blob: a6a6ab76cda392745af632d2eaae270f55b5b5c6 [file] [log] [blame]
// 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 "url/third_party/mozilla/url_parse.h"
#include "base/macros.h"
#include "testing/gtest/include/gtest/gtest.h"
#include "url/third_party/mozilla/url_parse.h"
#if defined(STARBOARD)
#include "starboard/client_porting/poem/string_poem.h"
#include "starboard/common/string.h"
#include "starboard/types.h"
#endif
// Interesting IE file:isms...
//
// file:/foo/bar file:///foo/bar
// The result here seems totally invalid!?!? This isn't UNC.
//
// file:/
// file:// or any other number of slashes
// IE6 doesn't do anything at all if you click on this link. No error:
// nothing. IE6's history system seems to always color this link, so I'm
// guessing that it maps internally to the empty URL.
//
// C:\ file:///C:/
// / file:///C:/
// /foo file:///C:/foo
// Interestingly, IE treats "/" as an alias for "c:\", which makes sense,
// but is weird to think about on Windows.
//
// file:foo/ file:foo/ (invalid?!?!?)
// file:/foo/ file:///foo/ (invalid?!?!?)
// file://foo/ file://foo/ (UNC to server "foo")
// file:///foo/ file:///foo/ (invalid)
// file:////foo/ file://foo/ (UNC to server "foo")
// Any more than four slashes is also treated as UNC.
//
// file:C:/ file://C:/
// file:/C:/ file://C:/
// The number of slashes after "file:" don't matter if the thing following
// it looks like an absolute drive path. Also, slashes and backslashes are
// equally valid here.
namespace url {
namespace {
// Used for regular URL parse cases.
struct URLParseCase {
const char* input;
const char* scheme;
const char* username;
const char* password;
const char* host;
int port;
const char* path;
const char* query;
const char* ref;
};
// Simpler version of URLParseCase for testing path URLs.
struct PathURLParseCase {
const char* input;
const char* scheme;
const char* path;
};
// Simpler version of URLParseCase for testing mailto URLs.
struct MailtoURLParseCase {
const char* input;
const char* scheme;
const char* path;
const char* query;
};
// More complicated version of URLParseCase for testing filesystem URLs.
struct FileSystemURLParseCase {
const char* input;
const char* inner_scheme;
const char* inner_username;
const char* inner_password;
const char* inner_host;
int inner_port;
const char* inner_path;
const char* path;
const char* query;
const char* ref;
};
bool ComponentMatches(const char* input,
const char* reference,
const Component& component) {
// If the component is nonexistent (length == -1), it should begin at 0.
EXPECT_TRUE(component.len >= 0 || component.len == -1);
// Begin should be valid.
EXPECT_LE(0, component.begin);
// A NULL reference means the component should be nonexistent.
if (!reference)
return component.len == -1;
if (component.len < 0)
return false; // Reference is not NULL but we don't have anything
if (strlen(reference) != static_cast<size_t>(component.len))
return false; // Lengths don't match
// Now check the actual characters.
return strncmp(reference, &input[component.begin], component.len) == 0;
}
void ExpectInvalidComponent(const Component& component) {
EXPECT_EQ(0, component.begin);
EXPECT_EQ(-1, component.len);
}
// Parsed ----------------------------------------------------------------------
TEST(URLParser, Length) {
const char* length_cases[] = {
// One with everything in it.
"http://user:pass@host:99/foo?bar#baz",
// One with nothing in it.
"",
// Working backwards, let's start taking off stuff from the full one.
"http://user:pass@host:99/foo?bar#",
"http://user:pass@host:99/foo?bar",
"http://user:pass@host:99/foo?",
"http://user:pass@host:99/foo",
"http://user:pass@host:99/",
"http://user:pass@host:99",
"http://user:pass@host:",
"http://user:pass@host",
"http://host",
"http://user@",
"http:",
};
for (size_t i = 0; i < arraysize(length_cases); i++) {
int true_length = static_cast<int>(strlen(length_cases[i]));
Parsed parsed;
ParseStandardURL(length_cases[i], true_length, &parsed);
EXPECT_EQ(true_length, parsed.Length());
}
}
TEST(URLParser, CountCharactersBefore) {
struct CountCase {
const char* url;
Parsed::ComponentType component;
bool include_delimiter;
int expected_count;
} count_cases[] = {
// Test each possibility in the case where all components are present.
