blob: fcb1d8816f47e6ab31ae21643363f8335ac94d1e [file] [log] [blame]
// Copyright 2014 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 "components/url_formatter/elide_url.h"
#include <stddef.h>
#include "base/cxx17_backports.h"
#include "base/logging.h"
#include "base/run_loop.h"
#include "base/strings/escape.h"
#include "base/strings/utf_string_conversions.h"
#include "base/test/task_environment.h"
#include "build/build_config.h"
#include "components/url_formatter/url_formatter.h"
#include "testing/gtest/include/gtest/gtest.h"
#include "ui/gfx/font_list.h"
#include "ui/gfx/text_elider.h"
#include "ui/gfx/text_utils.h"
#include "url/gurl.h"
#include "url/origin.h"
namespace {
struct Testcase {
const std::string input;
const std::string output;
};
struct ProgressiveTestcase {
const std::string input;
const std::vector<std::string> output;
};
struct UrlComponent {
url::Parsed::ComponentType type;
int begin;
int len;
};
struct ParsingTestcase {
const std::string input;
const std::string output;
const std::vector<UrlComponent> components;
};
#if !BUILDFLAG(IS_ANDROID)
// Returns the width of a utf8 or utf16 string using default UI font, or the
// provided |font_list|.
float GetWidth(const std::string& utf8,
const gfx::FontList& font_list = gfx::FontList()) {
return gfx::GetStringWidthF(base::UTF8ToUTF16(utf8), font_list);
}
float GetWidth(const std::u16string& utf16,
const gfx::FontList& font_list = gfx::FontList()) {
return gfx::GetStringWidthF(utf16, font_list);
}
// Verify that one or more URLs passes through an explicit sequence of elided
// strings as available space progressively decreases. This helps ensure that
// transitional corner cases are handled properly. To be tolerant of
// character-width variation across platforms, the test allows a limited number
// of expected strings to be skipped mid-run. The first and last expected
// strings must be matched. If the algorithm produces a string that isn't in the
// expected string list, the test fill fail. Example test expectations:
//
// google.com/intl/en/.../ads/ <-- Must match.
// google.com/intl/.../ads/
// google.com/.../ads/
// google.com/intl... <- Elider can skip this, in case the 'l' does not fit.
// google.com/int...
// google.com/in... <- Must match.
//
void RunProgressiveElisionTest(
const std::vector<ProgressiveTestcase>& testcases) {
const gfx::FontList font_list;
for (const auto& testcase : testcases) {
SCOPED_TRACE("Eliding " + testcase.input);
const GURL url(testcase.input);
// Occasionally, a parsed URL can grow in length before elision, such as
// when parsing a Windows file path with missing slashes.
ASSERT_FALSE(testcase.output.empty());
float width = std::max(GetWidth(testcase.input, font_list),
GetWidth(testcase.output.front(), font_list));
// Ideally, this test would iterate through all available field widths on a
// per-pixel basis, but this is slow. Instead, compute the next input field
// width as slightly less than the previous elided string. This approach
// misses coverage in cases where a smaller available width generates a
// longer string than some other larger available width, but the tradeoff
// feels acceptable.
int mismatches = 0;
const int kMaxConsecutiveMismatches = 3;
for (size_t i = 0; i < testcase.output.size(); i++) {
const auto& expected = testcase.output[i];
std::u16string expected_utf16 = base::UTF8ToUTF16(expected);
std::u16string elided = url_formatter::ElideUrl(url, font_list, width);
if (expected_utf16 != elided) {
if (i > 0 && i < testcase.output.size() - 1 &&
mismatches < kMaxConsecutiveMismatches) {
mismatches++;
continue;
}
EXPECT_EQ(expected_utf16, elided);
break;
}
mismatches = 0;
float new_width = GetWidth(elided, font_list);
// Elision rounds fractional available widths up.
EXPECT_LE(new_width, std::ceil(width)) << " at " << elided;
width = new_width - 1.0f;
}
}
}
void RunElisionTest(const std::vector<Testcase>& testcases) {
const gfx::FontList font_list;
for (const auto& testcase : testcases) {
SCOPED_TRACE("Eliding " + testcase.input);
const GURL url(testcase.input);
const float available_width = GetWidth(testcase.output, font_list);
EXPECT_EQ(base::UTF8ToUTF16(testcase.output),
url_formatter::ElideUrl(url, font_list, available_width));
}
}
// Test eliding of commonplace URLs.
