_cases[i].expected, out_str);
}
if (query_cases[i].input16) {
base::string16 input16(WStringToUTF16(query_cases[i].input16));
int len = static_cast<int>(input16.length());
Component in_comp(0, len);
std::string out_str;
StdStringCanonOutput output(&out_str);
CanonicalizeQuery(input16.c_str(), in_comp, NULL, &output, &out_comp);
output.Complete();
EXPECT_EQ(query_cases[i].expected, out_str);
}
}
// Extra test for input with embedded NULL;
std::string out_str;
StdStringCanonOutput output(&out_str);
Component out_comp;
CanonicalizeQuery("a \x00z\x01", Component(0, 5), NULL, &output, &out_comp);
output.Complete();
EXPECT_EQ("?a%20%00z%01", out_str);
}
TEST(URLCanonTest, Ref) {
// Refs are trivial, it just checks the encoding.
DualComponentCase ref_cases[] = {
// Regular one, we shouldn't escape spaces, et al.
{"hello, world", L"hello, world", "#hello, world", Component(1, 12), true},
// UTF-8/wide input should be preserved
{"\xc2\xa9", L"\xa9", "#\xc2\xa9", Component(1, 2), true},
// Test a characer that takes > 16 bits (U+10300 = old italic letter A)
{"\xF0\x90\x8C\x80ss", L"\xd800\xdf00ss", "#\xF0\x90\x8C\x80ss", Component(1, 6), true},
// Escaping should be preserved unchanged, even invalid ones
{"%41%a", L"%41%a", "#%41%a", Component(1, 5), true},
// Invalid UTF-8/16 input should be flagged and the input made valid
{"\xc2", NULL, "#\xef\xbf\xbd", Component(1, 3), true},
{NULL, L"\xd800\x597d", "#\xef\xbf\xbd\xe5\xa5\xbd", Component(1, 6), true},
// Test a Unicode invalid character.
{"a\xef\xb7\x90", L"a\xfdd0", "#a\xef\xbf\xbd", Component(1, 4), true},
// Refs can have # signs and we should preserve them.
{"asdf#qwer", L"asdf#qwer", "#asdf#qwer", Component(1, 9), true},
{"#asdf", L"#asdf", "##asdf", Component(1, 5), true},
};
for (size_t i = 0; i < arraysize(ref_cases); i++) {
// 8-bit input
if (ref_cases[i].input8) {
int len = static_cast<int>(strlen(ref_cases[i].input8));
Component in_comp(0, len);
Component out_comp;
std::string out_str;
StdStringCanonOutput output(&out_str);
CanonicalizeRef(ref_cases[i].input8, in_comp, &output, &out_comp);
output.Complete();
EXPECT_EQ(ref_cases[i].expected_component.begin, out_comp.begin);
EXPECT_EQ(ref_cases[i].expected_component.len, out_comp.len);
EXPECT_EQ(ref_cases[i].expected, out_str);
}
// 16-bit input
if (ref_cases[i].input16) {
base::string16 input16(WStringToUTF16(ref_cases[i].input16));
int len = static_cast<int>(input16.length());
Component in_comp(0, len);
Component out_comp;
std::string out_str;
StdStringCanonOutput output(&out_str);
CanonicalizeRef(input16.c_str(), in_comp, &output, &out_comp);
output.Complete();
EXPECT_EQ(ref_cases[i].expected_component.begin, out_comp.begin);
EXPECT_EQ(ref_cases[i].expected_component.len, out_comp.len);
EXPECT_EQ(ref_cases[i].expected, out_str);
}
}
// Try one with an embedded NULL. It should be stripped.
const char null_input[5] = "ab\x00z";
Component null_input_component(0, 4);
Component out_comp;
std::string out_str;
StdStringCanonOutput output(&out_str);
CanonicalizeRef(null_input, null_input_component, &output, &out_comp);
output.Complete();
EXPECT_EQ(1, out_comp.begin);
EXPECT_EQ(3, out_comp.len);
EXPECT_EQ("#abz", out_str);
}
TEST(URLCanonTest, CanonicalizeStandardURL) {
// The individual component canonicalize tests should have caught the cases
// for each of those components. Here, we just need to test that the various
// parts are included or excluded properly, and have the correct separators.
struct URLCase {
const char* input;
const char* expected;
bool expected_success;
} cases[] = {
{"http://www.google.com/foo?bar=baz#", "http://www.google.com/foo?bar=baz#", true},
{"http://[www.google.com]/", "http://[www.google.com]/", false},
{"ht\ttp:@www.google.com:80/;p?#", "ht%09tp://www.google.com:80/;p?#", false},
{"http:////////user:@google.com:99?foo", "http://user@google.com:99/?foo", true},
{"www.google.com", ":www.google.com/", true},
{"http://192.0x00A80001", "http://192.168.0.1/", true},
{"http://www/foo%2Ehtml", "http://www/foo.html", true},
{"http://user:pass@/", "http://user:pass@/", false},
{"http://%25DOMAIN:foobar@foodomain.com/", "http://%25DOMAIN:foobar@foodomain.com/", true},
// Backslashes should get converted to forward slashes.
{"http:\\\\www.google.com\\foo", "http://www.google.com/foo", true},
// Busted refs shouldn't make the whole thing fail.
{"http://www.google.com/asdf#\xc2", "http://www.google.com/asdf#\xef\xbf\xbd", true},
// Basic port tests.
{"http://foo:80/", "http://foo/", true},
{"http://foo:81/", "http://foo:81/", true},
{"httpa://foo:80/", "httpa://foo:80/", true},
{"http://foo:-80/", "http://foo:-80/", false},
{"https://foo:443/", "https://foo/", true},
{"https://foo:80/", "https://foo:80/", true},
{"ftp://foo:21/", "ftp://foo/", true},
{"ftp://foo:80/", "ftp://foo:80/", true},
{"gopher://foo:70/", "gopher://foo/", true},
{"gopher://foo:443/", "gopher://foo:443/", true},
{"ws://foo:80/", "ws://foo/", true},
{"ws://foo:81/", "ws://foo:81/", true},
{"ws://foo:443/", "ws://foo:443/", true},
{"ws://foo:815/", "ws://foo:815/", true},
{"wss://foo:80/", "wss://foo:80/", true},
{"wss://foo:81/", "wss://foo:81/", true},
{"wss://foo:443/", "wss://foo/", true},
{"wss://foo:815/", "wss://foo:815/", true},
};
for (size_t i = 0; i < ARRAYSIZE(cases); i++) {
int url_len = static_cast<int>(strlen(cases[i].input));
Parsed parsed;
ParseStandardURL(cases[i].input, url_len, &parsed);
Parsed out_parsed;
std::string out_str;
StdStringCanonOutput output(&out_str);
bool success = CanonicalizeStandardURL(
cases[i].input, url_len, parsed, NULL, &output, &out_parsed);
output.Complete();
EXPECT_EQ(cases[i].expected_success, success);
EXPECT_EQ(cases[i].expected, out_str);
}
}
// The codepath here is the same as for regular canonicalization, so we just
// need to test that things are replaced or not correctly.