// 0 1 2
// 0123456789012345678901
{"http://u:p@h:8/p?q#r", Parsed::SCHEME, true, 0},
{"http://u:p@h:8/p?q#r", Parsed::SCHEME, false, 0},
{"http://u:p@h:8/p?q#r", Parsed::USERNAME, true, 7},
{"http://u:p@h:8/p?q#r", Parsed::USERNAME, false, 7},
{"http://u:p@h:8/p?q#r", Parsed::PASSWORD, true, 9},
{"http://u:p@h:8/p?q#r", Parsed::PASSWORD, false, 9},
{"http://u:p@h:8/p?q#r", Parsed::HOST, true, 11},
{"http://u:p@h:8/p?q#r", Parsed::HOST, false, 11},
{"http://u:p@h:8/p?q#r", Parsed::PORT, true, 12},
{"http://u:p@h:8/p?q#r", Parsed::PORT, false, 13},
{"http://u:p@h:8/p?q#r", Parsed::PATH, false, 14},
{"http://u:p@h:8/p?q#r", Parsed::PATH, true, 14},
{"http://u:p@h:8/p?q#r", Parsed::QUERY, true, 16},
{"http://u:p@h:8/p?q#r", Parsed::QUERY, false, 17},
{"http://u:p@h:8/p?q#r", Parsed::REF, true, 18},
{"http://u:p@h:8/p?q#r", Parsed::REF, false, 19},
// Now test when the requested component is missing.
{"http://u:p@h:8/p?", Parsed::REF, true, 17},
{"http://u:p@h:8/p?q", Parsed::REF, true, 18},
{"http://u:p@h:8/p#r", Parsed::QUERY, true, 16},
{"http://u:p@h:8#r", Parsed::PATH, true, 14},
{"http://u:p@h/", Parsed::PORT, true, 12},
{"http://u:p@/", Parsed::HOST, true, 11},
// This case is a little weird. It will report that the password would
// start where the host begins. This is arguably correct, although you
// could also argue that it should start at the '@' sign. Doing it
// starting with the '@' sign is actually harder, so we don't bother.
{"http://u@h/", Parsed::PASSWORD, true, 9},
{"http://h/", Parsed::USERNAME, true, 7},
{"http:", Parsed::USERNAME, true, 5},
{"", Parsed::SCHEME, true, 0},
// Make sure a random component still works when there's nothing there.
{"", Parsed::REF, true, 0},
// File URLs are special with no host, so we test those.
{"file:///c:/foo", Parsed::USERNAME, true, 7},
{"file:///c:/foo", Parsed::PASSWORD, true, 7},
{"file:///c:/foo", Parsed::HOST, true, 7},
{"file:///c:/foo", Parsed::PATH, true, 7},
};
for (size_t i = 0; i < arraysize(count_cases); i++) {
int length = static_cast<int>(strlen(count_cases[i].url));
// Simple test to distinguish file and standard URLs.
Parsed parsed;
if (length > 0 && count_cases[i].url[0] == 'f')
ParseFileURL(count_cases[i].url, length, &parsed);
else
ParseStandardURL(count_cases[i].url, length, &parsed);
int chars_before = parsed.CountCharactersBefore(
count_cases[i].component, count_cases[i].include_delimiter);
EXPECT_EQ(count_cases[i].expected_count, chars_before);
}
}
// Standard --------------------------------------------------------------------
// Input Scheme Usrname Passwd Host Port Path Query Ref
// ------------------------------------ ------- ------- ---------- ------------ --- ---------- ------------ -----
static URLParseCase cases[] = {
// Regular URL with all the parts
{"http://user:pass@foo:21/bar;par?b#c", "http", "user", "pass", "foo", 21, "/bar;par","b", "c"},
// Known schemes should lean towards authority identification
{"http:foo.com", "http", NULL, NULL, "foo.com", -1, NULL, NULL, NULL},
// Spaces!
