in src/parser/ArcanistDiffParser.php [556:809]
protected function parseIndexHunk(ArcanistDiffChange $change) {
$is_git = $this->getIsGit();
$is_mercurial = $this->getIsMercurial();
$is_svn = (!$is_git && !$is_mercurial);
$move_source = null;
$line = $this->getLine();
if ($is_git) {
do {
$patterns = array(
'(?P<new>new) file mode (?P<newmode>\d+)',
'(?P<deleted>deleted) file mode (?P<oldmode>\d+)',
// These occur when someone uses `chmod` on a file.
'old mode (?P<oldmode>\d+)',
'new mode (?P<newmode>\d+)',
// These occur when you `mv` a file and git figures it out.
'similarity index ',
'rename from (?P<old>.*)',
'(?P<move>rename) to (?P<cur>.*)',
'copy from (?P<old>.*)',
'(?P<copy>copy) to (?P<cur>.*)',
);
$ok = false;
$match = null;
foreach ($patterns as $pattern) {
$ok = preg_match('@^'.$pattern.'@', $line, $match);
if ($ok) {
break;
}
}
if (!$ok) {
if ($line === null ||
preg_match('/^(diff --git|commit) /', $line)) {
// In this case, there are ONLY file mode changes, or this is a
// pure move. If it's a move, flag these changesets so we can build
// synthetic changes later, enabling us to show file contents in
// Differential -- git only gives us a block like this:
//
// diff --git a/README b/READYOU
// similarity index 100%
// rename from README
// rename to READYOU
//
// ...i.e., there is no associated diff.
// This allows us to distinguish between property changes only
// and actual moves. For property changes only, we can't currently
// build a synthetic diff correctly, so just skip it.
// TODO: Build synthetic diffs for property changes, too.
if ($change->getType() != ArcanistDiffChangeType::TYPE_CHANGE) {
$change->setNeedsSyntheticGitHunks(true);
if ($move_source) {
$move_source->setNeedsSyntheticGitHunks(true);
}
}
return;
}
break;
}
if (!empty($match['oldmode'])) {
$change->setOldProperty('unix:filemode', $match['oldmode']);
}
if (!empty($match['newmode'])) {
$change->setNewProperty('unix:filemode', $match['newmode']);
}
if (!empty($match['deleted'])) {
$change->setType(ArcanistDiffChangeType::TYPE_DELETE);
}
if (!empty($match['new'])) {
// If you replace a symlink with a normal file, git renders the change
// as a "delete" of the symlink plus an "add" of the new file. We
// prefer to represent this as a change.
if ($change->getType() == ArcanistDiffChangeType::TYPE_DELETE) {
$change->setType(ArcanistDiffChangeType::TYPE_CHANGE);
} else {
$change->setType(ArcanistDiffChangeType::TYPE_ADD);
}
}
if (!empty($match['old'])) {
$match['old'] = self::unescapeFilename($match['old']);
$change->setOldPath($match['old']);
}
if (!empty($match['cur'])) {
$match['cur'] = self::unescapeFilename($match['cur']);
$change->setCurrentPath($match['cur']);
}
if (!empty($match['copy'])) {
$change->setType(ArcanistDiffChangeType::TYPE_COPY_HERE);
$old = $this->buildChange($change->getOldPath());
$type = $old->getType();
if ($type == ArcanistDiffChangeType::TYPE_MOVE_AWAY) {
$old->setType(ArcanistDiffChangeType::TYPE_MULTICOPY);
} else {
$old->setType(ArcanistDiffChangeType::TYPE_COPY_AWAY);
}
$old->addAwayPath($change->getCurrentPath());
}
if (!empty($match['move'])) {
$change->setType(ArcanistDiffChangeType::TYPE_MOVE_HERE);
$old = $this->buildChange($change->getOldPath());
$type = $old->getType();
if ($type == ArcanistDiffChangeType::TYPE_MULTICOPY) {
// Great, no change.
