There was some loss - mainly source code for personal hacking. My Jython UnicodeData implementation was lost, although I did have a copy of that in my GMail Sent folder which has been contributed to the committers in the hope that someone with more time and enthusiasm than me finds it useful. Apparently they have. The other thing lost was my WideFinder sketching. I surprisingly did have an early warning that the disk was close to failure, when my implementation (which was clocking ten seconds or something) suddenly started taking twenty minutes to complete. Obviously I didn't pick up on this warning; instead blaming it on some horrible mistake in my code. Some learnings from my WideFinder experience.
- NIO can be tricky, and should probably have an easier API in comparison to classical IO. It was nice to look at this anyway, and I'll probably have a read of the Jython and JRuby IO stuff to see how others do it, and also the Apache NIO stuff.
- Java string matching is slow using regexp. I'd not got as far as implementing Boyer-Moore or similar, but I expected to get a big hike out of my 10 second time when switching to that.
- CAS is good, versus synchronized blocks. Brian Goetz has talked about the advantages of non-blocking forms elsewhere.
- Java is quite restrictive. I must have been doing too much Common Lisp, Erlang and Scheme recently!
No comments:
Post a Comment