TEST(TextEliderTest, TestGeneralEliding) {
const std::string kEllipsisStr(gfx::kEllipsis);
const std::vector<ProgressiveTestcase> progressive_testcases = {
// Elide a non-www URL (www URLs are handled differently). In this first
// case, elide down to nothing to test the terminal cases.
{"http://xyz.google.com/foo?bar",
{
/* clang-format off */
"xyz.google.com/foo?bar",
"xyz.google.com/foo?b" + kEllipsisStr,
"xyz.google.com/foo?" + kEllipsisStr,
"xyz.google.com/foo" + kEllipsisStr,
"xyz.google.com/fo" + kEllipsisStr,
"xyz.google.com/f" + kEllipsisStr,
kEllipsisStr + "google.com/foo" + kEllipsisStr,
kEllipsisStr + "google.com/fo" + kEllipsisStr,
kEllipsisStr + "google.com/f" + kEllipsisStr,
kEllipsisStr + "google.com/" + kEllipsisStr,
kEllipsisStr + "google.com" + kEllipsisStr,
kEllipsisStr + "google.co" + kEllipsisStr,
kEllipsisStr + "google.c" + kEllipsisStr,
kEllipsisStr + "google." + kEllipsisStr,
kEllipsisStr + "google" + kEllipsisStr,
kEllipsisStr + "googl" + kEllipsisStr,
kEllipsisStr + "goog" + kEllipsisStr,
kEllipsisStr + "goo" + kEllipsisStr,
kEllipsisStr + "go" + kEllipsisStr,
kEllipsisStr + "g" + kEllipsisStr,
kEllipsisStr + kEllipsisStr,
kEllipsisStr,
""
/* clang-format on */
}},
// The trailing directory name is preserved
{"http://www.google.com/intl/en/ads/",
{
/* clang-format off */
"www.google.com/intl/en/ads/",
"google.com/intl/en/ads/",
"google.com/intl/" + kEllipsisStr + "/ads/",
"google.com/" + kEllipsisStr + "/ads/",
"google.com/" + kEllipsisStr + "/ad" + kEllipsisStr,
"google.com/" + kEllipsisStr + "/a" + kEllipsisStr,
"google.com/intl/e" + kEllipsisStr,
"google.com/intl/" + kEllipsisStr,
"google.com/intl" + kEllipsisStr,
"google.com/int" + kEllipsisStr,
"google.com/in" + kEllipsisStr,
"google.com/i" + kEllipsisStr,
"google.com/" + kEllipsisStr,
"google.com" + kEllipsisStr,
"google.co" + kEllipsisStr,
"google.c" + kEllipsisStr,
"google." + kEllipsisStr,
"google" + kEllipsisStr,
"googl" + kEllipsisStr,
/* clang-format on */
}},
// Subdomain is completely elided if the last path element doesn't fit.
{"https://subdomain.foo.com/bar/filename.html",
{
/* clang-format off */
"https://subdomain.foo.com/bar/filename.html",
"subdomain.foo.com/bar/filename.html",
"subdomain.foo.com/" + kEllipsisStr + "/filename.html",
kEllipsisStr + "foo.com/bar/filename.html",
kEllipsisStr + "foo.com/" + kEllipsisStr + "/filename.html",
/* clang-format on */
}},
// Path eliding works when a query is present.
{"http://www.g.com/subdir/ads/?query",
{
/* clang-format off */
"www.g.com/subdir/ads/?query",
"www.g.com/subdir/ads/?que" + kEllipsisStr,
"www.g.com/subdir/ads/?qu" + kEllipsisStr,
"www.g.com/subdir/ads/?q" + kEllipsisStr,
"www.g.com/subdir/ads/?" + kEllipsisStr,
"www.g.com/subdir/ads/" + kEllipsisStr,
"www.g.com/subdir/ads" + kEllipsisStr,
"www.g.com/subdir/ad" + kEllipsisStr,
/* clang-format on */
}},
};
RunProgressiveElisionTest(progressive_testcases);
}
// When there is very little space available, the elision code will shorten
// both path AND file name to an ellipsis - ".../...". To avoid this result,
// there is a hack in place that simply treats them as one string in this
// case.