} else if ($type == ArcanistDiffChangeType::TYPE_MOVE_AWAY) {
$old->setType(ArcanistDiffChangeType::TYPE_MULTICOPY);
} else if ($type == ArcanistDiffChangeType::TYPE_COPY_AWAY) {
$old->setType(ArcanistDiffChangeType::TYPE_MULTICOPY);
} else {
$old->setType(ArcanistDiffChangeType::TYPE_MOVE_AWAY);
}
// We'll reference this above.
$move_source = $old;
$old->addAwayPath($change->getCurrentPath());
}
$line = $this->nextNonemptyLine();
} while (true);
}
$line = $this->getLine();
if ($is_svn) {
$ok = preg_match('/^=+\s*$/', $line);
if (!$ok) {
$this->didFailParse(pht(
"Expected '%s' divider line.",
'======================='));
} else {
// Adding an empty file in SVN can produce an empty line here.
$line = $this->nextNonemptyLine();
}
} else if ($is_git) {
$ok = preg_match('/^index .*$/', $line);
if (!$ok) {
// TODO: "hg diff -g" diffs ("mercurial git-style diffs") do not include
// this line, so we can't parse them if we fail on it. Maybe introduce
// a flag saying "parse this diff using relaxed git-style diff rules"?
// $this->didFailParse("Expected 'index af23f...a98bc' header line.");
} else {
// NOTE: In the git case, where this patch is the last change in the
// file, we may have a final terminal newline. Skip over it so that
// we'll hit the '$line === null' block below. This is covered by the
// 'git-empty-file.gitdiff' test case.
$line = $this->nextNonemptyLine();
}
}
// If there are files with only whitespace changes and -b or -w are
// supplied as command-line flags to `diff', svn and git both produce
// changes without any body.
if ($line === null ||
preg_match(
'/^(Index:|Property changes on:|diff --git|commit) /',
$line)) {
return;
}
$is_binary_add = preg_match(
'/^Cannot display: file marked as a binary type\.$/',
rtrim($line));
if ($is_binary_add) {
$this->nextLine(); // Cannot display: file marked as a binary type.
$this->nextNonemptyLine(); // svn:mime-type = application/octet-stream
$this->markBinary($change);
return;
}
// We can get this in git, or in SVN when a file exists in the repository
// WITHOUT a binary mime-type and is changed and given a binary mime-type.
$is_binary_diff = preg_match(
'/^(Binary files|Files) .* and .* differ$/',
rtrim($line));
if ($is_binary_diff) {
$this->nextNonemptyLine(); // Binary files x and y differ
$this->markBinary($change);
return;
}
// This occurs under "hg diff --git" when a binary file is removed. See
// test case "hg-binary-delete.hgdiff". (I believe it never occurs under
// git, which reports the "files X and /dev/null differ" string above. Git
// can not apply these patches.)
$is_hg_binary_delete = preg_match(
'/^Binary file .* has changed$/',
rtrim($line));
if ($is_hg_binary_delete) {
$this->nextNonemptyLine();
$this->markBinary($change);
return;
}
// With "git diff --binary" (not a normal mode, but one users may explicitly
// invoke and then, e.g., copy-paste into the web console) or "hg diff
// --git" (normal under hg workflows), we may encounter a literal binary
// patch.
$is_git_binary_patch = preg_match(
'/^GIT binary patch$/',
rtrim($line));
if ($is_git_binary_patch) {
$this->nextLine();
$this->parseGitBinaryPatch();
$line = $this->getLine();
if (preg_match('/^literal/', $line)) {
// We may have old/new binaries (change) or just a new binary (hg add).
// If there are two blocks, parse both.
$this->parseGitBinaryPatch();
}
$this->markBinary($change);
return;
}
if ($is_git) {
// "git diff -b" ignores whitespace, but has an empty hunk target
if (preg_match('@^diff --git .*$@', $line)) {
$this->nextLine();
return null;
}
}
if ($this->isRCS) {
// Skip the RCS headers.
$this->nextLine();
$this->nextLine();
$this->nextLine();
}
$old_file = $this->parseHunkTarget();
$new_file = $this->parseHunkTarget();
if ($this->isRCS) {
$change->setCurrentPath($new_file);
}
$change->setOldPath($old_file);
$this->parseChangeset($change);
}