TEST(URLCanonTest, ReplaceStandardURL) {
ReplaceCase replace_cases[] = {
// Common case of truncating the path.
{"http://www.google.com/foo?bar=baz#ref", NULL, NULL, NULL, NULL, NULL, "/", kDeleteComp, kDeleteComp, "http://www.google.com/"},
// Replace everything
{"http://a:b@google.com:22/foo;bar?baz@cat", "https", "me", "pw", "host.com", "99", "/path", "query", "ref", "https://me:pw@host.com:99/path?query#ref"},
// Replace nothing
{"http://a:b@google.com:22/foo?baz@cat", NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, "http://a:b@google.com:22/foo?baz@cat"},
// Replace scheme with filesystem. The result is garbage, but you asked
// for it.
{"http://a:b@google.com:22/foo?baz@cat", "filesystem", NULL, NULL, NULL, NULL, NULL, NULL, NULL, "filesystem://a:b@google.com:22/foo?baz@cat"},
};
for (size_t i = 0; i < arraysize(replace_cases); i++) {
const ReplaceCase& cur = replace_cases[i];
int base_len = static_cast<int>(strlen(cur.base));
Parsed parsed;
ParseStandardURL(cur.base, base_len, &parsed);
Replacements<char> r;
typedef Replacements<char> R; // Clean up syntax.
// Note that for the scheme we pass in a different clear function since
// there is no function to clear the scheme.
SetupReplComp(&R::SetScheme, &R::ClearRef, &r, cur.scheme);
SetupReplComp(&R::SetUsername, &R::ClearUsername, &r, cur.username);
SetupReplComp(&R::SetPassword, &R::ClearPassword, &r, cur.password);
SetupReplComp(&R::SetHost, &R::ClearHost, &r, cur.host);
SetupReplComp(&R::SetPort, &R::ClearPort, &r, cur.port);
SetupReplComp(&R::SetPath, &R::ClearPath, &r, cur.path);
SetupReplComp(&R::SetQuery, &R::ClearQuery, &r, cur.query);
SetupReplComp(&R::SetRef, &R::ClearRef, &r, cur.ref);
std::string out_str;
StdStringCanonOutput output(&out_str);
Parsed out_parsed;
ReplaceStandardURL(replace_cases[i].base, parsed, r, NULL, &output,
&out_parsed);
output.Complete();
EXPECT_EQ(replace_cases[i].expected, out_str);
}
// The path pointer should be ignored if the address is invalid.
{
const char src[] = "http://www.google.com/here_is_the_path";
int src_len = static_cast<int>(strlen(src));
Parsed parsed;
ParseStandardURL(src, src_len, &parsed);
// Replace the path to 0 length string. By using 1 as the string address,
// the test should get an access violation if it tries to dereference it.
Replacements<char> r;
r.SetPath(reinterpret_cast<char*>(0x00000001), Component(0, 0));
std::string out_str1;
StdStringCanonOutput output1(&out_str1);
Parsed new_parsed;
ReplaceStandardURL(src, parsed, r, NULL, &output1, &new_parsed);
output1.Complete();
EXPECT_STREQ("http://www.google.com/", out_str1.c_str());
// Same with an "invalid" path.
r.SetPath(reinterpret_cast<char*>(0x00000001), Component());
std::string out_str2;
StdStringCanonOutput output2(&out_str2);
ReplaceStandardURL(src, parsed, r, NULL, &output2, &new_parsed);
output2.Complete();
EXPECT_STREQ("http://www.google.com/", out_str2.c_str());
}
}
TEST(URLCanonTest, ReplaceFileURL) {
ReplaceCase replace_cases[] = {
// Replace everything
{"file:///C:/gaba?query#ref", NULL, NULL, NULL, "filer", NULL, "/foo", "b", "c", "file://filer/foo?b#c"},
// Replace nothing
{"file:///C:/gaba?query#ref", NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, "file:///C:/gaba?query#ref"},
// Clear non-path components (common)
{"file:///C:/gaba?query#ref", NULL, NULL, NULL, NULL, NULL, NULL, kDeleteComp, kDeleteComp, "file:///C:/gaba"},
// Replace path with something that doesn't begin with a slash and make
// sure it gets added properly.
{"file:///C:/gaba", NULL, NULL, NULL, NULL, NULL, "interesting/", NULL, NULL, "file:///interesting/"},
{"file:///home/gaba?query#ref", NULL, NULL, NULL, "filer", NULL, "/foo", "b", "c", "file://filer/foo?b#c"},
{"file:///home/gaba?query#ref", NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, "file:///home/gaba?query#ref"},
{"file:///home/gaba?query#ref", NULL, NULL, NULL, NULL, NULL, NULL, kDeleteComp, kDeleteComp, "file:///home/gaba"},
{"file:///home/gaba", NULL, NULL, NULL, NULL, NULL, "interesting/", NULL, NULL, "file:///interesting/"},
// Replace scheme -- shouldn't do anything.
{"file:///C:/gaba?query#ref", "http", NULL, NULL, NULL, NULL, NULL, NULL, NULL, "file:///C:/gaba?query#ref"},
};
for (size_t i = 0; i < arraysize(replace_cases); i++) {
const ReplaceCase& cur = replace_cases[i];
int base_len = static_cast<int>(strlen(cur.base));
Parsed parsed;
ParseFileURL(cur.base, base_len, &parsed);
Replacements<char> r;
typedef Replacements<char> R; // Clean up syntax.
