The lvalue substr doesn't actually return the string but a special lvalue indicating the offset and length. Creating this and setting it to '' involves a fair amount of overhead.

Which is actually more efficient will probably depend on how much you are truncating.

It was originally part of a log reader. I found that reading in blocks and working pattern matching on those blocks was much more efficient than on line by line. Obviously you can get partial lines, so I rindex the last /n, save the fragment at the end for the next pass and return the main block truncated. It works like a charm. I was reinventing it and seing how I could make it more efficient.

Then you probably want to use 4-arg substr to truncate the buffer and save the fragment all at once:

$end = rindex($buffer, "\n");
if ($end < 0) { whoops, incomplete line }
$fragment = substr($buffer, $end+1, length($buffer), '');
[download]

Note that length($buffer) is bigger than the actual length(buffer)-$end-1 you want, but specifying a length larger than is available is guaranteed to not cause problems; you'll just get what is there. If the buffer ends with "\n", you'll get an empty $fragment.