{"\t :foo.com \n", "", NULL, NULL, "foo.com", -1, NULL, NULL, NULL},
{" foo.com ", NULL, NULL, NULL, "foo.com", -1, NULL, NULL, NULL},
{"a:\t foo.com", "a", NULL, NULL, "\t foo.com", -1, NULL, NULL, NULL},
{"http://f:21/ b ? d # e ", "http", NULL, NULL, "f", 21, "/ b ", " d ", " e"},
// Invalid port numbers should be identified and turned into -2, empty port
// numbers should be -1. Spaces aren't allowed in port numbers
{"http://f:/c", "http", NULL, NULL, "f", -1, "/c", NULL, NULL},
{"http://f:0/c", "http", NULL, NULL, "f", 0, "/c", NULL, NULL},
{"http://f:00000000000000/c", "http", NULL, NULL, "f", 0, "/c", NULL, NULL},
{"http://f:00000000000000000000080/c", "http", NULL, NULL, "f", 80, "/c", NULL, NULL},
{"http://f:b/c", "http", NULL, NULL, "f", -2, "/c", NULL, NULL},
{"http://f: /c", "http", NULL, NULL, "f", -2, "/c", NULL, NULL},
{"http://f:\n/c", "http", NULL, NULL, "f", -2, "/c", NULL, NULL},
{"http://f:fifty-two/c", "http", NULL, NULL, "f", -2, "/c", NULL, NULL},
{"http://f:999999/c", "http", NULL, NULL, "f", -2, "/c", NULL, NULL},
{"http://f: 21 / b ? d # e ", "http", NULL, NULL, "f", -2, "/ b ", " d ", " e"},
// Creative URLs missing key elements
{"", NULL, NULL, NULL, NULL, -1, NULL, NULL, NULL},
{" \t", NULL, NULL, NULL, NULL, -1, NULL, NULL, NULL},
{":foo.com/", "", NULL, NULL, "foo.com", -1, "/", NULL, NULL},
{":foo.com\\", "", NULL, NULL, "foo.com", -1, "\\", NULL, NULL},
{":", "", NULL, NULL, NULL, -1, NULL, NULL, NULL},
{":a", "", NULL, NULL, "a", -1, NULL, NULL, NULL},
{":/", "", NULL, NULL, NULL, -1, NULL, NULL, NULL},
{":\\", "", NULL, NULL, NULL, -1, NULL, NULL, NULL},
{":#", "", NULL, NULL, NULL, -1, NULL, NULL, ""},
{"#", NULL, NULL, NULL, NULL, -1, NULL, NULL, ""},
{"#/", NULL, NULL, NULL, NULL, -1, NULL, NULL, "/"},
{"#\\", NULL, NULL, NULL, NULL, -1, NULL, NULL, "\\"},
{"#;?", NULL, NULL, NULL, NULL, -1, NULL, NULL, ";?"},
{"?", NULL, NULL, NULL, NULL, -1, NULL, "", NULL},
{"/", NULL, NULL, NULL, NULL, -1, NULL, NULL, NULL},
{":23", "", NULL, NULL, "23", -1, NULL, NULL, NULL},
{"/:23", "/", NULL, NULL, "23", -1, NULL, NULL, NULL},
{"//", NULL, NULL, NULL, NULL, -1, NULL, NULL, NULL},
{"::", "", NULL, NULL, NULL, -1, NULL, NULL, NULL},
{"::23", "", NULL, NULL, NULL, 23, NULL, NULL, NULL},
{"foo://", "foo", NULL, NULL, NULL, -1, NULL, NULL, NULL},
// Username/passwords and things that look like them
{"http://a:b@c:29/d", "http", "a", "b", "c", 29, "/d", NULL, NULL},
{"http::@c:29", "http", "", "", "c", 29, NULL, NULL, NULL},
// ... "]" in the password field isn't allowed, but we tolerate it here...
{"http://&a:foo(b]c@d:2/", "http", "&a", "foo(b]c", "d", 2, "/", NULL, NULL},
{"http://::@c@d:2", "http", "", ":@c", "d", 2, NULL, NULL, NULL},
{"http://foo.com:b@d/", "http", "foo.com", "b", "d", -1, "/", NULL, NULL},
{"http://foo.com/\\@", "http", NULL, NULL, "foo.com", -1, "/\\@", NULL, NULL},
{"http:\\\\foo.com\\", "http", NULL, NULL, "foo.com", -1, "\\", NULL, NULL},
{"http:\\\\a\\b:c\\d@foo.com\\", "http", NULL, NULL, "a", -1, "\\b:c\\d@foo.com\\", NULL, NULL},
// Tolerate different numbers of slashes.
{"foo:/", "foo", NULL, NULL, NULL, -1, NULL, NULL, NULL},
{"foo:/bar.com/", "foo", NULL, NULL, "bar.com", -1, "/", NULL, NULL},
{"foo://///////", "foo", NULL, NULL, NULL, -1, NULL, NULL, NULL},
{"foo://///////bar.com/", "foo", NULL, NULL, "bar.com", -1, "/", NULL, NULL},
{"foo:////://///", "foo", NULL, NULL, NULL, -1, "/////", NULL, NULL},
// Raw file paths on Windows aren't handled by the parser.
{"c:/foo", "c", NULL, NULL, "foo", -1, NULL, NULL, NULL},
{"//foo/bar", NULL, NULL, NULL, "foo", -1, "/bar", NULL, NULL},
// Use the first question mark for the query and the ref.
