I have searched about this, but I couldn't find why StringBuilder's ensureCapacity() method won't lengthen the old capacity by just doubling but also adding two.
So, when default capacity of 16 is full, next lengthened value will be 34 unless whole string length doesn't exceed 34. Why shouldn't it be 32?
My best guess is considering for a null character, '\u0000', but I'm not sure. Can anyone tell me why?
I believe it has to do with a simple, if somewhat dumb, way to ensure the corner case of very small strings.
For example, if I have the string
""
and I double it only, I will not have a sufficient size to store anything else in it. If I double it and add a small constant number of spaces, I can assure that my new value is larger than my old one.
Why increment it by two then? Probably a small performance improvement. By adding two instead of 1, I can avoid an intermediate expansion for small expansions (0 to 10 chars detailed below)
"" => expand => "1" => expand => "123" expand => "1234567" expand => "123456789012345"
which is 4 expands compared to
"" => expand => "12" => expand => "123456" => expand => "123456789012"
which is 3 expands. This also works nicely for one char strings (expanding to 10 chars)
"1" => expand => "1234" => expand => "1234567890"
while the 1 char expansion routine looks like
"1" => expand => "123" => expand => "1234567" => expand => "123456789012345"
Finally, an added increment of two tends to word align about 50% of the time, while added increments of one or three would do so about 25% of the time. While this might not seem like a big deal, some architectures cannot accommodate non-aligned reads without expensive interrupt calls to rewrite the read in the CPU, leading to all sorts of performance issues.