private function breakHunkIntoSmallHunks(ArcanistDiffHunk $base_hunk)
{
$context = 3;
$results = array();
$lines = phutil_split_lines($base_hunk->getCorpus());
$n = count($lines);
$old_offset = $base_hunk->getOldOffset();
$new_offset = $base_hunk->getNewOffset();
$ii = 0;
$jj = 0;
while ($ii < $n) {
// Skip lines until we find the next line with changes. Note: this skips
// both ' ' (no changes) and '\' (no newline at end of file) lines. If we
// don't skip the latter, we may incorrectly generate a terminal hunk
// that has no actual change information when a file doesn't have a
// terminal newline and not changed near the end of the file. 'patch' will
// fail to apply the diff if we generate a hunk that does not actually
// contain changes.
for ($jj = $ii; $jj < $n; ++$jj) {
$char = $lines[$jj][0];
if ($char == '-' || $char == '+') {
break;
}
}
if ($jj >= $n) {
break;
}
$hunk_start = max($jj - $context, 0);
// NOTE: There are two tricky considerations here.
// We can not generate a patch with overlapping hunks, or 'git apply'
// rejects it after 1.7.3.4.
// We can not generate a patch with too much trailing context, or
// 'patch' rejects it.
// So we need to ensure that we generate disjoint hunks, but don't
// generate any hunks with too much context.
$old_lines = 0;
$new_lines = 0;
$hunk_adjust = 0;
$last_change = $jj;
$break_here = null;
for (; $jj < $n; ++$jj) {
if ($lines[$jj][0] == ' ') {
if ($jj - $last_change > $context) {
if ($break_here === null) {
// We haven't seen a change in $context lines, so this is a
// potential place to break the hunk. However, we need to keep
// looking in case there is another change fewer than $context
// lines away, in which case we have to merge the hunks.
$break_here = $jj;
}
}
if ($jj - $last_change > ($context + 1) * 2) {
// We definitely aren't going to merge this with the next hunk, so
// break out of the loop. We'll end the hunk at $break_here.
break;
}
} else {
$break_here = null;
$last_change = $jj;
if ($lines[$jj][0] == '\\') {
// When we have a "\ No newline at end of file" line, it does not
// contribute to either hunk length.
++$hunk_adjust;
} else {
if ($lines[$jj][0] == '-') {
++$old_lines;
} else {
if ($lines[$jj][0] == '+') {
++$new_lines;
}
}
}
}
}
if ($break_here !== null) {
$jj = $break_here;
}
$hunk_length = min($jj, $n) - $hunk_start;
$count_length = $hunk_length - $hunk_adjust;
$hunk = new ArcanistDiffHunk();
$hunk->setOldOffset($old_offset + $hunk_start - $ii);
$hunk->setNewOffset($new_offset + $hunk_start - $ii);
$hunk->setOldLength($count_length - $new_lines);
$hunk->setNewLength($count_length - $old_lines);
$corpus = array_slice($lines, $hunk_start, $hunk_length);
$corpus = implode('', $corpus);
$hunk->setCorpus($corpus);
$results[] = $hunk;
$old_offset += $jj - $ii - $new_lines;
$new_offset += $jj - $ii - $old_lines;
$ii = $jj;
}
return $results;
}
protected function parseChangeset(ArcanistDiffChange $change)
{
// If a diff includes two sets of changes to the same file, let the
// second one win. In particular, this occurs when adding subdirectories
// in Subversion that contain files: the file text will be present in
// both the directory diff and the file diff. See T5555. Dropping the
// hunks lets whichever one shows up later win instead of showing changes
// twice.
$change->dropHunks();
$all_changes = array();
do {
$hunk = new ArcanistDiffHunk();
$line = $this->getLineTrimmed();
$real = array();
// In the case where only one line is changed, the length is omitted.
// The final group is for git, which appends a guess at the function
// context to the diff.
$matches = null;
$ok = preg_match('/^@@ -(\\d+)(?:,(\\d+))? \\+(\\d+)(?:,(\\d+))? @@(?: .*?)?$/U', $line, $matches);
if (!$ok) {
// It's possible we hit the style of an svn1.7 property change.