{"http://foo/path;a??e#f#g", "http", NULL, NULL, "foo", -1, "/path;a", "?e", "f#g"},
{"http://foo/abcd?efgh?ijkl", "http", NULL, NULL, "foo", -1, "/abcd", "efgh?ijkl", NULL},
{"http://foo/abcd#foo?bar", "http", NULL, NULL, "foo", -1, "/abcd", NULL, "foo?bar"},
// IPv6, check also interesting uses of colons.
{"[61:24:74]:98", "[61", NULL, NULL, "24:74]", 98, NULL, NULL, NULL},
{"http://[61:27]:98", "http", NULL, NULL, "[61:27]", 98, NULL, NULL, NULL},
{"http:[61:27]/:foo", "http", NULL, NULL, "[61:27]", -1, "/:foo", NULL, NULL},
{"http://[1::2]:3:4", "http", NULL, NULL, "[1::2]:3", 4, NULL, NULL, NULL},
// Partially-complete IPv6 literals, and related cases.
{"http://2001::1", "http", NULL, NULL, "2001:", 1, NULL, NULL, NULL},
{"http://[2001::1", "http", NULL, NULL, "[2001::1", -1, NULL, NULL, NULL},
{"http://2001::1]", "http", NULL, NULL, "2001::1]", -1, NULL, NULL, NULL},
{"http://2001::1]:80", "http", NULL, NULL, "2001::1]", 80, NULL, NULL, NULL},
{"http://[2001::1]", "http", NULL, NULL, "[2001::1]", -1, NULL, NULL, NULL},
{"http://[2001::1]:80", "http", NULL, NULL, "[2001::1]", 80, NULL, NULL, NULL},
{"http://[[::]]", "http", NULL, NULL, "[[::]]", -1, NULL, NULL, NULL},
};
TEST(URLParser, Standard) {
// Declared outside for loop to try to catch cases in init() where we forget
// to reset something that is reset by the constructor.
Parsed parsed;
for (size_t i = 0; i < arraysize(cases); i++) {
const char* url = cases[i].input;
ParseStandardURL(url, static_cast<int>(strlen(url)), &parsed);
int port = ParsePort(url, parsed.port);
EXPECT_TRUE(ComponentMatches(url, cases[i].scheme, parsed.scheme));
EXPECT_TRUE(ComponentMatches(url, cases[i].username, parsed.username));
EXPECT_TRUE(ComponentMatches(url, cases[i].password, parsed.password));
EXPECT_TRUE(ComponentMatches(url, cases[i].host, parsed.host));
EXPECT_EQ(cases[i].port, port);
EXPECT_TRUE(ComponentMatches(url, cases[i].path, parsed.path));
EXPECT_TRUE(ComponentMatches(url, cases[i].query, parsed.query));
EXPECT_TRUE(ComponentMatches(url, cases[i].ref, parsed.ref));
}
}
// PathURL --------------------------------------------------------------------
// Various incarnations of path URLs.
static PathURLParseCase path_cases[] = {
{"", NULL, NULL},
{":", "", NULL},
{":/", "", "/"},
{"/", NULL, "/"},
{" This is \\interesting// \t", NULL, "This is \\interesting// \t"},
{"about:", "about", NULL},
{"about:blank", "about", "blank"},
{" about: blank ", "about", " blank "},
{"javascript :alert(\"He:/l\\l#o?foo\"); ", "javascript ", "alert(\"He:/l\\l#o?foo\"); "},
};
TEST(URLParser, PathURL) {
// Declared outside for loop to try to catch cases in init() where we forget
// to reset something that is reset by the constructor.
Parsed parsed;
for (size_t i = 0; i < arraysize(path_cases); i++) {
const char* url = path_cases[i].input;
ParsePathURL(url, static_cast<int>(strlen(url)), false, &parsed);
EXPECT_TRUE(ComponentMatches(url, path_cases[i].scheme, parsed.scheme))
<< i;
EXPECT_TRUE(ComponentMatches(url, path_cases[i].path, parsed.GetContent()))
<< i;
// The remaining components are never used for path URLs.
