As I was writing file TODO.md I was trying to figure out how fast or
slow I actually am at programming and how predictable this is.
Ur-Scheme was 1553 lines of Scheme and mostly took me four weeks and
three days, from February 3 to March 4 of 02008, exactly a
person-month. David A. Wheeler's "SLOCCount" says it should have
taken 3.8 person-months using the basic COCOMO formula 2.4 *
(KSLOC**1.05), so let's suppose that instead of being the famous "10x
programmer" I am a 3.8x programmer, not compared to normal modern
programmers but to whatever losers the COCOMO model was calibrated on.
As another more recent data point, Dercuano's genpdf.py took me 5 days (0.25 person-months), and it's 550 lines of code, suggesting 1.28 person-months --- a productivity factor of about 5x for me. 5 is pretty close to 3.8.
I'd like to evaluate StoneKnifeForth's development speed in this way, but I don't have any reasonable way to do so, since I don't know how to evaluate either how many lines of code it is or how long it took me to write.
SLOCCount says that the part of BubbleOS I have written so far should have taken 17 person-months since it contains 6473 lines of code, although about 800 of those are the actuarial tables in the death clock Toki. In fact, I wrote most of it from October 12, 02018, to February 22, 02019, which is almost four months; this is also lower than SLOCCount's estimate by only a factor of about 3-5.
I wrote Dumpulse mostly October 15-17, 02017, about 0.14 person-months. Checking out the commit 52f10e5a5d22c9ef8f78992cdc95cbdb8ed4ee79 I get 646 lines of code, nominally 1.52 person-months. In this case the multiplier is closer to 10x. I think this is partly because I'd already been thinking and talking about how to do it and partly because I was able to stay pretty focused for three days --- although the git commits cluster into only four or five hours on each of those days.
I might be able to speed things up by taking advantage of new programming technology like generative testing, or by choosing especially conservative and well-understood designs.