If `n` is the number of `substr` replace operations, the old
implementation did `n+1` allocations and always first copied start,
then replacement, then the remaining (unsearched) part of the original
string.
Now calculate the number of occurences of `substr` before allocating the
resulting string.
Change the algorithm to parse through the original string and only
copy the parts that must be kept into the resulting string plus the
replacements where they belong at.
Now we also only have to go through the original string twice plus we only
do a single allocation.
Signed-off-by: Steffen Jaeckel <jaeckel-floss@eyet-services.de>
d3b6ebb1d7