ExpectInvalidComponent(parsed.username);
ExpectInvalidComponent(parsed.password);
ExpectInvalidComponent(parsed.host);
ExpectInvalidComponent(parsed.port);
}
}
// Various incarnations of file URLs.
static URLParseCase file_cases[] = {
#ifdef WIN32
{"file:server", "file", NULL, NULL, "server", -1, NULL, NULL, NULL},
{" file: server \t", "file", NULL, NULL, " server",-1, NULL, NULL, NULL},
{"FiLe:c|", "FiLe", NULL, NULL, NULL, -1, "c|", NULL, NULL},
{"FILE:/\\\\/server/file", "FILE", NULL, NULL, "server", -1, "/file", NULL, NULL},
{"file://server/", "file", NULL, NULL, "server", -1, "/", NULL, NULL},
{"file://localhost/c:/", "file", NULL, NULL, NULL, -1, "/c:/", NULL, NULL},
{"file://127.0.0.1/c|\\", "file", NULL, NULL, NULL, -1, "/c|\\", NULL, NULL},
{"file:/", "file", NULL, NULL, NULL, -1, NULL, NULL, NULL},
{"file:", "file", NULL, NULL, NULL, -1, NULL, NULL, NULL},
// If there is a Windows drive letter, treat any number of slashes as the
// path part.
{"file:c:\\fo\\b", "file", NULL, NULL, NULL, -1, "c:\\fo\\b", NULL, NULL},
{"file:/c:\\foo/bar", "file", NULL, NULL, NULL, -1, "/c:\\foo/bar",NULL, NULL},
{"file://c:/f\\b", "file", NULL, NULL, NULL, -1, "/c:/f\\b", NULL, NULL},
{"file:///C:/foo", "file", NULL, NULL, NULL, -1, "/C:/foo", NULL, NULL},
{"file://///\\/\\/c:\\f\\b", "file", NULL, NULL, NULL, -1, "/c:\\f\\b", NULL, NULL},
// If there is not a drive letter, we should treat is as UNC EXCEPT for
// three slashes, which we treat as a Unix style path.
{"file:server/file", "file", NULL, NULL, "server", -1, "/file", NULL, NULL},
{"file:/server/file", "file", NULL, NULL, "server", -1, "/file", NULL, NULL},
{"file://server/file", "file", NULL, NULL, "server", -1, "/file", NULL, NULL},
{"file:///server/file", "file", NULL, NULL, NULL, -1, "/server/file",NULL, NULL},
{"file://\\server/file", "file", NULL, NULL, NULL, -1, "\\server/file",NULL, NULL},
{"file:////server/file", "file", NULL, NULL, "server", -1, "/file", NULL, NULL},
// Queries and refs are valid for file URLs as well.
{"file:///C:/foo.html?#", "file", NULL, NULL, NULL, -1, "/C:/foo.html", "", ""},
{"file:///C:/foo.html?query=yes#ref", "file", NULL, NULL, NULL, -1, "/C:/foo.html", "query=yes", "ref"},
#else // WIN32
// No slashes.
{"file:", "file", NULL, NULL, NULL, -1, NULL, NULL, NULL},
{"file:path", "file", NULL, NULL, NULL, -1, "path", NULL, NULL},
{"file:path/", "file", NULL, NULL, NULL, -1, "path/", NULL, NULL},
{"file:path/f.txt", "file", NULL, NULL, NULL, -1, "path/f.txt", NULL, NULL},
// One slash.
{"file:/", "file", NULL, NULL, NULL, -1, "/", NULL, NULL},
{"file:/path", "file", NULL, NULL, NULL, -1, "/path", NULL, NULL},
{"file:/path/", "file", NULL, NULL, NULL, -1, "/path/", NULL, NULL},
{"file:/path/f.txt", "file", NULL, NULL, NULL, -1, "/path/f.txt", NULL, NULL},
// Two slashes.
{"file://", "file", NULL, NULL, NULL, -1, NULL, NULL, NULL},
{"file://server", "file", NULL, NULL, "server", -1, NULL, NULL, NULL},
{"file://server/", "file", NULL, NULL, "server", -1, "/", NULL, NULL},
{"file://server/f.txt", "file", NULL, NULL, "server", -1, "/f.txt", NULL, NULL},
// Three slashes.
{"file:///", "file", NULL, NULL, NULL, -1, "/", NULL, NULL},
{"file:///path", "file", NULL, NULL, NULL, -1, "/path", NULL, NULL},
{"file:///path/", "file", NULL, NULL, NULL, -1, "/path/", NULL, NULL},
{"file:///path/f.txt", "file", NULL, NULL, NULL, -1, "/path/f.txt", NULL, NULL},
// More than three slashes.
