Depends on D19829. Ref T13216. See PHI908. The current implementation is kind of a lot to live in CommitHookEngine and will likely fail if git diff-tree produces more than 2GB of output.

Pull it out and make it slightly more robust against enormous commits. It's probably limited by this, now:

implode("\n", $every_path)

We could replace that with some PhutilReverseRopeSource primitive or something but since we don't have one of those and it seems unlikely that we'll hit this case in practice, I left it here for now with just the easy stuff converted to be stream-oriented.

Used this script to test the query against various commits, got good results:

<?php

require_once 'scripts/init/init-script.php';

$viewer = PhabricatorUser::getOmnipotentUser();

$repository = id(new PhabricatorRepositoryQuery())
  ->setViewer($viewer)
  ->withCallsigns(array('P'))
  ->executeOne();

var_dump(
  id(new DiffusionLowLevelFilesizeQuery())
    ->setRepository($repository)
    ->withIdentifier($argv[1])
    ->execute());

Used this to find large commits in history and pull filesizes (worked great, although our largest commit only touches a couple thousand paths):

for hash in `git log --format=%H`; do echo -n $hash; echo -n ' '; git diff-tree -r --no-commit-id $hash | wc -l | awk '{print $1}'; done | awk '{print $2 " " $1}' | sort -n