mirror of
https://github.com/catchorg/Catch2.git
synced 2024-11-25 06:46:10 +01:00
8ce2426e53
This PR adds functionality to skip around ANSI escape sequences in catch_textflow so they do not contribute to line length and line wrapping code does not split escape sequences in the middle. I've implemented this by creating a AnsiSkippingString abstraction that has a bidirectional iterator that can skip around escape sequences while iterating. Additionally I refactored Column::const_iterator to be iterator-based rather than index-based so this abstraction is a simple drop-in for std::string. Currently only color sequences are handled, other escape sequences are left unaffected. Motivation: Text with ANSI color sequences gets messed up when being output by Catch2 #2833.
401 lines
13 KiB
C++
401 lines
13 KiB
C++
|
|
// Copyright Catch2 Authors
|
|
// Distributed under the Boost Software License, Version 1.0.
|
|
// (See accompanying file LICENSE.txt or copy at
|
|
// https://www.boost.org/LICENSE_1_0.txt)
|
|
|
|
// SPDX-License-Identifier: BSL-1.0
|
|
|
|
#include <catch2/catch_test_macros.hpp>
|
|
#include <catch2/internal/catch_textflow.hpp>
|
|
|
|
#include <sstream>
|
|
|
|
using Catch::TextFlow::Column;
|
|
using Catch::TextFlow::AnsiSkippingString;
|
|
|
|
namespace {
|
|
static std::string as_written(Column const& c) {
|
|
std::stringstream sstr;
|
|
sstr << c;
|
|
return sstr.str();
|
|
}
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::Column one simple line",
|
|
"[TextFlow][column][approvals]" ) {
|
|
Column col( "simple short line" );
|
|
|
|
REQUIRE(as_written(col) == "simple short line");
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::Column respects already present newlines",
|
|
"[TextFlow][column][approvals]" ) {
|
|
Column col( "abc\ndef" );
|
|
REQUIRE( as_written( col ) == "abc\ndef" );
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::Column respects width setting",
|
|
"[TextFlow][column][approvals]" ) {
|
|
Column col( "The quick brown fox jumped over the lazy dog" );
|
|
|
|
SECTION( "width=20" ) {
|
|
col.width( 20 );
|
|
REQUIRE( as_written( col ) == "The quick brown fox\n"
|
|
"jumped over the lazy\n"
|
|
"dog" );
|
|
}
|
|
SECTION("width=10") {
|
|
col.width( 10 );
|
|
REQUIRE( as_written( col ) == "The quick\n"
|
|
"brown fox\n"
|
|
"jumped\n"
|
|
"over the\n"
|
|
"lazy dog" );
|
|
}
|
|
SECTION("width=5") {
|
|
// This is so small some words will have to be split with hyphen
|
|
col.width(5);
|
|
REQUIRE( as_written( col ) == "The\n"
|
|
"quick\n"
|
|
"brown\n"
|
|
"fox\n"
|
|
"jump-\n"
|
|
"ed\n"
|
|
"over\n"
|
|
"the\n"
|
|
"lazy\n"
|
|
"dog" );
|
|
}
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::Column respects indentation setting",
|
|
"[TextFlow][column][approvals]" ) {
|
|
Column col( "First line\nSecond line\nThird line" );
|
|
|
|
SECTION("Default: no indentation at all") {
|
|
REQUIRE(as_written(col) == "First line\nSecond line\nThird line");
|
|
}
|
|
SECTION("Indentation on first line only") {
|
|
col.initialIndent(3);
|
|
REQUIRE(as_written(col) == " First line\nSecond line\nThird line");
|
|
}
|
|