TEST(TextEliderTest, TestTrailingEllipsisSlashEllipsisHack) {
const std::string kEllipsisStr(gfx::kEllipsis);
// Very little space, would cause double ellipsis.
gfx::FontList font_list;
GURL url("http://battersbox.com/directory/foo/peter_paul_and_mary.html");
float available_width = GetWidth(
"battersbox.com/" + kEllipsisStr + "/" + kEllipsisStr, font_list);
// Create the expected string, after elision. Depending on font size, the
// directory might become /dir... or /di... or/d... - it never should be
// shorter than that. (If it is, the font considers d... to be longer
// than .../... - that should never happen).
ASSERT_GT(GetWidth(kEllipsisStr + "/" + kEllipsisStr, font_list),
GetWidth("d" + kEllipsisStr, font_list));
GURL long_url("http://battersbox.com/directorynameisreallylongtoforcetrunc");
std::u16string expected =
url_formatter::ElideUrl(long_url, font_list, available_width);
// Ensure that the expected result still contains part of the directory name.
ASSERT_GT(expected.length(), std::string("battersbox.com/d").length());
EXPECT_EQ(expected, url_formatter::ElideUrl(url, font_list, available_width));
// Regression test for https://crbug.com/756717. An empty path, eliding to a
// width in between the full domain ("www.angelfire.lycos.com") and a bit
// longer than the ETLD+1 ("…lycos.com…/…UV"). This previously crashed due to
// the path being empty.
url = GURL("http://www.angelfire.lycos.com/");
available_width = GetWidth(kEllipsisStr + "angelfire.lycos.com", font_list);
EXPECT_EQ(base::UTF8ToUTF16(kEllipsisStr + "lycos.com"),
url_formatter::ElideUrl(url, font_list, available_width));
// More space available - elide directories, partially elide filename.
const std::vector<Testcase> testcases = {
{"http://battersbox.com/directory/foo/peter_paul_and_mary.html",
"battersbox.com/" + kEllipsisStr + "/peter" + kEllipsisStr},
};
RunElisionTest(testcases);
}
// Test eliding of empty strings, URLs with ports, passwords, queries, etc.
TEST(TextEliderTest, TestElisionSpecialCases) {
#if BUILDFLAG(IS_WIN)
// Needed to bypass DCHECK in GetFallbackFont.
base::test::SingleThreadTaskEnvironment task_environment(
base::test::SingleThreadTaskEnvironment::MainThreadType::UI);
#endif
const std::string kEllipsisStr(gfx::kEllipsis);
const std::vector<Testcase> testcases = {
// URL with "www" subdomain (gets removed specially).
{"http://www.google.com/foo?bar", "www.google.com/foo?bar"},
{"http://www.google.com/foo?bar", "google.com/foo?bar"},
// URL with no path.
{"http://xyz.google.com", kEllipsisStr + "google.com"},
{"https://xyz.google.com", kEllipsisStr + "google.com"},
{"http://a.b.com/pathname/c?d", "a.b.com/" + kEllipsisStr + "/c?d"},
{"", ""},
{"http://foo.bar..example.com...hello/test/filename.html",
"foo.bar..example.com...hello/" + kEllipsisStr + "/filename.html"},
{"http://foo.bar../", "foo.bar.."},
{"http://xn--1lq90i.cn/foo", "\xe5\x8c\x97\xe4\xba\xac.cn/foo"},
{"http://me:mypass@secrethost.com:99/foo?bar#baz",
"secrethost.com:99/foo?bar#baz"},
{"http://me:mypass@ss%xxfdsf.com/foo", "ss%25xxfdsf.com/foo"},
{"mailto:elgoato@elgoato.com", "mailto:elgoato@elgoato.com"},
{"javascript:click(0)", "javascript:click(0)"},
{"https://chess.eecs.berkeley.edu:4430/login/arbitfilename",
"chess.eecs.berkeley.edu:4430/login/arbitfilename"},
{"https://chess.eecs.berkeley.edu:4430/login/arbitfilename",
kEllipsisStr + "berkeley.edu:4430/" + kEllipsisStr + "/arbitfilename"},
// Unescaping.
{"http://www/%E4%BD%A0%E5%A5%BD?"
"q=%E4%BD%A0%E5%A5%BD#\xe4\xbd\xa0\xe4\xbd\xa0\xe4\xbd\xa0",
"www/\xe4\xbd\xa0\xe5\xa5\xbd?q=\xe4\xbd\xa0\xe5\xa5\xbd#\xe4\xbd\xa0" +
kEllipsisStr},
// Invalid unescaping for path. The ref will always be valid UTF-8. We
// don't bother to do too many edge cases, since these are handled by the
// escaper unittest.
