
SPDX License Identifiers are a more compact / standardized way of representing file license information. See: https://spdx.dev/resources/use/#identifiers This was done with the `ambr` search and replace tool. ambr --no-parent-ignore --key-from-file --rep-from-file key.txt rep.txt *
128 lines
4.5 KiB
C++
128 lines
4.5 KiB
C++
/*
|
|
* Copyright (c) 2020, Itamar S. <itamar8910@gmail.com>
|
|
*
|
|
* SPDX-License-Identifier: BSD-2-Clause
|
|
*/
|
|
|
|
#include "Hunks.h"
|
|
#include <AK/Debug.h>
|
|
|
|
namespace Diff {
|
|
Vector<Hunk> parse_hunks(const String& diff)
|
|
{
|
|
Vector<String> diff_lines = diff.split('\n');
|
|
if (diff_lines.is_empty())
|
|
return {};
|
|
|
|
Vector<Hunk> hunks;
|
|
|
|
size_t line_index = 0;
|
|
HunkLocation current_location {};
|
|
|
|
// Skip to first hunk
|
|
while (diff_lines[line_index][0] != '@') {
|
|
++line_index;
|
|
}
|
|
|
|
while (line_index < diff_lines.size()) {
|
|
if (diff_lines[line_index][0] == '@') {
|
|
current_location = parse_hunk_location(diff_lines[line_index]);
|
|
++line_index;
|
|
continue;
|
|
}
|
|
if (diff_lines[line_index][0] == ' ') {
|
|
current_location.apply_offset(1, HunkLocation::LocationType::Both);
|
|
++line_index;
|
|
continue;
|
|
}
|
|
Hunk hunk {};
|
|
hunk.original_start_line = current_location.original_start_line;
|
|
hunk.target_start_line = current_location.target_start_line;
|
|
|
|
while (line_index < diff_lines.size() && diff_lines[line_index][0] == '-') {
|
|
hunk.removed_lines.append(diff_lines[line_index].substring(1, diff_lines[line_index].length() - 1));
|
|
current_location.apply_offset(1, HunkLocation::LocationType::Original);
|
|
++line_index;
|
|
}
|
|
while (line_index < diff_lines.size() && diff_lines[line_index][0] == '+') {
|
|
hunk.added_lines.append(diff_lines[line_index].substring(1, diff_lines[line_index].length() - 1));
|
|
current_location.apply_offset(1, HunkLocation::LocationType::Target);
|
|
++line_index;
|
|
}
|
|
|
|
while (line_index < diff_lines.size() && diff_lines[line_index][0] == ' ') {
|
|
current_location.apply_offset(1, HunkLocation::LocationType::Both);
|
|
++line_index;
|
|
}
|
|
hunks.append(hunk);
|
|
}
|
|
|
|
if constexpr (HUNKS_DEBUG) {
|
|
for (const auto& hunk : hunks) {
|
|
dbgln("Hunk location:");
|
|
dbgln(" orig: {}", hunk.original_start_line);
|
|
dbgln(" target: {}", hunk.target_start_line);
|
|
dbgln(" removed:");
|
|
for (const auto& line : hunk.removed_lines)
|
|
dbgln("- {}", line);
|
|
dbgln(" added:");
|
|
for (const auto& line : hunk.added_lines)
|
|
dbgln("+ {}", line);
|
|
}
|
|
}
|
|
|
|
return hunks;
|
|
}
|
|
|
|
HunkLocation parse_hunk_location(const String& location_line)
|
|
{
|
|
size_t char_index = 0;
|
|
struct StartAndLength {
|
|
size_t start { 0 };
|
|
size_t length { 0 };
|
|
};
|
|
auto parse_start_and_length_pair = [](const String& raw) {
|
|
auto index_of_separator = raw.index_of(",").value();
|
|
auto start = raw.substring(0, index_of_separator);
|
|
auto length = raw.substring(index_of_separator + 1, raw.length() - index_of_separator - 1);
|
|
auto res = StartAndLength { start.to_uint().value() - 1, length.to_uint().value() - 1 };
|
|
return res;
|
|
};
|
|
while (char_index < location_line.length() && location_line[char_index++] != '-') {
|
|
}
|
|
VERIFY(char_index < location_line.length());
|
|
|
|
size_t original_location_start_index = char_index;
|
|
|
|
while (char_index < location_line.length() && location_line[char_index++] != ' ') {
|
|
}
|
|
VERIFY(char_index < location_line.length() && location_line[char_index] == '+');
|
|
size_t original_location_end_index = char_index - 2;
|
|
|
|
size_t target_location_start_index = char_index + 1;
|
|
|
|
char_index += 1;
|
|
while (char_index < location_line.length() && location_line[char_index++] != ' ') {
|
|
}
|
|
VERIFY(char_index < location_line.length());
|
|
|
|
size_t target_location_end_index = char_index - 2;
|
|
|
|
auto original_pair = parse_start_and_length_pair(location_line.substring(original_location_start_index, original_location_end_index - original_location_start_index + 1));
|
|
auto target_pair = parse_start_and_length_pair(location_line.substring(target_location_start_index, target_location_end_index - target_location_start_index + 1));
|
|
return { original_pair.start, original_pair.length, target_pair.start, target_pair.length };
|
|
}
|
|
|
|
void HunkLocation::apply_offset(size_t offset, HunkLocation::LocationType type)
|
|
{
|
|
if (type == LocationType::Original || type == LocationType::Both) {
|
|
original_start_line += offset;
|
|
original_length -= offset;
|
|
}
|
|
if (type == LocationType::Target || type == LocationType::Both) {
|
|
target_start_line += offset;
|
|
target_length -= offset;
|
|
}
|
|
}
|
|
|
|
};
|