SECTION("Indentation on all lines") {
|
|
col.indent(3);
|
|
REQUIRE(as_written(col) == " First line\n Second line\n Third line");
|
|
}
|
|
SECTION("Indentation on later lines only") {
|
|
col.indent(5).initialIndent(0);
|
|
REQUIRE(as_written(col) == "First line\n Second line\n Third line");
|
|
}
|
|
SECTION("Different indentation on first and later lines") {
|
|
col.initialIndent(1).indent(2);
|
|
REQUIRE(as_written(col) == " First line\n Second line\n Third line");
|
|
}
|
|
}
|
|
|
|
TEST_CASE("TextFlow::Column indentation respects whitespace", "[TextFlow][column][approvals]") {
|
|
Column col(" text with whitespace\n after newlines");
|
|
|
|
SECTION("No extra indentation") {
|
|
col.initialIndent(0).indent(0);
|
|
REQUIRE(as_written(col) == " text with whitespace\n after newlines");
|
|
}
|
|
SECTION("Different indentation on first and later lines") {
|
|
col.initialIndent(1).indent(2);
|
|
REQUIRE(as_written(col) == " text with whitespace\n after newlines");
|
|
}
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::Column linebreaking prefers boundary characters",
|
|
"[TextFlow][column][approvals]" ) {
|
|
SECTION("parentheses") {
|
|
Column col("(Hello)aaa(World)");
|
|
SECTION("width=20") {
|
|
col.width(20);
|
|
REQUIRE(as_written(col) == "(Hello)aaa(World)");
|
|
}
|
|
SECTION("width=15") {
|
|
col.width(15);
|
|
REQUIRE(as_written(col) == "(Hello)aaa\n(World)");
|
|
}
|
|
SECTION("width=8") {
|
|
col.width(8);
|
|
REQUIRE(as_written(col) == "(Hello)\naaa\n(World)");
|
|
}
|
|
}
|
|
SECTION("commas") {
|
|
Column col("Hello, world");
|
|
col.width(8);
|
|
|
|
REQUIRE(as_written(col) == "Hello,\nworld");
|
|
}
|
|
}
|
|
|
|
|
|
TEST_CASE( "TextFlow::Column respects indentation for empty lines",
|
|
"[TextFlow][column][approvals][!shouldfail]" ) {
|
|
// This is currently bugged and does not do what it should
|
|
Column col("\n\nthird line");
|
|
col.indent(2);
|
|
|
|
//auto b = col.begin();
|
|
//auto e = col.end();
|
|
|
|
//auto b1 = *b;
|
|
//++b;
|
|
//auto b2 = *b;
|
|
//++b;
|
|
//auto b3 = *b;
|
|
//++b;
|
|
|
|
//REQUIRE(b == e);
|
|
|
|
std::string written = as_written(col);
|
|
|
|
REQUIRE(written == " \n \n third line");
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::Column leading/trailing whitespace",
|
|
"[TextFlow][column][approvals]" ) {
|
|
SECTION("Trailing whitespace") {
|
|
Column col("some trailing whitespace: \t");
|
|
REQUIRE(as_written(col) == "some trailing whitespace: \t");
|
|
}
|
|
SECTION("Some leading whitespace") {
|
|
Column col("\t \t whitespace wooo");
|
|
REQUIRE(as_written(col) == "\t \t whitespace wooo");
|
|
}
|
|
SECTION("both") {
|
|
Column col(" abc ");
|
|
REQUIRE(as_written(col) == " abc ");
|
|
}
|
|
SECTION("whitespace only") {
|
|
Column col("\t \t");
|
|
REQUIRE(as_written(col) == "\t \t");
|
|
}
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::Column can handle empty string",
|
|
"[TextFlow][column][approvals]" ) {
|
|
Column col("");
|
|
REQUIRE(as_written(col) == "");
|
|
}
|
|
|
|
TEST_CASE( "#1400 - TextFlow::Column wrapping would sometimes duplicate words",
|
|
"[TextFlow][column][regression][approvals]" ) {
|
|