{"http://www/%E4%A0%E5%A5%BD?q=%E4%BD%A0%E5%A5%BD#\xe4\xbd\xa0",
"www/%E4%A0%E5%A5%BD?q=\xe4\xbd\xa0\xe5\xa5\xbd#\xe4\xbd\xa0"},
};
RunElisionTest(testcases);
}
// Test eliding of file: URLs.
TEST(TextEliderTest, TestFileURLEliding) {
const std::string kEllipsisStr(gfx::kEllipsis);
const std::vector<ProgressiveTestcase> progressive_testcases = {
{"file:///C:/path1/path2/path3/filename",
{
/* clang-format off */
"file:///C:/path1/path2/path3/filename",
"/C:/path1/path2/path3/filename",
"C:/path1/path2/path3/filename",
"C:/path1/path2/" + kEllipsisStr + "/filename",
/* clang-format on */
}},
// GURL parses "file:///C:path" differently on windows than it does on posix.
#if BUILDFLAG(IS_WIN)
{"file:///C:path1/path2/path3/filename",
{
/* clang-format off */
"C:/path1/path2/path3/filename",
"C:/path1/path2/" + kEllipsisStr + "/filename",
"C:/path1/" + kEllipsisStr + "/filename",
"C:/" + kEllipsisStr + "/filename",
/* clang-format on */
}},
#endif // BUILDFLAG(IS_WIN)
{"file://filer/foo/bar/file",
{
/* clang-format off */
"file://filer/foo/bar/file",
"filer/foo/bar/file",
"filer/foo/" + kEllipsisStr + "/file",
"filer/" + kEllipsisStr + "/file",
"filer/foo" + kEllipsisStr,
"filer/fo" + kEllipsisStr,
"filer/f" + kEllipsisStr,
"filer/" + kEllipsisStr,
"filer" + kEllipsisStr,
"file" + kEllipsisStr,
/* clang-format on */
}},
};
RunProgressiveElisionTest(progressive_testcases);
const std::vector<Testcase> testcases = {
// Eliding file URLs with nothing after the ':' shouldn't crash.
{"file:///aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa:", "aaa" + kEllipsisStr},
{"file:///aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa:/", "aaa" + kEllipsisStr},
};
RunElisionTest(testcases);
}
TEST(TextEliderTest, TestHostEliding) {
const std::string kEllipsisStr(gfx::kEllipsis);
Testcase testcases[] = {
{"http://google.com", "google.com"},
{"http://reallyreallyreallylongdomainname.com",
"reallyreallyreallylongdomainname.com"},
{"http://foo", "foo"},
{"http://foo.bar", "foo.bar"},
{"http://subdomain.google.com", kEllipsisStr + ".google.com"},
{"http://a.b.c.d.e.f.com", kEllipsisStr + "f.com"},
{"http://subdomain.foo.bar", kEllipsisStr + "in.foo.bar"},
{"http://subdomain.reallylongdomainname.com",
kEllipsisStr + "ain.reallylongdomainname.com"},
{"http://a.b.c.d.e.f.com", kEllipsisStr + ".e.f.com"},
// IDN - Greek alpha.beta.gamma.delta.epsilon.zeta.com
{"http://xn--mxa.xn--nxa.xn--oxa.xn--pxa.xn--qxa.xn--rxa.com",
kEllipsisStr + ".\xCE\xB5.\xCE\xB6.com"},
};
for (const auto& testcase : testcases) {
const float available_width = GetWidth(testcase.output);
EXPECT_EQ(base::UTF8ToUTF16(testcase.output),
url_formatter::ElideHost(GURL(testcase.input), gfx::FontList(),
available_width));
}
// Trying to elide to a really short length will still keep the full TLD+1
EXPECT_EQ(u"google.com", url_formatter::ElideHost(GURL("http://google.com"),
gfx::FontList(), 2));
EXPECT_EQ(base::UTF8ToUTF16(kEllipsisStr + ".google.com"),
url_formatter::ElideHost(GURL("http://subdomain.google.com"),
gfx::FontList(), 2));
EXPECT_EQ(u"foo.bar", url_formatter::ElideHost(GURL("http://foo.bar"),
gfx::FontList(), 2));
}
#endif // !BUILDFLAG(IS_ANDROID)
struct OriginTestData {
const char* const description;
const char* const input;
const char16_t* const output;
const char16_t* const output_omit_web_scheme;
const char16_t* const output_omit_cryptographic_scheme;
};
// Common test data for both FormatUrlForSecurityDisplay() and
// FormatOriginForSecurityDisplay()
const OriginTestData common_tests[] = {
{"Empty URL", "", u"", u"", u""},
{"HTTP URL", "http://www.google.com/", u"http://www.google.com",
u"www.google.com", u"http://www.google.com"},
{"HTTPS URL", "https://www.google.com/", u"https://www.google.com",
u"www.google.com", u"www.google.com"},
{"Standard HTTP port", "http://www.google.com:80/",
u"http://www.google.com", u"www.google.com", u"http://www.google.com"},
{"Standard HTTPS port", "https://www.google.com:443/",
u"https://www.google.com", u"www.google.com", u"www.google.com"},
{"Standard HTTP port, IDN Chinese",
"http://\xe4\xb8\xad\xe5\x9b\xbd.icom.museum:80",
u"http://中国.icom.museum", u"中国.icom.museum",
u"http://中国.icom.museum"},
{"HTTP URL, IDN Hebrew (RTL)",
"http://"
"\xd7\x90\xd7\x99\xd7\xa7\xd7\x95\xd7\xb4\xd7\x9d."