SetupReplComp(&R::SetScheme, &R::ClearRef, &r, cur.scheme);
SetupReplComp(&R::SetUsername, &R::ClearUsername, &r, cur.username);
SetupReplComp(&R::SetPassword, &R::ClearPassword, &r, cur.password);
SetupReplComp(&R::SetHost, &R::ClearHost, &r, cur.host);
SetupReplComp(&R::SetPort, &R::ClearPort, &r, cur.port);
SetupReplComp(&R::SetPath, &R::ClearPath, &r, cur.path);
SetupReplComp(&R::SetQuery, &R::ClearQuery, &r, cur.query);
SetupReplComp(&R::SetRef, &R::ClearRef, &r, cur.ref);
std::string out_str;
StdStringCanonOutput output(&out_str);
Parsed out_parsed;
ReplaceFileURL(cur.base, parsed, r, NULL, &output, &out_parsed);
output.Complete();
EXPECT_EQ(replace_cases[i].expected, out_str);
}
}
TEST(URLCanonTest, ReplaceFileSystemURL) {
ReplaceCase replace_cases[] = {
// Replace everything in the outer URL.
{"filesystem:file:///temporary/gaba?query#ref", NULL, NULL, NULL, NULL, NULL, "/foo", "b", "c", "filesystem:file:///temporary/foo?b#c"},
// Replace nothing
{"filesystem:file:///temporary/gaba?query#ref", NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, "filesystem:file:///temporary/gaba?query#ref"},
// Clear non-path components (common)
{"filesystem:file:///temporary/gaba?query#ref", NULL, NULL, NULL, NULL, NULL, NULL, kDeleteComp, kDeleteComp, "filesystem:file:///temporary/gaba"},
// Replace path with something that doesn't begin with a slash and make
// sure it gets added properly.
{"filesystem:file:///temporary/gaba?query#ref", NULL, NULL, NULL, NULL, NULL, "interesting/", NULL, NULL, "filesystem:file:///temporary/interesting/?query#ref"},
// Replace scheme -- shouldn't do anything.
{"filesystem:http://u:p@bar.com/t/gaba?query#ref", "http", NULL, NULL, NULL, NULL, NULL, NULL, NULL, "filesystem:http://u:p@bar.com/t/gaba?query#ref"},
// Replace username -- shouldn't do anything.
{"filesystem:http://u:p@bar.com/t/gaba?query#ref", NULL, "u2", NULL, NULL, NULL, NULL, NULL, NULL, "filesystem:http://u:p@bar.com/t/gaba?query#ref"},
// Replace password -- shouldn't do anything.
{"filesystem:http://u:p@bar.com/t/gaba?query#ref", NULL, NULL, "pw2", NULL, NULL, NULL, NULL, NULL, "filesystem:http://u:p@bar.com/t/gaba?query#ref"},
// Replace host -- shouldn't do anything.
{"filesystem:http://u:p@bar.com/t/gaba?query#ref", NULL, NULL, NULL, "foo.com", NULL, NULL, NULL, NULL, "filesystem:http://u:p@bar.com/t/gaba?query#ref"},
// Replace port -- shouldn't do anything.
{"filesystem:http://u:p@bar.com:40/t/gaba?query#ref", NULL, NULL, NULL, NULL, "41", NULL, NULL, NULL, "filesystem:http://u:p@bar.com:40/t/gaba?query#ref"},
};
for (size_t i = 0; i < arraysize(replace_cases); i++) {
const ReplaceCase& cur = replace_cases[i];
int base_len = static_cast<int>(strlen(cur.base));
Parsed parsed;
ParseFileSystemURL(cur.base, base_len, &parsed);
Replacements<char> r;
typedef Replacements<char> R; // Clean up syntax.
SetupReplComp(&R::SetScheme, &R::ClearRef, &r, cur.scheme);
SetupReplComp(&R::SetUsername, &R::ClearUsername, &r, cur.username);
SetupReplComp(&R::SetPassword, &R::ClearPassword, &r, cur.password);
SetupReplComp(&R::SetHost, &R::ClearHost, &r, cur.host);
SetupReplComp(&R::SetPort, &R::ClearPort, &r, cur.port);
SetupReplComp(&R::SetPath, &R::ClearPath, &r, cur.path);
SetupReplComp(&R::SetQuery, &R::ClearQuery, &r, cur.query);
SetupReplComp(&R::SetRef, &R::ClearRef, &r, cur.ref);
std::string out_str;
StdStringCanonOutput output(&out_str);
Parsed out_parsed;
ReplaceFileSystemURL(cur.base, parsed, r, NULL, &output, &out_parsed);
output.Complete();
EXPECT_EQ(replace_cases[i].expected, out_str);
}
}
TEST(URLCanonTest, ReplacePathURL) {
ReplaceCase replace_cases[] = {
// Replace everything
{"data:foo", "javascript", NULL, NULL, NULL, NULL, "alert('foo?');", NULL, NULL, "javascript:alert('foo?');"},
// Replace nothing
{"data:foo", NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, "data:foo"},
// Replace one or the other
{"data:foo", "javascript", NULL, NULL, NULL, NULL, NULL, NULL, NULL, "javascript:foo"},
{"data:foo", NULL, NULL, NULL, NULL, NULL, "bar", NULL, NULL, "data:bar"},
{"data:foo", NULL, NULL, NULL, NULL, NULL, kDeleteComp, NULL, NULL, "data:"},
};
for (size_t i = 0; i < arraysize(replace_cases); i++) {
const ReplaceCase& cur = replace_cases[i];
int base_len = static_cast<int>(strlen(cur.base));
Parsed parsed;
ParsePathURL(cur.base, base_len, false, &parsed);
Replacements<char> r;
typedef Replacements<char> R; // Clean up syntax.