// This is a 4-line Index block, followed by an empty line, followed
// by a "Property changes on:" section similar to svn1.6.
if ($line == '') {
$line = $this->nextNonemptyLine();
$ok = preg_match('/^Property changes on:/', $line);
if (!$ok) {
$this->didFailParse(pht('Confused by empty line'));
}
$line = $this->nextLine();
return $this->parsePropertyHunk($change);
}
$this->didFailParse(pht("Expected hunk header '%s'.", '@@ -NN,NN +NN,NN @@'));
}
$hunk->setOldOffset($matches[1]);
$hunk->setNewOffset($matches[3]);
// Cover for the cases where length wasn't present (implying one line).
$old_len = idx($matches, 2);
if (!strlen($old_len)) {
$old_len = 1;
}
$new_len = idx($matches, 4);
if (!strlen($new_len)) {
$new_len = 1;
}
$hunk->setOldLength($old_len);
$hunk->setNewLength($new_len);
$add = 0;
$del = 0;
$hit_next_hunk = false;
while (($line = $this->nextLine()) !== null) {
if (strlen(rtrim($line, "\r\n"))) {
$char = $line[0];
} else {
// Normally, we do not encouter empty lines in diffs, because
// unchanged lines have an initial space. However, in Git, with
// the option `diff.suppress-blank-empty` set, unchanged blank lines
// emit as completely empty. If we encounter a completely empty line,
// treat it as a ' ' (i.e., unchanged empty line) line.
$char = ' ';
}
switch ($char) {
case '\\':
if (!preg_match('@\\ No newline at end of file@', $line)) {
$this->didFailParse(pht("Expected '\\ No newline at end of file'."));
}
if ($new_len) {
$real[] = $line;
$hunk->setIsMissingOldNewline(true);
} else {
$real[] = $line;
$hunk->setIsMissingNewNewline(true);
}
if (!$new_len) {
break 2;
}
break;
case '+':
++$add;
--$new_len;
$real[] = $line;
break;
case '-':
if (!$old_len) {
// In this case, we've hit "---" from a new file. So don't
// advance the line cursor.
$hit_next_hunk = true;
break 2;
}
++$del;
--$old_len;
$real[] = $line;
break;
case ' ':
if (!$old_len && !$new_len) {
break 2;
}
--$old_len;
--$new_len;
$real[] = $line;
break;
default:
// We hit something, likely another hunk.
$hit_next_hunk = true;
break 2;
}
}
if ($old_len || $new_len) {
$this->didFailParse(pht('Found the wrong number of hunk lines.'));
}
$corpus = implode('', $real);
$is_binary = false;
if ($this->detectBinaryFiles) {
$is_binary = !phutil_is_utf8($corpus);
$try_encoding = $this->tryEncoding;
if ($is_binary && $try_encoding) {
$is_binary = ArcanistDiffUtils::isHeuristicBinaryFile($corpus);
if (!$is_binary) {
$corpus = phutil_utf8_convert($corpus, 'UTF-8', $try_encoding);
if (!phutil_is_utf8($corpus)) {
throw new Exception(pht("Failed to convert a hunk from '%s' to UTF-8. " . "Check that the specified encoding is correct.", $try_encoding));
}
}
}
}
if ($is_binary) {
// SVN happily treats binary files which aren't marked with the right
// mime type as text files. Detect that junk here and mark the file
// binary. We'll catch stuff with unicode too, but that's verboten
// anyway. If there are too many false positives with this we might
// need to make it threshold-triggered instead of triggering on any
// unprintable byte.
$change->setFileType(ArcanistDiffChangeType::FILE_BINARY);
} else {
$hunk->setCorpus($corpus);
$hunk->setAddLines($add);
$hunk->setDelLines($del);
$change->addHunk($hunk);
}
if (!$hit_next_hunk) {
$line = $this->nextNonemptyLine();
}
} while (preg_match('/^@@ /', $line));
}
protected function parseChangeset(ArcanistDiffChange $change)
{
$all_changes = array();
do {
$hunk = new ArcanistDiffHunk();
$line = $this->getLine();
$real = array();
// In the case where only one line is changed, the length is omitted.