{"file:////", "file", NULL, NULL, NULL, -1, "/", NULL, NULL},
{"file:////path", "file", NULL, NULL, NULL, -1, "/path", NULL, NULL},
{"file:////path/", "file", NULL, NULL, NULL, -1, "/path/", NULL, NULL},
{"file:////path/f.txt", "file", NULL, NULL, NULL, -1, "/path/f.txt", NULL, NULL},
// Schemeless URLs
{"path/f.txt", NULL, NULL, NULL, NULL, -1, "path/f.txt", NULL, NULL},
{"path:80/f.txt", "path", NULL, NULL, NULL, -1, "80/f.txt", NULL, NULL},
{"path/f.txt:80", "path/f.txt",NULL, NULL, NULL, -1, "80", NULL, NULL}, // Wrong.
{"/path/f.txt", NULL, NULL, NULL, NULL, -1, "/path/f.txt", NULL, NULL},
{"/path:80/f.txt", NULL, NULL, NULL, NULL, -1, "/path:80/f.txt",NULL, NULL},
{"/path/f.txt:80", NULL, NULL, NULL, NULL, -1, "/path/f.txt:80",NULL, NULL},
{"//server/f.txt", NULL, NULL, NULL, "server", -1, "/f.txt", NULL, NULL},
{"//server:80/f.txt", NULL, NULL, NULL, "server:80",-1, "/f.txt", NULL, NULL},
{"//server/f.txt:80", NULL, NULL, NULL, "server", -1, "/f.txt:80", NULL, NULL},
{"///path/f.txt", NULL, NULL, NULL, NULL, -1, "/path/f.txt", NULL, NULL},
{"///path:80/f.txt", NULL, NULL, NULL, NULL, -1, "/path:80/f.txt",NULL, NULL},
{"///path/f.txt:80", NULL, NULL, NULL, NULL, -1, "/path/f.txt:80",NULL, NULL},
{"////path/f.txt", NULL, NULL, NULL, NULL, -1, "/path/f.txt", NULL, NULL},
{"////path:80/f.txt", NULL, NULL, NULL, NULL, -1, "/path:80/f.txt",NULL, NULL},
{"////path/f.txt:80", NULL, NULL, NULL, NULL, -1, "/path/f.txt:80",NULL, NULL},
// Queries and refs are valid for file URLs as well.
{"file:///foo.html?#", "file", NULL, NULL, NULL, -1, "/foo.html", "", ""},
{"file:///foo.html?q=y#ref", "file", NULL, NULL, NULL, -1, "/foo.html", "q=y", "ref"},
#endif // WIN32
};
TEST(URLParser, ParseFileURL) {
// Declared outside for loop to try to catch cases in init() where we forget
// to reset something that is reset by the construtor.
Parsed parsed;
for (size_t i = 0; i < arraysize(file_cases); i++) {
const char* url = file_cases[i].input;
ParseFileURL(url, static_cast<int>(strlen(url)), &parsed);
int port = ParsePort(url, parsed.port);
EXPECT_TRUE(ComponentMatches(url, file_cases[i].scheme, parsed.scheme))
<< " for case #" << i << " [" << url << "] "
<< parsed.scheme.begin << ", " << parsed.scheme.len;
EXPECT_TRUE(ComponentMatches(url, file_cases[i].username, parsed.username))
<< " for case #" << i << " [" << url << "] "
<< parsed.username.begin << ", " << parsed.username.len;
EXPECT_TRUE(ComponentMatches(url, file_cases[i].password, parsed.password))
<< " for case #" << i << " [" << url << "] "
<< parsed.password.begin << ", " << parsed.password.len;
EXPECT_TRUE(ComponentMatches(url, file_cases[i].host, parsed.host))
<< " for case #" << i << " [" << url << "] "
<< parsed.host.begin << ", " << parsed.host.len;
EXPECT_EQ(file_cases[i].port, port)
<< " for case #" << i << " [ " << url << "] " << port;
EXPECT_TRUE(ComponentMatches(url, file_cases[i].path, parsed.path))
<< " for case #" << i << " [" << url << "] "
<< parsed.path.begin << ", " << parsed.path.len;
EXPECT_TRUE(ComponentMatches(url, file_cases[i].query, parsed.query))
<< " for case #" << i << " [" << url << "] "
<< parsed.query.begin << ", " << parsed.query.len;
EXPECT_TRUE(ComponentMatches(url, file_cases[i].ref, parsed.ref))
<< " for case #" << i << " [ "<< url << "] "
<< parsed.query.begin << ", " << parsed.scheme.len;
}
}
TEST(URLParser, ExtractFileName) {
struct FileCase {
const char* input;
const char* expected;
} file_cases[] = {
{"http://www.google.com", NULL},
{"http://www.google.com/", ""},
{"http://www.google.com/search", "search"},
{"http://www.google.com/search/", ""},
{"http://www.google.com/foo/bar.html?baz=22", "bar.html"},
{"http://www.google.com/foo/bar.html#ref", "bar.html"},
{"http://www.google.com/search/;param", ""},
{"http://www.google.com/foo/bar.html;param#ref", "bar.html"},
{"http://www.google.com/foo/bar.html;foo;param#ref", "bar.html"},
{"http://www.google.com/foo/bar.html?query#ref", "bar.html"},
{"http://www.google.com/foo;/bar.html", "bar.html"},
{"http://www.google.com/foo;/", ""},
{"http://www.google.com/foo;", "foo"},
{"http://www.google.com/;", ""},
{"http://www.google.com/foo;bar;html", "foo"},
};
for (size_t i = 0; i < arraysize(file_cases); i++) {
const char* url = file_cases[i].input;
int len = static_cast<int>(strlen(url));
Parsed parsed;
ParseStandardURL(url, len, &parsed);
Component file_name;
ExtractFileName(url, parsed.path, &file_name);
EXPECT_TRUE(ComponentMatches(url, file_cases[i].expected, file_name));
}
}
// Returns true if the parameter with index |parameter| in the given URL's
// query string. The expected key can be NULL to indicate no such key index
// should exist. The parameter number is 1-based.