SetupReplComp(&R::SetScheme, &R::ClearRef, &r, cur.scheme);
SetupReplComp(&R::SetUsername, &R::ClearUsername, &r, cur.username);
SetupReplComp(&R::SetPassword, &R::ClearPassword, &r, cur.password);
SetupReplComp(&R::SetHost, &R::ClearHost, &r, cur.host);
SetupReplComp(&R::SetPort, &R::ClearPort, &r, cur.port);
SetupReplComp(&R::SetPath, &R::ClearPath, &r, cur.path);
SetupReplComp(&R::SetQuery, &R::ClearQuery, &r, cur.query);
SetupReplComp(&R::SetRef, &R::ClearRef, &r, cur.ref);
std::string out_str;
StdStringCanonOutput output(&out_str);
Parsed out_parsed;
ReplacePathURL(cur.base, parsed, r, &output, &out_parsed);
output.Complete();
EXPECT_EQ(replace_cases[i].expected, out_str);
}
}
TEST(URLCanonTest, ReplaceMailtoURL) {
ReplaceCase replace_cases[] = {
// Replace everything
{"mailto:jon@foo.com?body=sup", "mailto", NULL, NULL, NULL, NULL, "addr1", "to=tony", NULL, "mailto:addr1?to=tony"},
// Replace nothing
{"mailto:jon@foo.com?body=sup", NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, "mailto:jon@foo.com?body=sup"},
// Replace the path
{"mailto:jon@foo.com?body=sup", NULL, NULL, NULL, NULL, NULL, "jason", NULL, NULL, "mailto:jason?body=sup"},
// Replace the query
{"mailto:jon@foo.com?body=sup", NULL, NULL, NULL, NULL, NULL, NULL, "custom=1", NULL, "mailto:jon@foo.com?custom=1"},
// Replace the path and query
{"mailto:jon@foo.com?body=sup", NULL, NULL, NULL, NULL, NULL, "jason", "custom=1", NULL, "mailto:jason?custom=1"},
// Set the query to empty (should leave trailing question mark)
{"mailto:jon@foo.com?body=sup", NULL, NULL, NULL, NULL, NULL, NULL, "", NULL, "mailto:jon@foo.com?"},
// Clear the query
{"mailto:jon@foo.com?body=sup", NULL, NULL, NULL, NULL, NULL, NULL, "|", NULL, "mailto:jon@foo.com"},
// Clear the path
{"mailto:jon@foo.com?body=sup", NULL, NULL, NULL, NULL, NULL, "|", NULL, NULL, "mailto:?body=sup"},
// Clear the path + query
{"mailto:", NULL, NULL, NULL, NULL, NULL, "|", "|", NULL, "mailto:"},
// Setting the ref should have no effect
{"mailto:addr1", NULL, NULL, NULL, NULL, NULL, NULL, NULL, "BLAH", "mailto:addr1"},
};
for (size_t i = 0; i < arraysize(replace_cases); i++) {
const ReplaceCase& cur = replace_cases[i];
int base_len = static_cast<int>(strlen(cur.base));
Parsed parsed;
ParseMailtoURL(cur.base, base_len, &parsed);
Replacements<char> r;
typedef Replacements<char> R;
SetupReplComp(&R::SetScheme, &R::ClearRef, &r, cur.scheme);
SetupReplComp(&R::SetUsername, &R::ClearUsername, &r, cur.username);
SetupReplComp(&R::SetPassword, &R::ClearPassword, &r, cur.password);
SetupReplComp(&R::SetHost, &R::ClearHost, &r, cur.host);
SetupReplComp(&R::SetPort, &R::ClearPort, &r, cur.port);
SetupReplComp(&R::SetPath, &R::ClearPath, &r, cur.path);
SetupReplComp(&R::SetQuery, &R::ClearQuery, &r, cur.query);
SetupReplComp(&R::SetRef, &R::ClearRef, &r, cur.ref);
std::string out_str;
StdStringCanonOutput output(&out_str);
Parsed out_parsed;
ReplaceMailtoURL(cur.base, parsed, r, &output, &out_parsed);
output.Complete();
EXPECT_EQ(replace_cases[i].expected, out_str);
}
}
TEST(URLCanonTest, CanonicalizeFileURL) {
struct URLCase {
const char* input;
const char* expected;
bool expected_success;
Component expected_host;
Component expected_path;
} cases[] = {
#ifdef _WIN32
// Windows-style paths
{"file:c:\\foo\\bar.html", "file:///C:/foo/bar.html", true, Component(), Component(7, 16)},
{" File:c|////foo\\bar.html", "file:///C:////foo/bar.html", true, Component(), Component(7, 19)},
{"file:", "file:///", true, Component(), Component(7, 1)},
{"file:UNChost/path", "file://unchost/path", true, Component(7, 7), Component(14, 5)},
// CanonicalizeFileURL supports absolute Windows style paths for IE
// compatability. Note that the caller must decide that this is a file
// URL itself so it can call the file canonicalizer. This is usually
// done automatically as part of relative URL resolving.
{"c:\\foo\\bar", "file:///C:/foo/bar", true, Component(), Component(7, 11)},
{"C|/foo/bar", "file:///C:/foo/bar", true, Component(), Component(7, 11)},
{"/C|\\foo\\bar", "file:///C:/foo/bar", true, Component(), Component(7, 11)},
{"//C|/foo/bar", "file:///C:/foo/bar", true, Component(), Component(7, 11)},
{"//server/file", "file://server/file", true, Component(7, 6), Component(13, 5)},
{"\\\\server\\file", "file://server/file", true, Component(7, 6), Component(13, 5)},
{"/\\server/file", "file://server/file", true, Component(7, 6), Component(13, 5)},
// We should preserve the number of slashes after the colon for IE
// compatability, except when there is none, in which case we should
// add one.
{"file:c:foo/bar.html", "file:///C:/foo/bar.html", true, Component(), Component(7, 16)},
{"file:/\\/\\C:\\\\//foo\\bar.html", "file:///C:////foo/bar.html", true, Component(), Component(7, 19)},
// Three slashes should be non-UNC, even if there is no drive spec (IE
// does this, which makes the resulting request invalid).
{"file:///foo/bar.txt", "file:///foo/bar.txt", true, Component(), Component(7, 12)},
// TODO(brettw) we should probably fail for invalid host names, which
// would change the expected result on this test. We also currently allow
// colon even though it's probably invalid, because its currently the
// "natural" result of the way the canonicalizer is written. There doesn't
// seem to be a strong argument for why allowing it here would be bad, so
// we just tolerate it and the load will fail later.
{"FILE:/\\/\\7:\\\\//foo\\bar.html", "file://7:////foo/bar.html", false, Component(7, 2), Component(9, 16)},
{"file:filer/home\\me", "file://filer/home/me", true, Component(7, 5), Component(12, 8)},
// Make sure relative paths can't go above the "C:"
{"file:///C:/foo/../../../bar.html", "file:///C:/bar.html", true, Component(), Component(7, 12)},
// Busted refs shouldn't make the whole thing fail.
{"file:///C:/asdf#\xc2", "file:///C:/asdf#\xef\xbf\xbd", true, Component(), Component(7, 8)},
#else
// Unix-style paths
{"file:///home/me", "file:///home/me", true, Component(), Component(7, 8)},
// Windowsy ones should get still treated as Unix-style.