"\xd7\x99\xd7\xa9\xd7\xa8\xd7\x90\xd7\x9c.museum/",
u"http://xn--4dbklr2c8d.xn--4dbrk0ce.museum",
u"xn--4dbklr2c8d.xn--4dbrk0ce.museum",
u"http://xn--4dbklr2c8d.xn--4dbrk0ce.museum"},
{"HTTP URL with query string, IDN Arabic (RTL)",
"http://\xd9\x85\xd8\xb5\xd8\xb1.icom.museum/foo.html?yes=no",
u"http://xn--wgbh1c.icom.museum", u"xn--wgbh1c.icom.museum",
u"http://xn--wgbh1c.icom.museum"},
{"Non-standard HTTP port", "http://www.google.com:9000/",
u"http://www.google.com:9000", u"www.google.com:9000",
u"http://www.google.com:9000"},
{"Non-standard HTTPS port", "https://www.google.com:9000/",
u"https://www.google.com:9000", u"www.google.com:9000",
u"www.google.com:9000"},
{"HTTP URL with path", "http://www.google.com/test.html",
u"http://www.google.com", u"www.google.com", u"http://www.google.com"},
{"HTTPS URL with path", "https://www.google.com/test.html",
u"https://www.google.com", u"www.google.com", u"www.google.com"},
{"Unusual secure scheme (wss)", "wss://www.google.com/",
u"wss://www.google.com", u"wss://www.google.com", u"www.google.com"},
{"Unusual non-secure scheme (ftp)", "ftp://www.google.com/",
u"ftp://www.google.com", u"ftp://www.google.com", u"ftp://www.google.com"},
{"Unlisted scheme (chrome)", "chrome://version", u"chrome://version",
u"chrome://version", u"chrome://version"},
{"HTTP IP address", "http://173.194.65.103", u"http://173.194.65.103",
u"173.194.65.103", u"http://173.194.65.103"},
{"HTTPS IP address", "https://173.194.65.103", u"https://173.194.65.103",
u"173.194.65.103", u"173.194.65.103"},
{"HTTP IPv6 address", "http://[FE80:0000:0000:0000:0202:B3FF:FE1E:8329]/",
u"http://[fe80::202:b3ff:fe1e:8329]", u"[fe80::202:b3ff:fe1e:8329]",
u"http://[fe80::202:b3ff:fe1e:8329]"},
{"HTTPs IPv6 address", "https://[FE80:0000:0000:0000:0202:B3FF:FE1E:8329]/",
u"https://[fe80::202:b3ff:fe1e:8329]", u"[fe80::202:b3ff:fe1e:8329]",
u"[fe80::202:b3ff:fe1e:8329]"},
{"HTTP IPv6 address with port",
"http://[FE80:0000:0000:0000:0202:B3FF:FE1E:8329]:80/",
u"http://[fe80::202:b3ff:fe1e:8329]", u"[fe80::202:b3ff:fe1e:8329]",
u"http://[fe80::202:b3ff:fe1e:8329]"},
{"HTTPs IPv6 address with port",
"https://[FE80:0000:0000:0000:0202:B3FF:FE1E:8329]:443/",
u"https://[fe80::202:b3ff:fe1e:8329]", u"[fe80::202:b3ff:fe1e:8329]",
u"[fe80::202:b3ff:fe1e:8329]"},
{"HTTPS IP address, non-default port", "https://173.194.65.103:8443",
u"https://173.194.65.103:8443", u"173.194.65.103:8443",
u"173.194.65.103:8443"},
{"Invalid host 1", "https://www.cyber../wow.php", u"https://www.cyber..",
u"www.cyber..", u"www.cyber.."},
{"Invalid host 2", "https://www...cyber/wow.php", u"https://www...cyber",
u"www...cyber", u"www...cyber"},
{"Invalid port 3", "https://173.194.65.103:/hello.aspx",
u"https://173.194.65.103", u"173.194.65.103", u"173.194.65.103"},
{"Trailing dot in DNS name", "https://www.example.com./get/goat",
u"https://www.example.com.", u"www.example.com.", u"www.example.com."}};
TEST(TextEliderTest, FormatUrlForSecurityDisplay) {
for (const auto& common_test : common_tests) {
std::u16string formatted =
url_formatter::FormatUrlForSecurityDisplay(GURL(common_test.input));
