ladybird/Libraries/LibDiff/Hunks.cpp

197 lines
5.3 KiB
C++
Raw Normal View History

/*
* Copyright (c) 2020, Itamar S. <itamar8910@gmail.com>
* Copyright (c) 2023, Shannon Booth <shannon@serenityos.org>
*
* SPDX-License-Identifier: BSD-2-Clause
*/
#include "Hunks.h"
#include <AK/Debug.h>
#include <AK/LexicalPath.h>
namespace Diff {
Optional<HunkLocation> Parser::consume_unified_location()
{
auto consume_range = [this](Range& range) {
if (!consume_line_number(range.start_line))
return false;
if (consume_specific(',')) {
if (!consume_line_number(range.number_of_lines))
return false;
} else {
range.number_of_lines = 1;
}
return true;
};
if (!consume_specific("@@ -"sv))
return {};
HunkLocation location;
if (!consume_range(location.old_range))
return {};
if (!consume_specific(" +"sv))
return {};
if (!consume_range(location.new_range))
return {};
if (!consume_specific(" @@"sv))
return {};
return location;
}
bool Parser::consume_line_number(size_t& number)
{
auto line = consume_while(is_ascii_digit);
auto maybe_number = line.to_number<size_t>();
if (!maybe_number.has_value())
return false;
number = maybe_number.value();
return true;
}
ErrorOr<String> Parser::parse_file_line(Optional<size_t> const& strip_count)
{
// FIXME: handle parsing timestamps as well.
auto line = consume_line();
GenericLexer line_parser(line);
auto path = line_parser.consume_until('\t');
// No strip count given. Default to basename of file.
if (!strip_count.has_value())
return String::from_byte_string(LexicalPath::basename(path));
// NOTE: We cannot use LexicalPath::parts as we want to strip the non-canonicalized path.
auto const& parts = path.split_view('/');
// More components to strip than the filename has. Just pretend it is missing.
if (strip_count.value() >= parts.size())
return String();
// Remove given number of leading components from the path.
size_t components = parts.size() - strip_count.value();
StringBuilder stripped_path;
for (size_t i = parts.size() - components; i < parts.size(); ++i) {
TRY(stripped_path.try_append(parts[i]));
if (i != parts.size() - 1)
TRY(stripped_path.try_append("/"sv));
}
return stripped_path.to_string();
}
ErrorOr<Patch> Parser::parse_patch(Optional<size_t> const& strip_count)
{
Patch patch;
patch.header = TRY(parse_header(strip_count));
patch.hunks = TRY(parse_hunks());
return patch;
}
ErrorOr<Header> Parser::parse_header(Optional<size_t> const& strip_count)
{
Header header;
while (!is_eof()) {
if (consume_specific("+++ "sv)) {
header.new_file_path = TRY(parse_file_line(strip_count));
continue;
}
if (consume_specific("--- "sv)) {
header.old_file_path = TRY(parse_file_line(strip_count));
continue;
}
if (next_is("@@ ")) {
header.format = Format::Unified;
return header;
}
consume_line();
}
return header;
}
ErrorOr<Vector<Hunk>> Parser::parse_hunks()
{
Vector<Hunk> hunks;
while (next_is("@@ ")) {
// Try an locate a hunk location in this hunk. It may be prefixed with information.
auto maybe_location = consume_unified_location();
consume_line();
if (!maybe_location.has_value())
break;
Hunk hunk { *maybe_location, {} };
auto old_lines_expected = hunk.location.old_range.number_of_lines;
auto new_lines_expected = hunk.location.new_range.number_of_lines;
// We've found a location. Now parse out all of the expected content lines.
while (old_lines_expected != 0 || new_lines_expected != 0) {
StringView line = consume_line();
if (line.is_empty())
return Error::from_string_literal("Malformed empty content line in patch");
if (line[0] != ' ' && line[0] != '+' && line[0] != '-')
return Error::from_string_literal("Invaid operation in patch");
auto const operation = Line::operation_from_symbol(line[0]);
if (operation != Line::Operation::Removal) {
if (new_lines_expected == 0)
return Error::from_string_literal("Found more removal and context lines in patch than expected");
--new_lines_expected;
}
if (operation != Line::Operation::Addition) {
if (old_lines_expected == 0)
return Error::from_string_literal("Found more addition and context lines in patch than expected");
--old_lines_expected;
}
auto const content = line.substring_view(1, line.length() - 1);
TRY(hunk.lines.try_append(Line { operation, TRY(String::from_utf8(content)) }));
}
TRY(hunks.try_append(hunk));
}
if constexpr (HUNKS_DEBUG) {
2022-04-01 17:58:27 +00:00
for (auto const& hunk : hunks) {
dbgln("{}", hunk.location);
for (auto const& line : hunk.lines)
dbgln("{}", line);
}
}
return hunks;
}
ErrorOr<Vector<Hunk>> parse_hunks(StringView diff)
{
Parser lexer(diff);
while (!lexer.next_is("@@ ") && !lexer.is_eof())
lexer.consume_line();
return lexer.parse_hunks();
}
}