{"file:c:\\foo\\bar.html", "file:///c:/foo/bar.html", true, Component(), Component(7, 16)},
{"file:c|//foo\\bar.html", "file:///c%7C//foo/bar.html", true, Component(), Component(7, 19)},
// file: tests from WebKit (LayoutTests/fast/loader/url-parse-1.html)
{"//", "file:///", true, Component(), Component(7, 1)},
{"///", "file:///", true, Component(), Component(7, 1)},
{"///test", "file:///test", true, Component(), Component(7, 5)},
{"file://test", "file://test/", true, Component(7, 4), Component(11, 1)},
{"file://localhost", "file://localhost/", true, Component(7, 9), Component(16, 1)},
{"file://localhost/", "file://localhost/", true, Component(7, 9), Component(16, 1)},
{"file://localhost/test", "file://localhost/test", true, Component(7, 9), Component(16, 5)},
#endif // _WIN32
};
for (size_t i = 0; i < ARRAYSIZE(cases); i++) {
int url_len = static_cast<int>(strlen(cases[i].input));
Parsed parsed;
ParseFileURL(cases[i].input, url_len, &parsed);
Parsed out_parsed;
std::string out_str;
StdStringCanonOutput output(&out_str);
bool success = CanonicalizeFileURL(cases[i].input, url_len, parsed, NULL,
&output, &out_parsed);
output.Complete();
EXPECT_EQ(cases[i].expected_success, success);
EXPECT_EQ(cases[i].expected, out_str);
// Make sure the spec was properly identified, the file canonicalizer has
// different code for writing the spec.
EXPECT_EQ(0, out_parsed.scheme.begin);
EXPECT_EQ(4, out_parsed.scheme.len);
EXPECT_EQ(cases[i].expected_host.begin, out_parsed.host.begin);
EXPECT_EQ(cases[i].expected_host.len, out_parsed.host.len);
EXPECT_EQ(cases[i].expected_path.begin, out_parsed.path.begin);
EXPECT_EQ(cases[i].expected_path.len, out_parsed.path.len);
}
}
TEST(URLCanonTest, CanonicalizeFileSystemURL) {
struct URLCase {
const char* input;
const char* expected;
bool expected_success;
} cases[] = {
{"Filesystem:htTp://www.Foo.com:80/tempoRary", "filesystem:http://www.foo.com/tempoRary/", true},
{"filesystem:httpS://www.foo.com/temporary/", "filesystem:https://www.foo.com/temporary/", true},
{"filesystem:http://www.foo.com//", "filesystem:http://www.foo.com//", false},
{"filesystem:http://www.foo.com/persistent/bob?query#ref", "filesystem:http://www.foo.com/persistent/bob?query#ref", true},
{"filesystem:fIle://\\temporary/", "filesystem:file:///temporary/", true},
{"filesystem:fiLe:///temporary", "filesystem:file:///temporary/", true},
{"filesystem:File:///temporary/Bob?qUery#reF", "filesystem:file:///temporary/Bob?qUery#reF", true},
};
for (size_t i = 0; i < ARRAYSIZE(cases); i++) {
int url_len = static_cast<int>(strlen(cases[i].input));
Parsed parsed;
ParseFileSystemURL(cases[i].input, url_len, &parsed);
Parsed out_parsed;
std::string out_str;
StdStringCanonOutput output(&out_str);
bool success = CanonicalizeFileSystemURL(cases[i].input, url_len, parsed,
NULL, &output, &out_parsed);
output.Complete();
EXPECT_EQ(cases[i].expected_success, success);
EXPECT_EQ(cases[i].expected, out_str);
// Make sure the spec was properly identified, the filesystem canonicalizer
// has different code for writing the spec.
EXPECT_EQ(0, out_parsed.scheme.begin);
EXPECT_EQ(10, out_parsed.scheme.len);
if (success)
EXPECT_GT(out_parsed.path.len, 0);
}
}
TEST(URLCanonTest, CanonicalizePathURL) {
// Path URLs should get canonicalized schemes but nothing else.
struct PathCase {
const char* input;
const char* expected;
} path_cases[] = {
{"javascript:", "javascript:"},
{"JavaScript:Foo", "javascript:Foo"},
{":\":This /is interesting;?#", ":\":This /is interesting;?#"},
};
for (size_t i = 0; i < ARRAYSIZE(path_cases); i++) {
int url_len = static_cast<int>(strlen(path_cases[i].input));
Parsed parsed;
ParsePathURL(path_cases[i].input, url_len, true, &parsed);
Parsed out_parsed;
std::string out_str;
StdStringCanonOutput output(&out_str);
bool success = CanonicalizePathURL(path_cases[i].input, url_len, parsed,
&output, &out_parsed);
output.Complete();
EXPECT_TRUE(success);
EXPECT_EQ(path_cases[i].expected, out_str);
EXPECT_EQ(0, out_parsed.host.begin);
EXPECT_EQ(-1, out_parsed.host.len);
// When we end with a colon at the end, there should be no path.
if (path_cases[i].input[url_len - 1] == ':') {
EXPECT_EQ(0, out_parsed.GetContent().begin);
EXPECT_EQ(-1, out_parsed.GetContent().len);
}
}
}
TEST(URLCanonTest, CanonicalizeMailtoURL) {
struct URLCase {
const char* input;
const char* expected;
bool expected_success;
Component expected_path;
Component expected_query;
} cases[] = {
{"mailto:addr1", "mailto:addr1", true, Component(7, 5), Component()},
{"mailto:addr1@foo.com", "mailto:addr1@foo.com", true, Component(7, 13), Component()},
// Trailing whitespace is stripped.
{"MaIlTo:addr1 \t ", "mailto:addr1", true, Component(7, 5), Component()},
{"MaIlTo:addr1?to=jon", "mailto:addr1?to=jon", true, Component(7, 5), Component(13,6)},
{"mailto:addr1,addr2", "mailto:addr1,addr2", true, Component(7, 11), Component()},
{"mailto:addr1, addr2", "mailto:addr1, addr2", true, Component(7, 12), Component()},
{"mailto:addr1%2caddr2", "mailto:addr1%2caddr2", true, Component(7, 13), Component()},
{"mailto:\xF0\x90\x8C\x80", "mailto:%F0%90%8C%80", true, Component(7, 12), Component()},
// Null character should be escaped to %00
{"mailto:addr1\0addr2?foo", "mailto:addr1%00addr2?foo", true, Component(7, 13), Component(21, 3)},
// Invalid -- UTF-8 encoded surrogate value.