EXPECT_EQ(common_test.output, formatted) << common_test.description;
std::u16string formatted_omit_web_scheme =
url_formatter::FormatUrlForSecurityDisplay(
GURL(common_test.input),
url_formatter::SchemeDisplay::OMIT_HTTP_AND_HTTPS);
EXPECT_EQ(common_test.output_omit_web_scheme, formatted_omit_web_scheme)
<< common_test.description;
std::u16string formatted_omit_cryptographic_scheme =
url_formatter::FormatUrlForSecurityDisplay(
GURL(common_test.input),
url_formatter::SchemeDisplay::OMIT_CRYPTOGRAPHIC);
EXPECT_EQ(common_test.output_omit_cryptographic_scheme,
formatted_omit_cryptographic_scheme)
<< common_test.description;
}
const OriginTestData tests[] = {
{"File URI", "file:///usr/example/file.html",
u"file:///usr/example/file.html", u"file:///usr/example/file.html",
u"file:///usr/example/file.html"},
{"File URI with hostname", "file://localhost/usr/example/file.html",
u"file:///usr/example/file.html", u"file:///usr/example/file.html",
u"file:///usr/example/file.html"},
{"UNC File URI 1", "file:///CONTOSO/accounting/money.xls",
u"file:///CONTOSO/accounting/money.xls",
u"file:///CONTOSO/accounting/money.xls",
u"file:///CONTOSO/accounting/money.xls"},
{"UNC File URI 2",
"file:///C:/Program%20Files/Music/Web%20Sys/main.html?REQUEST=RADIO",
u"file:///C:/Program%20Files/Music/Web%20Sys/main.html",
u"file:///C:/Program%20Files/Music/Web%20Sys/main.html",
u"file:///C:/Program%20Files/Music/Web%20Sys/main.html"},
{"Invalid IPv6 address", "https://[2001:db8:0:1]/",
u"https://[2001:db8:0:1]", u"https://[2001:db8:0:1]",
u"https://[2001:db8:0:1]"},
{"HTTP filesystem: URL with path",
"filesystem:http://www.google.com/temporary/test.html",
u"filesystem:http://www.google.com", u"filesystem:http://www.google.com",
u"filesystem:http://www.google.com"},
{"File filesystem: URL with path",
"filesystem:file://localhost/temporary/stuff/"
"test.html?z=fun&goat=billy",
u"filesystem:file:///temporary/stuff/test.html",
u"filesystem:file:///temporary/stuff/test.html",
u"filesystem:file:///temporary/stuff/test.html"},
{"Invalid scheme 1", "twelve://www.cyber.org/wow.php",
u"twelve://www.cyber.org/wow.php", u"twelve://www.cyber.org/wow.php",
u"twelve://www.cyber.org/wow.php"},
{"Invalid scheme 2", "://www.cyber.org/wow.php",
u"://www.cyber.org/wow.php", u"://www.cyber.org/wow.php",
u"://www.cyber.org/wow.php"},
{"Invalid port 1", "https://173.194.65.103:000",
u"https://173.194.65.103:0", u"173.194.65.103:0", u"173.194.65.103:0"},
{"Invalid port 2", "https://173.194.65.103:gruffle",
u"https://173.194.65.103:gruffle", u"https://173.194.65.103:gruffle",
u"https://173.194.65.103:gruffle"},
{"Blob URL",
"blob:http://www.html5rocks.com/4d4ff040-6d61-4446-86d3-13ca07ec9ab9",
u"blob:http://www.html5rocks.com/"
u"4d4ff040-6d61-4446-86d3-13ca07ec9ab9",
u"blob:http://www.html5rocks.com/"
u"4d4ff040-6d61-4446-86d3-13ca07ec9ab9",
u"blob:http://www.html5rocks.com/"
u"4d4ff040-6d61-4446-86d3-13ca07ec9ab9"}};
for (const auto& test : tests) {
std::u16string formatted =
url_formatter::FormatUrlForSecurityDisplay(GURL(test.input));
EXPECT_EQ(test.output, formatted) << test.description;
std::u16string formatted_omit_web_scheme =
