On Fri, Sep 21, 2012 at 5:03 AM, <r...
> do you have the updated version?
No, not yet. I am still thinking about what the right resizing
strategy is. I can make the code much faster on the benchmark by
estimating the new array size based on fields/byte processed so far,
cutting out almost all reallocations, but that gives it the chance to
wildly overestimate and overallocate what it would otherwise need. The
current code, while slower by being two passes, at least does not
overallocate. I'm still not quite sure how to balance those. Probably
I will give up that particular speedup and go back to ordinary append.