{"mailto:\xed\xa0\x80", "mailto:%EF%BF%BD", false, Component(7, 9), Component()},
{"mailto:addr1?", "mailto:addr1?", true, Component(7, 5), Component(13, 0)},
};
// Define outside of loop to catch bugs where components aren't reset
Parsed parsed;
Parsed out_parsed;
for (size_t i = 0; i < ARRAYSIZE(cases); i++) {
int url_len = static_cast<int>(strlen(cases[i].input));
if (i == 8) {
// The 9th test case purposely has a '\0' in it -- don't count it
// as the string terminator.
url_len = 22;
}
ParseMailtoURL(cases[i].input, url_len, &parsed);
std::string out_str;
StdStringCanonOutput output(&out_str);
bool success = CanonicalizeMailtoURL(cases[i].input, url_len, parsed,
&output, &out_parsed);
output.Complete();
EXPECT_EQ(cases[i].expected_success, success);
EXPECT_EQ(cases[i].expected, out_str);
// Make sure the spec was properly identified
EXPECT_EQ(0, out_parsed.scheme.begin);
EXPECT_EQ(6, out_parsed.scheme.len);
EXPECT_EQ(cases[i].expected_path.begin, out_parsed.path.begin);
EXPECT_EQ(cases[i].expected_path.len, out_parsed.path.len);
EXPECT_EQ(cases[i].expected_query.begin, out_parsed.query.begin);
EXPECT_EQ(cases[i].expected_query.len, out_parsed.query.len);
}
}
#ifndef WIN32
TEST(URLCanonTest, _itoa_s) {
// We fill the buffer with 0xff to ensure that it's getting properly
// null-terminated. We also allocate one byte more than what we tell
// _itoa_s about, and ensure that the extra byte is untouched.
char buf[6];
memset(buf, 0xff, sizeof(buf));
EXPECT_EQ(0, _itoa_s(12, buf, sizeof(buf) - 1, 10));
EXPECT_STREQ("12", buf);
EXPECT_EQ('\xFF', buf[3]);
// Test the edge cases - exactly the buffer size and one over
memset(buf, 0xff, sizeof(buf));
EXPECT_EQ(0, _itoa_s(1234, buf, sizeof(buf) - 1, 10));
EXPECT_STREQ("1234", buf);
EXPECT_EQ('\xFF', buf[5]);
memset(buf, 0xff, sizeof(buf));
EXPECT_EQ(EINVAL, _itoa_s(12345, buf, sizeof(buf) - 1, 10));
EXPECT_EQ('\xFF', buf[5]); // should never write to this location
// Test the template overload (note that this will see the full buffer)
memset(buf, 0xff, sizeof(buf));
EXPECT_EQ(0, _itoa_s(12, buf, 10));
EXPECT_STREQ("12", buf);
EXPECT_EQ('\xFF', buf[3]);
memset(buf, 0xff, sizeof(buf));
EXPECT_EQ(0, _itoa_s(12345, buf, 10));
EXPECT_STREQ("12345", buf);
EXPECT_EQ(EINVAL, _itoa_s(123456, buf, 10));
// Test that radix 16 is supported.
memset(buf, 0xff, sizeof(buf));
EXPECT_EQ(0, _itoa_s(1234, buf, sizeof(buf) - 1, 16));
EXPECT_STREQ("4d2", buf);
EXPECT_EQ('\xFF', buf[5]);
}
TEST(URLCanonTest, _itow_s) {
// We fill the buffer with 0xff to ensure that it's getting properly
// null-terminated. We also allocate one byte more than what we tell
// _itoa_s about, and ensure that the extra byte is untouched.
base::char16 buf[6];
const char fill_mem = 0xff;
const base::char16 fill_char = 0xffff;
memset(buf, fill_mem, sizeof(buf));
EXPECT_EQ(0, _itow_s(12, buf, sizeof(buf) / 2 - 1, 10));
EXPECT_EQ(WStringToUTF16(L"12"), base::string16(buf));
EXPECT_EQ(fill_char, buf[3]);
// Test the edge cases - exactly the buffer size and one over
EXPECT_EQ(0, _itow_s(1234, buf, sizeof(buf) / 2 - 1, 10));
EXPECT_EQ(WStringToUTF16(L"1234"), base::string16(buf));
EXPECT_EQ(fill_char, buf[5]);
memset(buf, fill_mem, sizeof(buf));
EXPECT_EQ(EINVAL, _itow_s(12345, buf, sizeof(buf) / 2 - 1, 10));
EXPECT_EQ(fill_char, buf[5]); // should never write to this location
// Test the template overload (note that this will see the full buffer)
memset(buf, fill_mem, sizeof(buf));
EXPECT_EQ(0, _itow_s(12, buf, 10));
EXPECT_EQ(WStringToUTF16(L"12"), base::string16(buf));
EXPECT_EQ(fill_char, buf[3]);
memset(buf, fill_mem, sizeof(buf));
EXPECT_EQ(0, _itow_s(12345, buf, 10));
EXPECT_EQ(WStringToUTF16(L"12345"), base::string16(buf));
EXPECT_EQ(EINVAL, _itow_s(123456, buf, 10));
}
#endif // !WIN32
// Returns true if the given two structures are the same.
static bool ParsedIsEqual(const Parsed& a, const Parsed& b) {
return a.scheme.begin == b.scheme.begin && a.scheme.len == b.scheme.len &&
a.username.begin == b.username.begin && a.username.len == b.username.len &&
a.password.begin == b.password.begin && a.password.len == b.password.len &&
a.host.begin == b.host.begin && a.host.len == b.host.len &&
a.port.begin == b.port.begin && a.port.len == b.port.len &&
a.path.begin == b.path.begin && a.path.len == b.path.len &&
a.query.begin == b.query.begin && a.query.len == b.query.len &&
a.ref.begin == b.ref.begin && a.ref.len == b.ref.len;
}
TEST(URLCanonTest, ResolveRelativeURL) {
struct RelativeCase {
const char* base; // Input base URL: MUST BE CANONICAL
bool is_base_hier; // Is the base URL hierarchical
bool is_base_file; // Tells us if the base is a file URL.
const char* test; // Input URL to test against.
bool succeed_relative; // Whether we expect IsRelativeURL to succeed
bool is_rel; // Whether we expect |test| to be relative or not.
bool succeed_resolve; // Whether we expect ResolveRelativeURL to succeed.
const char* resolved; // What we expect in the result when resolving.
} rel_cases[] = {
// Basic absolute input.
{"http://host/a", true, false, "http://another/", true, false, false, NULL},
{"http://host/a", true, false, "http:////another/", true, false, false, NULL},
// Empty relative URLs should only remove the ref part of the URL,
// leaving the rest unchanged.