static bool NthParameterIs(const char* url,
int parameter,
const char* expected_key,
const char* expected_value) {
Parsed parsed;
ParseStandardURL(url, static_cast<int>(strlen(url)), &parsed);
Component query = parsed.query;
for (int i = 1; i <= parameter; i++) {
Component key, value;
if (!ExtractQueryKeyValue(url, &query, &key, &value)) {
if (parameter >= i && !expected_key)
return true; // Expected nonexistent key, got one.
return false; // Not enough keys.
}
if (i == parameter) {
if (!expected_key)
return false;
if (strncmp(&url[key.begin], expected_key, key.len) != 0)
return false;
if (strncmp(&url[value.begin], expected_value, value.len) != 0)
return false;
return true;
}
}
return expected_key == NULL; // We didn't find that many parameters.
}
TEST(URLParser, ExtractQueryKeyValue) {
EXPECT_TRUE(NthParameterIs("http://www.google.com", 1, NULL, NULL));
// Basic case.
char a[] = "http://www.google.com?arg1=1&arg2=2&bar";
EXPECT_TRUE(NthParameterIs(a, 1, "arg1", "1"));
EXPECT_TRUE(NthParameterIs(a, 2, "arg2", "2"));
EXPECT_TRUE(NthParameterIs(a, 3, "bar", ""));
EXPECT_TRUE(NthParameterIs(a, 4, NULL, NULL));
// Empty param at the end.
char b[] = "http://www.google.com?foo=bar&";
EXPECT_TRUE(NthParameterIs(b, 1, "foo", "bar"));
EXPECT_TRUE(NthParameterIs(b, 2, NULL, NULL));
// Empty param at the beginning.
char c[] = "http://www.google.com?&foo=bar";
EXPECT_TRUE(NthParameterIs(c, 1, "", ""));
EXPECT_TRUE(NthParameterIs(c, 2, "foo", "bar"));
EXPECT_TRUE(NthParameterIs(c, 3, NULL, NULL));
// Empty key with value.
char d[] = "http://www.google.com?=foo";
EXPECT_TRUE(NthParameterIs(d, 1, "", "foo"));
EXPECT_TRUE(NthParameterIs(d, 2, NULL, NULL));
// Empty value with key.
char e[] = "http://www.google.com?foo=";
EXPECT_TRUE(NthParameterIs(e, 1, "foo", ""));
EXPECT_TRUE(NthParameterIs(e, 2, NULL, NULL));
// Empty key and values.
char f[] = "http://www.google.com?&&==&=";
EXPECT_TRUE(NthParameterIs(f, 1, "", ""));
EXPECT_TRUE(NthParameterIs(f, 2, "", ""));
EXPECT_TRUE(NthParameterIs(f, 3, "", "="));
EXPECT_TRUE(NthParameterIs(f, 4, "", ""));
EXPECT_TRUE(NthParameterIs(f, 5, NULL, NULL));
}
// MailtoURL --------------------------------------------------------------------
static MailtoURLParseCase mailto_cases[] = {
//|input |scheme |path |query
{"mailto:foo@gmail.com", "mailto", "foo@gmail.com", NULL},
{" mailto: to \t", "mailto", " to", NULL},
{"mailto:addr1%2C%20addr2 ", "mailto", "addr1%2C%20addr2", NULL},
{"Mailto:addr1, addr2 ", "Mailto", "addr1, addr2", NULL},
{"mailto:addr1:addr2 ", "mailto", "addr1:addr2", NULL},
{"mailto:?to=addr1,addr2", "mailto", NULL, "to=addr1,addr2"},
{"mailto:?to=addr1%2C%20addr2", "mailto", NULL, "to=addr1%2C%20addr2"},
{"mailto:addr1?to=addr2", "mailto", "addr1", "to=addr2"},
{"mailto:?body=#foobar#", "mailto", NULL, "body=#foobar#",},
{"mailto:#?body=#foobar#", "mailto", "#", "body=#foobar#"},
};
TEST(URLParser, MailtoUrl) {
// Declared outside for loop to try to catch cases in init() where we forget
// to reset something that is reset by the constructor.