url_formatter::FormatUrlForSecurityDisplay(
GURL(test.input),
url_formatter::SchemeDisplay::OMIT_HTTP_AND_HTTPS);
EXPECT_EQ(test.output_omit_web_scheme, formatted_omit_web_scheme)
<< test.description;
std::u16string formatted_omit_cryptographic_scheme =
url_formatter::FormatUrlForSecurityDisplay(
GURL(test.input), url_formatter::SchemeDisplay::OMIT_CRYPTOGRAPHIC);
EXPECT_EQ(test.output_omit_cryptographic_scheme,
formatted_omit_cryptographic_scheme)
<< test.description;
}
std::u16string formatted = url_formatter::FormatUrlForSecurityDisplay(GURL());
EXPECT_EQ(std::u16string(), formatted)
<< "Explicitly test the 0-argument GURL constructor";
std::u16string formatted_omit_scheme =
url_formatter::FormatUrlForSecurityDisplay(
GURL(), url_formatter::SchemeDisplay::OMIT_HTTP_AND_HTTPS);
EXPECT_EQ(std::u16string(), formatted_omit_scheme)
<< "Explicitly test the 0-argument GURL constructor";
formatted_omit_scheme = url_formatter::FormatUrlForSecurityDisplay(
GURL(), url_formatter::SchemeDisplay::OMIT_CRYPTOGRAPHIC);
EXPECT_EQ(std::u16string(), formatted_omit_scheme)
<< "Explicitly test the 0-argument GURL constructor";
}
TEST(TextEliderTest, FormatOriginForSecurityDisplay) {
for (const auto& common_test : common_tests) {
std::u16string formatted = url_formatter::FormatOriginForSecurityDisplay(
url::Origin::Create(GURL(common_test.input)));
EXPECT_EQ(common_test.output, formatted) << common_test.description;
std::u16string formatted_omit_web_scheme =
url_formatter::FormatOriginForSecurityDisplay(
url::Origin::Create(GURL(common_test.input)),
url_formatter::SchemeDisplay::OMIT_HTTP_AND_HTTPS);
EXPECT_EQ(common_test.output_omit_web_scheme, formatted_omit_web_scheme)
<< common_test.description;
std::u16string formatted_omit_cryptographic_scheme =
url_formatter::FormatOriginForSecurityDisplay(
url::Origin::Create(GURL(common_test.input)),
url_formatter::SchemeDisplay::OMIT_CRYPTOGRAPHIC);
EXPECT_EQ(common_test.output_omit_cryptographic_scheme,
formatted_omit_cryptographic_scheme)
<< common_test.description;
}
const OriginTestData tests[] = {
{"File URI", "file:///usr/example/file.html", u"file://", u"file://",
u"file://"},
{"File URI with hostname", "file://localhost/usr/example/file.html",
u"file://localhost", u"file://localhost", u"file://localhost"},
{"UNC File URI 1", "file:///CONTOSO/accounting/money.xls", u"file://",
u"file://", u"file://"},
{"UNC File URI 2",
"file:///C:/Program%20Files/Music/Web%20Sys/main.html?REQUEST=RADIO",
u"file://", u"file://", u"file://"},
{"Invalid IPv6 address", "https://[2001:db8:0:1]/", u"", u"", u""},
{"HTTP filesystem: URL with path",
"filesystem:http://www.google.com/temporary/test.html",
u"http://www.google.com", u"www.google.com", u"http://www.google.com"},
{"File filesystem: URL with path",
"filesystem:file://localhost/temporary/stuff/test.html?z=fun&goat=billy",
u"file://", u"file://", u"file://"},
{"Invalid scheme 1", "twelve://www.cyber.org/wow.php", u"", u"", u""},
{"Invalid scheme 2", "://www.cyber.org/wow.php", u"", u"", u""},
{"Invalid port 1", "https://173.194.65.103:99999", u"", u"", u""},