{"http://foo/bar", true, false, "", true, true, true, "http://foo/bar"},
{"http://foo/bar#ref", true, false, "", true, true, true, "http://foo/bar"},
{"http://foo/bar#", true, false, "", true, true, true, "http://foo/bar"},
// Spaces at the ends of the relative path should be ignored.
{"http://foo/bar", true, false, " another ", true, true, true, "http://foo/another"},
{"http://foo/bar", true, false, " . ", true, true, true, "http://foo/"},
{"http://foo/bar", true, false, " \t ", true, true, true, "http://foo/bar"},
// Matching schemes without two slashes are treated as relative.
{"http://host/a", true, false, "http:path", true, true, true, "http://host/path"},
{"http://host/a/", true, false, "http:path", true, true, true, "http://host/a/path"},
{"http://host/a", true, false, "http:/path", true, true, true, "http://host/path"},
{"http://host/a", true, false, "HTTP:/path", true, true, true, "http://host/path"},
// Nonmatching schemes are absolute.
{"http://host/a", true, false, "https:host2", true, false, false, NULL},
{"http://host/a", true, false, "htto:/host2", true, false, false, NULL},
// Absolute path input
{"http://host/a", true, false, "/b/c/d", true, true, true, "http://host/b/c/d"},
{"http://host/a", true, false, "\\b\\c\\d", true, true, true, "http://host/b/c/d"},
{"http://host/a", true, false, "/b/../c", true, true, true, "http://host/c"},
{"http://host/a?b#c", true, false, "/b/../c", true, true, true, "http://host/c"},
{"http://host/a", true, false, "\\b/../c?x#y", true, true, true, "http://host/c?x#y"},
{"http://host/a?b#c", true, false, "/b/../c?x#y", true, true, true, "http://host/c?x#y"},
// Relative path input
{"http://host/a", true, false, "b", true, true, true, "http://host/b"},
{"http://host/a", true, false, "bc/de", true, true, true, "http://host/bc/de"},
{"http://host/a/", true, false, "bc/de?query#ref", true, true, true, "http://host/a/bc/de?query#ref"},
{"http://host/a/", true, false, ".", true, true, true, "http://host/a/"},
{"http://host/a/", true, false, "..", true, true, true, "http://host/"},
{"http://host/a/", true, false, "./..", true, true, true, "http://host/"},
{"http://host/a/", true, false, "../.", true, true, true, "http://host/"},
{"http://host/a/", true, false, "././.", true, true, true, "http://host/a/"},
{"http://host/a?query#ref", true, false, "../../../foo", true, true, true, "http://host/foo"},
// Query input
{"http://host/a", true, false, "?foo=bar", true, true, true, "http://host/a?foo=bar"},
{"http://host/a?x=y#z", true, false, "?", true, true, true, "http://host/a?"},
{"http://host/a?x=y#z", true, false, "?foo=bar#com", true, true, true, "http://host/a?foo=bar#com"},
// Ref input
{"http://host/a", true, false, "#ref", true, true, true, "http://host/a#ref"},
{"http://host/a#b", true, false, "#", true, true, true, "http://host/a#"},
{"http://host/a?foo=bar#hello", true, false, "#bye", true, true, true, "http://host/a?foo=bar#bye"},
// Non-hierarchical base: no relative handling. Relative input should
// error, and if a scheme is present, it should be treated as absolute.
{"data:foobar", false, false, "baz.html", false, false, false, NULL},
{"data:foobar", false, false, "data:baz", true, false, false, NULL},
{"data:foobar", false, false, "data:/base", true, false, false, NULL},
// Non-hierarchical base: absolute input should succeed.
{"data:foobar", false, false, "http://host/", true, false, false, NULL},
{"data:foobar", false, false, "http:host", true, false, false, NULL},
// Invalid schemes should be treated as relative.
{"http://foo/bar", true, false, "./asd:fgh", true, true, true, "http://foo/asd:fgh"},
{"http://foo/bar", true, false, ":foo", true, true, true, "http://foo/:foo"},
{"http://foo/bar", true, false, " hello world", true, true, true, "http://foo/hello%20world"},
{"data:asdf", false, false, ":foo", false, false, false, NULL},
{"data:asdf", false, false, "bad(':foo')", false, false, false, NULL},
// We should treat semicolons like any other character in URL resolving
{"http://host/a", true, false, ";foo", true, true, true, "http://host/;foo"},
{"http://host/a;", true, false, ";foo", true, true, true, "http://host/;foo"},
{"http://host/a", true, false, ";/../bar", true, true, true, "http://host/bar"},
// Relative URLs can also be written as "//foo/bar" which is relative to
// the scheme. In this case, it would take the old scheme, so for http
// the example would resolve to "http://foo/bar".
{"http://host/a", true, false, "//another", true, true, true, "http://another/"},
{"http://host/a", true, false, "//another/path?query#ref", true, true, true, "http://another/path?query#ref"},
{"http://host/a", true, false, "///another/path", true, true, true, "http://another/path"},
{"http://host/a", true, false, "//Another\\path", true, true, true, "http://another/path"},
{"http://host/a", true, false, "//", true, true, false, "http:"},
// IE will also allow one or the other to be a backslash to get the same
// behavior.
{"http://host/a", true, false, "\\/another/path", true, true, true, "http://another/path"},
{"http://host/a", true, false, "/\\Another\\path", true, true, true, "http://another/path"},
#ifdef WIN32
// Resolving against Windows file base URLs.
{"file:///C:/foo", true, true, "http://host/", true, false, false, NULL},
{"file:///C:/foo", true, true, "bar", true, true, true, "file:///C:/bar"},
{"file:///C:/foo", true, true, "../../../bar.html", true, true, true, "file:///C:/bar.html"},
{"file:///C:/foo", true, true, "/../bar.html", true, true, true, "file:///C:/bar.html"},
// But two backslashes on Windows should be UNC so should be treated
// as absolute.
{"http://host/a", true, false, "\\\\another\\path", true, false, false, NULL},
// IE doesn't support drive specs starting with two slashes. It fails
// immediately and doesn't even try to load. We fix it up to either
// an absolute path or UNC depending on what it looks like.
{"file:///C:/something", true, true, "//c:/foo", true, true, true, "file:///C:/foo"},
{"file:///C:/something", true, true, "//localhost/c:/foo", true, true, true, "file:///C:/foo"},
// Windows drive specs should be allowed and treated as absolute.
{"file:///C:/foo", true, true, "c:", true, false, false, NULL},
{"file:///C:/foo", true, true, "c:/foo", true, false, false, NULL},
{"http://host/a", true, false, "c:\\foo", true, false, false, NULL},
// Relative paths with drive letters should be allowed when the base is
// also a file.