// The final group is for git, which appends a guess at the function
// context to the diff.
$matches = null;
$ok = preg_match('/^@@ -(\\d+)(?:,(\\d+))? \\+(\\d+)(?:,(\\d+))? @@(?: .*?)?$/U', $line, $matches);
if (!$ok) {
// It's possible we hit the style of an svn1.7 property change.
// This is a 4-line Index block, followed by an empty line, followed
// by a "Property changes on:" section similar to svn1.6.
if ($line == '') {
$line = $this->nextNonemptyLine();
$ok = preg_match('/^Property changes on:/', $line);
if (!$ok) {
$this->didFailParse("Confused by empty line");
}
$line = $this->nextLine();
return $this->parsePropertyHunk($change);
}
$this->didFailParse("Expected hunk header '@@ -NN,NN +NN,NN @@'.");
}
$hunk->setOldOffset($matches[1]);
$hunk->setNewOffset($matches[3]);
// Cover for the cases where length wasn't present (implying one line).
$old_len = idx($matches, 2);
if (!strlen($old_len)) {
$old_len = 1;
}
$new_len = idx($matches, 4);
if (!strlen($new_len)) {
$new_len = 1;
}
$hunk->setOldLength($old_len);
$hunk->setNewLength($new_len);
$add = 0;
$del = 0;
$advance = false;
while (($line = $this->nextLine()) !== null) {
if (strlen($line)) {
$char = $line[0];
} else {
$char = '~';
}
switch ($char) {
case '\\':
if (!preg_match('@\\ No newline at end of file@', $line)) {
$this->didFailParse("Expected '\\ No newline at end of file'.");
}
if ($new_len) {
$real[] = $line;
$hunk->setIsMissingOldNewline(true);
} else {
$real[] = $line;
$hunk->setIsMissingNewNewline(true);
}
if (!$new_len) {
$advance = true;
break 2;
}
break;
case '+':
if (!$new_len) {
break 2;
}
++$add;
--$new_len;
$real[] = $line;
break;
case '-':
if (!$old_len) {
break 2;
}
++$del;
--$old_len;
$real[] = $line;
break;
case ' ':
if (!$old_len && !$new_len) {
break 2;
}
--$old_len;
--$new_len;
$real[] = $line;
break;
case '~':
$advance = true;
break 2;
default:
break 2;
}
}
if ($old_len != 0 || $new_len != 0) {
$this->didFailParse("Found the wrong number of hunk lines.");
}
$corpus = implode("\n", $real);
$is_binary = false;
if ($this->detectBinaryFiles) {
$is_binary = !phutil_is_utf8($corpus);
if ($is_binary && $this->tryEncoding) {
$is_binary = ArcanistDiffUtils::isHeuristicBinaryFile($corpus);
if (!$is_binary) {
// NOTE: This feature is HIGHLY EXPERIMENTAL and will cause a lot
// of issues. Use it at your own risk.
$corpus = mb_convert_encoding($corpus, 'UTF-8', $this->tryEncoding);
if (!phutil_is_utf8($corpus)) {
throw new Exception('Failed converting hunk to ' . $this->tryEncoding);
}
}
}
}
if ($is_binary) {
// SVN happily treats binary files which aren't marked with the right
// mime type as text files. Detect that junk here and mark the file
// binary. We'll catch stuff with unicode too, but that's verboten
// anyway. If there are too many false positives with this we might
// need to make it threshold-triggered instead of triggering on any
// unprintable byte.
$change->setFileType(ArcanistDiffChangeType::FILE_BINARY);
} else {
$hunk->setCorpus($corpus);
$hunk->setAddLines($add);
$hunk->setDelLines($del);
$change->addHunk($hunk);
}
if ($advance) {
$line = $this->nextNonemptyLine();
}
} while (preg_match('/^@@ /', $line));
}