{"Invalid port 2", "https://173.194.65.103:gruffle", u"", u"", u""},
{"Blob URL",
"blob:http://www.html5rocks.com/4d4ff040-6d61-4446-86d3-13ca07ec9ab9",
u"http://www.html5rocks.com", u"www.html5rocks.com",
u"http://www.html5rocks.com"}};
for (const auto& test : tests) {
std::u16string formatted = url_formatter::FormatOriginForSecurityDisplay(
url::Origin::Create(GURL(test.input)));
EXPECT_EQ(test.output, formatted) << test.description;
std::u16string formatted_omit_web_scheme =
url_formatter::FormatOriginForSecurityDisplay(
url::Origin::Create(GURL(test.input)),
url_formatter::SchemeDisplay::OMIT_HTTP_AND_HTTPS);
EXPECT_EQ(test.output_omit_web_scheme, formatted_omit_web_scheme)
<< test.description;
std::u16string formatted_omit_cryptographic_scheme =
url_formatter::FormatOriginForSecurityDisplay(
url::Origin::Create(GURL(test.input)),
url_formatter::SchemeDisplay::OMIT_CRYPTOGRAPHIC);
EXPECT_EQ(test.output_omit_cryptographic_scheme,
formatted_omit_cryptographic_scheme)
<< test.description;
}
std::u16string formatted = url_formatter::FormatOriginForSecurityDisplay(
url::Origin::Create(GURL()));
EXPECT_EQ(std::u16string(), formatted)
<< "Explicitly test the url::Origin which takes an empty, invalid URL";
std::u16string formatted_omit_scheme =
url_formatter::FormatOriginForSecurityDisplay(
url::Origin::Create(GURL()),
url_formatter::SchemeDisplay::OMIT_HTTP_AND_HTTPS);
EXPECT_EQ(std::u16string(), formatted_omit_scheme)
<< "Explicitly test the url::Origin which takes an empty, invalid URL";
formatted_omit_scheme = url_formatter::FormatOriginForSecurityDisplay(
url::Origin::Create(GURL()),
url_formatter::SchemeDisplay::OMIT_CRYPTOGRAPHIC);
EXPECT_EQ(std::u16string(), formatted_omit_scheme)
<< "Explicitly test the url::Origin which takes an empty, invalid URL";
}
TEST(TextEliderTest, FormatUrlForDisplayOmitSchemePathAndTrivialSubdomains) {
EXPECT_EQ(
u"google.com",
url_formatter::FormatUrlForDisplayOmitSchemePathAndTrivialSubdomains(
GURL("http://user:pass@google.com/path")));
EXPECT_EQ(
u"chrome://version",
url_formatter::FormatUrlForDisplayOmitSchemePathAndTrivialSubdomains(
GURL("chrome://version")));
EXPECT_EQ(
u"äää.de",
url_formatter::FormatUrlForDisplayOmitSchemePathAndTrivialSubdomains(
GURL("https://äää.de")));
EXPECT_EQ(
u"xn--4caaa.com",
url_formatter::FormatUrlForDisplayOmitSchemePathAndTrivialSubdomains(
GURL("https://äää.com")));
EXPECT_EQ(
u"مثال.إختبار",
url_formatter::FormatUrlForDisplayOmitSchemePathAndTrivialSubdomains(
GURL("https://xn--mgbh0fb.xn--kgbechtv/")));
}
TEST(TextEliderTest,
FormatUrlForDisplayOmitSchemePathTrivialSubdomainsAndMobilePrefix) {
#if BUILDFLAG(IS_IOS)
EXPECT_EQ(
u"google.com",
url_formatter::
FormatUrlForDisplayOmitSchemePathTrivialSubdomainsAndMobilePrefix(
GURL("http://m.google.com/example")));
EXPECT_EQ(
u"google.com",
url_formatter::
FormatUrlForDisplayOmitSchemePathTrivialSubdomainsAndMobilePrefix(
GURL("http://www.m.google.com/example")));
EXPECT_EQ(
u"google.com",
url_formatter::
FormatUrlForDisplayOmitSchemePathTrivialSubdomainsAndMobilePrefix(
GURL("http://m.www.google.com/example")));
#else
GTEST_SKIP();
#endif
}
} // namespace