{"file:///C:/foo", true, true, "/z:/bar", true, true, true, "file:///Z:/bar"},
// Treat absolute paths as being off of the drive.
{"file:///C:/foo", true, true, "/bar", true, true, true, "file:///C:/bar"},
{"file://localhost/C:/foo", true, true, "/bar", true, true, true, "file://localhost/C:/bar"},
{"file:///C:/foo/com/", true, true, "/bar", true, true, true, "file:///C:/bar"},
// On Windows, two slashes without a drive letter when the base is a file
// means that the path is UNC.
{"file:///C:/something", true, true, "//somehost/path", true, true, true, "file://somehost/path"},
{"file:///C:/something", true, true, "/\\//somehost/path", true, true, true, "file://somehost/path"},
#else
// On Unix we fall back to relative behavior since there's nothing else
// reasonable to do.
{"http://host/a", true, false, "\\\\Another\\path", true, true, true, "http://another/path"},
#endif
// Even on Windows, we don't allow relative drive specs when the base
// is not file.
{"http://host/a", true, false, "/c:\\foo", true, true, true, "http://host/c:/foo"},
{"http://host/a", true, false, "//c:\\foo", true, true, true, "http://c/foo"},
// Ensure that ports aren't allowed for hosts relative to a file url.
// Although the result string shows a host:port portion, the call to
// resolve the relative URL returns false, indicating parse failure,
// which is what is required.
{"file:///foo.txt", true, true, "//host:80/bar.txt", true, true, false, "file://host:80/bar.txt"},
// Filesystem URL tests; filesystem URLs are only valid and relative if
// they have no scheme, e.g. "./index.html". There's no valid equivalent
// to http:index.html.
{"filesystem:http://host/t/path", true, false, "filesystem:http://host/t/path2", true, false, false, NULL},
{"filesystem:http://host/t/path", true, false, "filesystem:https://host/t/path2", true, false, false, NULL},
{"filesystem:http://host/t/path", true, false, "http://host/t/path2", true, false, false, NULL},
{"http://host/t/path", true, false, "filesystem:http://host/t/path2", true, false, false, NULL},
{"filesystem:http://host/t/path", true, false, "./path2", true, true, true, "filesystem:http://host/t/path2"},
{"filesystem:http://host/t/path/", true, false, "path2", true, true, true, "filesystem:http://host/t/path/path2"},
{"filesystem:http://host/t/path", true, false, "filesystem:http:path2", true, false, false, NULL},
// Absolute URLs are still not relative to a non-standard base URL.
{"about:blank", false, false, "http://X/A", true, false, true, ""},
{"about:blank", false, false, "content://content.Provider/", true, false, true, ""},
};
for (size_t i = 0; i < ARRAYSIZE(rel_cases); i++) {
const RelativeCase& cur_case = rel_cases[i];
Parsed parsed;
int base_len = static_cast<int>(strlen(cur_case.base));
if (cur_case.is_base_file)
ParseFileURL(cur_case.base, base_len, &parsed);
else if (cur_case.is_base_hier)
ParseStandardURL(cur_case.base, base_len, &parsed);
else
ParsePathURL(cur_case.base, base_len, false, &parsed);
// First see if it is relative.
int test_len = static_cast<int>(strlen(cur_case.test));
bool is_relative;
Component relative_component;
bool succeed_is_rel = IsRelativeURL(
cur_case.base, parsed, cur_case.test, test_len, cur_case.is_base_hier,
&is_relative, &relative_component);
EXPECT_EQ(cur_case.succeed_relative, succeed_is_rel) <<
"succeed is rel failure on " << cur_case.test;
EXPECT_EQ(cur_case.is_rel, is_relative) <<
"is rel failure on " << cur_case.test;
// Now resolve it.
if (succeed_is_rel && is_relative && cur_case.is_rel) {
std::string resolved;
StdStringCanonOutput output(&resolved);
Parsed resolved_parsed;
bool succeed_resolve = ResolveRelativeURL(
cur_case.base, parsed, cur_case.is_base_file, cur_case.test,
relative_component, NULL, &output, &resolved_parsed);
output.Complete();
EXPECT_EQ(cur_case.succeed_resolve, succeed_resolve);
EXPECT_EQ(cur_case.resolved, resolved) << " on " << cur_case.test;
// Verify that the output parsed structure is the same as parsing a
// the URL freshly.
Parsed ref_parsed;
int resolved_len = static_cast<int>(resolved.size());
if (cur_case.is_base_file) {
ParseFileURL(resolved.c_str(), resolved_len, &ref_parsed);
} else if (cur_case.is_base_hier) {
ParseStandardURL(resolved.c_str(), resolved_len, &ref_parsed);
} else {
ParsePathURL(resolved.c_str(), resolved_len, false, &ref_parsed);
}
EXPECT_TRUE(ParsedIsEqual(ref_parsed, resolved_parsed));
}
}
}
// It used to be when we did a replacement with a long buffer of UTF-16
// characters, we would get invalid data in the URL. This is because the buffer
// it used to hold the UTF-8 data was resized, while some pointers were still
// kept to the old buffer that was removed.
TEST(URLCanonTest, ReplacementOverflow) {
const char src[] = "file:///C:/foo/bar";
int src_len = static_cast<int>(strlen(src));
Parsed parsed;
ParseFileURL(src, src_len, &parsed);
// Override two components, the path with something short, and the query with
// sonething long enough to trigger the bug.
Replacements<base::char16> repl;
base::string16 new_query;
for (int i = 0; i < 4800; i++)
new_query.push_back('a');
base::string16 new_path(WStringToUTF16(L"/foo"));
repl.SetPath(new_path.c_str(), Component(0, 4));
repl.SetQuery(new_query.c_str(),
Component(0, static_cast<int>(new_query.length())));
// Call ReplaceComponents on the string. It doesn't matter if we call it for
// standard URLs, file URLs, etc, since they will go to the same replacement
// function that was buggy.
Parsed repl_parsed;
std::string repl_str;
StdStringCanonOutput repl_output(&repl_str);
ReplaceFileURL(src, parsed, repl, NULL, &repl_output, &repl_parsed);
repl_output.Complete();
// Generate the expected string and check.
std::string expected("file:///foo?");
for (size_t i = 0; i < new_query.length(); i++)
expected.push_back('a');
EXPECT_TRUE(expected == repl_str);
}
} // namespace url