Parsed parsed;
for (size_t i = 0; i < arraysize(mailto_cases); ++i) {
const char* url = mailto_cases[i].input;
ParseMailtoURL(url, static_cast<int>(strlen(url)), &parsed);
int port = ParsePort(url, parsed.port);
EXPECT_TRUE(ComponentMatches(url, mailto_cases[i].scheme, parsed.scheme));
EXPECT_TRUE(ComponentMatches(url, mailto_cases[i].path, parsed.path));
EXPECT_TRUE(ComponentMatches(url, mailto_cases[i].query, parsed.query));
EXPECT_EQ(PORT_UNSPECIFIED, port);
// The remaining components are never used for mailto URLs.
ExpectInvalidComponent(parsed.username);
ExpectInvalidComponent(parsed.password);
ExpectInvalidComponent(parsed.port);
ExpectInvalidComponent(parsed.ref);
}
}
// Various incarnations of filesystem URLs.
static FileSystemURLParseCase filesystem_cases[] = {
// Regular URL with all the parts
{"filesystem:http://user:pass@foo:21/temporary/bar;par?b#c", "http", "user", "pass", "foo", 21, "/temporary", "/bar;par", "b", "c"},
{"filesystem:https://foo/persistent/bar;par/", "https", NULL, NULL, "foo", -1, "/persistent", "/bar;par/", NULL, NULL},
{"filesystem:file:///persistent/bar;par/", "file", NULL, NULL, NULL, -1, "/persistent", "/bar;par/", NULL, NULL},
{"filesystem:file:///persistent/bar;par/?query#ref", "file", NULL, NULL, NULL, -1, "/persistent", "/bar;par/", "query", "ref"},
{"filesystem:file:///persistent", "file", NULL, NULL, NULL, -1, "/persistent", "", NULL, NULL},
};
TEST(URLParser, FileSystemURL) {
// Declared outside for loop to try to catch cases in init() where we forget
// to reset something that is reset by the constructor.
Parsed parsed;
for (size_t i = 0; i < arraysize(filesystem_cases); i++) {
const FileSystemURLParseCase* parsecase = &filesystem_cases[i];
const char* url = parsecase->input;
ParseFileSystemURL(url, static_cast<int>(strlen(url)), &parsed);
EXPECT_TRUE(ComponentMatches(url, "filesystem", parsed.scheme));
EXPECT_EQ(!parsecase->inner_scheme, !parsed.inner_parsed());
// Only check the inner_parsed if there is one.
if (parsed.inner_parsed()) {
EXPECT_TRUE(ComponentMatches(url, parsecase->inner_scheme,
parsed.inner_parsed()->scheme));
EXPECT_TRUE(ComponentMatches(url, parsecase->inner_username,
parsed.inner_parsed()->username));
EXPECT_TRUE(ComponentMatches(url, parsecase->inner_password,
parsed.inner_parsed()->password));
EXPECT_TRUE(ComponentMatches(url, parsecase->inner_host,
parsed.inner_parsed()->host));
int port = ParsePort(url, parsed.inner_parsed()->port);
EXPECT_EQ(parsecase->inner_port, port);
// The remaining components are never used for filesystem URLs.
ExpectInvalidComponent(parsed.inner_parsed()->query);
ExpectInvalidComponent(parsed.inner_parsed()->ref);
}
EXPECT_TRUE(ComponentMatches(url, parsecase->path, parsed.path));
EXPECT_TRUE(ComponentMatches(url, parsecase->query, parsed.query));
EXPECT_TRUE(ComponentMatches(url, parsecase->ref, parsed.ref));
// The remaining components are never used for filesystem URLs.
ExpectInvalidComponent(parsed.username);
ExpectInvalidComponent(parsed.password);
ExpectInvalidComponent(parsed.host);
ExpectInvalidComponent(parsed.port);
}
}
} // namespace
} // namespace url