const auto long_string = std::string(
|
|
"Lorem ipsum dolor sit amet, consectetur adipiscing elit. Pellentesque nisl \n"
|
|
"massa, luctus ut ligula vitae, suscipit tempus velit. Vivamus sodales, quam in \n"
|
|
"convallis posuere, libero nisi ultricies orci, nec lobortis.\n");
|
|
|
|
auto col = Column(long_string)
|
|
.width(79)
|
|
.indent(2);
|
|
|
|
REQUIRE(as_written(col) ==
|
|
" Lorem ipsum dolor sit amet, consectetur adipiscing elit. Pellentesque nisl \n"
|
|
" massa, luctus ut ligula vitae, suscipit tempus velit. Vivamus sodales, quam\n"
|
|
" in \n"
|
|
" convallis posuere, libero nisi ultricies orci, nec lobortis.");
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::AnsiSkippingString skips ansi sequences",
|
|
"[TextFlow][ansiskippingstring][approvals]" ) {
|
|
|
|
SECTION("basic string") {
|
|
std::string text = "a\033[38;2;98;174;239mb\033[38mc\033[0md\033[me";
|
|
AnsiSkippingString str(text);
|
|
|
|
SECTION( "iterates forward" ) {
|
|
auto it = str.begin();
|
|
CHECK(*it == 'a');
|
|
++it;
|
|
CHECK(*it == 'b');
|
|
++it;
|
|
CHECK(*it == 'c');
|
|
++it;
|
|
CHECK(*it == 'd');
|
|
++it;
|
|
CHECK(*it == 'e');
|
|
++it;
|
|
CHECK(it == str.end());
|
|
}
|
|
SECTION( "iterates backwards" ) {
|
|
auto it = str.end();
|
|
--it;
|
|
CHECK(*it == 'e');
|
|
--it;
|
|
CHECK(*it == 'd');
|
|
--it;
|
|
CHECK(*it == 'c');
|
|
--it;
|
|
CHECK(*it == 'b');
|
|
--it;
|
|
CHECK(*it == 'a');
|
|
CHECK(it == str.begin());
|
|
}
|
|
}
|
|
|
|
SECTION( "ansi escape sequences at the start" ) {
|
|
std::string text = "\033[38;2;98;174;239ma\033[38;2;98;174;239mb\033[38mc\033[0md\033[me";
|
|
AnsiSkippingString str(text);
|
|
auto it = str.begin();
|
|
CHECK(*it == 'a');
|
|
++it;
|
|
CHECK(*it == 'b');
|
|
++it;
|
|
CHECK(*it == 'c');
|
|
++it;
|
|
CHECK(*it == 'd');
|
|
++it;
|
|
CHECK(*it == 'e');
|
|
++it;
|
|
CHECK(it == str.end());
|
|
--it;
|
|
CHECK(*it == 'e');
|
|
--it;
|
|
CHECK(*it == 'd');
|
|
--it;
|
|
CHECK(*it == 'c');
|
|
--it;
|
|
CHECK(*it == 'b');
|
|
--it;
|
|
CHECK(*it == 'a');
|
|
CHECK(it == str.begin());
|
|
}
|
|
|
|
SECTION( "ansi escape sequences at the end" ) {
|
|
std::string text = "a\033[38;2;98;174;239mb\033[38mc\033[0md\033[me\033[38;2;98;174;239m";
|
|
AnsiSkippingString str(text);
|
|
auto it = str.begin();
|
|
CHECK(*it == 'a');
|
|
++it;
|
|
CHECK(*it == 'b');
|
|
++it;
|
|
CHECK(*it == 'c');
|
|
++it;
|
|
CHECK(*it == 'd');
|
|
++it;
|
|
CHECK(*it == 'e');
|
|
++it;
|
|
CHECK(it == str.end());
|
|
--it;
|
|
CHECK(*it == 'e');
|
|
--it;
|
|
CHECK(*it == 'd');
|
|
--it;
|
|
CHECK(*it == 'c');
|
|
--it;
|
|
CHECK(*it == 'b');
|
|
--it;
|
|
CHECK(*it == 'a');
|
|
CHECK(it == str.begin());
|
|
}
|
|
|
|
SECTION( "skips consecutive escapes" ) {
|
|
std::string text = "\033[38;2;98;174;239m\033[38;2;98;174;239ma\033[38;2;98;174;239mb\033[38m\033[38m\033[38mc\033[0md\033[me";
|
|
AnsiSkippingString str(text);
|
|
auto it = str.begin();
|
|
CHECK(*it == 'a');
|
|
++it;
|
|
CHECK(*it == 'b');
|
|
++it;
|
|
CHECK(*it == 'c');
|
|
++it;
|
|
CHECK(*it == 'd');
|
|
++it;
|
|
CHECK(*it == 'e');
|
|
++it;
|
|
CHECK(it == str.end());
|
|
--it;
|
|
CHECK(*it == 'e');
