No subject

1. this test program doesn't really match the code in knit.py that was
discussed recently since it has a list of strings to start with.  So
the best solution for fastimport might be different to the best
solution for knit.py.

2. Rather than measuring the final memory usage, you might want to use
a tool like valgrind's massif to look for peaks over the program's
run.

3. for the use case of reading blobs from a fast-import input stream,
the best solution might be to read the entire blob (since the size is
included in the stream) then count line endings afterwards.

James.