|
|
--it;
|
|
CHECK(*it == 'd');
|
|
--it;
|
|
CHECK(*it == 'c');
|
|
--it;
|
|
CHECK(*it == 'b');
|
|
--it;
|
|
CHECK(*it == 'a');
|
|
CHECK(it == str.begin());
|
|
}
|
|
|
|
SECTION( "handles incomplete ansi sequences" ) {
|
|
std::string text = "a\033[b\033[30c\033[30;d\033[30;2e";
|
|
AnsiSkippingString str(text);
|
|
CHECK(std::string(str.begin(), str.end()) == text);
|
|
}
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::AnsiSkippingString computes the size properly",
|
|
"[TextFlow][ansiskippingstring][approvals]" ) {
|
|
std::string text = "\033[38;2;98;174;239m\033[38;2;98;174;239ma\033[38;2;98;174;239mb\033[38m\033[38m\033[38mc\033[0md\033[me";
|
|
AnsiSkippingString str(text);
|
|
CHECK(str.size() == 5);
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::AnsiSkippingString substrings properly",
|
|
"[TextFlow][ansiskippingstring][approvals]" ) {
|
|
SECTION("basic test") {
|
|
std::string text = "a\033[38;2;98;174;239mb\033[38mc\033[0md\033[me";
|
|
AnsiSkippingString str(text);
|
|
auto a = str.begin();
|
|
auto b = str.begin();
|
|
++b;
|
|
++b;
|
|
CHECK(str.substring(a, b) == "a\033[38;2;98;174;239mb\033[38m");
|
|
++a;
|
|
++b;
|
|
CHECK(str.substring(a, b) == "b\033[38mc\033[0m");
|
|
CHECK(str.substring(a, str.end()) == "b\033[38mc\033[0md\033[me");
|
|
CHECK(str.substring(str.begin(), str.end()) == text);
|
|
}
|
|
SECTION("escapes at the start") {
|
|
std::string text = "\033[38;2;98;174;239m\033[38;2;98;174;239ma\033[38;2;98;174;239mb\033[38m\033[38m\033[38mc\033[0md\033[me";
|
|
AnsiSkippingString str(text);
|
|
auto a = str.begin();
|
|
auto b = str.begin();
|
|
++b;
|
|
++b;
|
|
CHECK(str.substring(a, b) == "\033[38;2;98;174;239m\033[38;2;98;174;239ma\033[38;2;98;174;239mb\033[38m\033[38m\033[38m");
|
|
++a;
|
|
++b;
|
|
CHECK(str.substring(a, b) == "b\033[38m\033[38m\033[38mc\033[0m");
|
|
CHECK(str.substring(a, str.end()) == "b\033[38m\033[38m\033[38mc\033[0md\033[me");
|
|
CHECK(str.substring(str.begin(), str.end()) == text);
|
|
}
|
|
SECTION("escapes at the end") {
|
|
std::string text = "a\033[38;2;98;174;239mb\033[38mc\033[0md\033[me\033[38m";
|
|
AnsiSkippingString str(text);
|
|
auto a = str.begin();
|
|
auto b = str.begin();
|
|
++b;
|
|
++b;
|
|
CHECK(str.substring(a, b) == "a\033[38;2;98;174;239mb\033[38m");
|
|
++a;
|
|
++b;
|
|
CHECK(str.substring(a, b) == "b\033[38mc\033[0m");
|
|
CHECK(str.substring(a, str.end()) == "b\033[38mc\033[0md\033[me\033[38m");
|
|
CHECK(str.substring(str.begin(), str.end()) == text);
|
|
}
|
|
}
|
|
|
|
TEST_CASE( "TextFlow::Column skips ansi escape sequences",
|
|
"[TextFlow][column][approvals]" ) {
|
|
std::string text = "\033[38;2;98;174;239m\033[38;2;198;120;221mThe quick brown \033[38;2;198;120;221mfox jumped over the lazy dog\033[0m";
|
|
Column col(text);
|
|
|
|
SECTION( "width=20" ) {
|
|
col.width( 20 );
|
|
REQUIRE( as_written( col ) == "\033[38;2;98;174;239m\033[38;2;198;120;221mThe quick brown \033[38;2;198;120;221mfox\n"
|
|
"jumped over the lazy\n"
|
|
"dog\033[0m" );
|
|
}
|
|
|
|
SECTION( "width=80" ) {
|
|
col.width( 80 );
|
|
REQUIRE( as_written( col ) == text );
|
|
}
|
|
}
|