I bring you, "Greetings!", from the Year 2016! (You are NOT--in a million years--going to believe who's running for President!! (well, one of them, anyway...))
I am time-traveling back to investigate what may have happened to RobA's random-density-fill.scm script. Reference is made in this thread to two versions (types, really) of the script: I gather that the original paints in random locations, while a later version aligns locations to a (pre-defined, likely) grid.
Lyle's creation appears to have used the former and PhotoMaster's more regular composition the latter (although his later question suggests he may not have been aware of it).
But I did not come to sow discord here in the past, but only to report that my problem in the future will be (er, "will have been"??) with a version identified internally as "v2.1". I also came back to see if Rod ever figured it out, but I see that his problem was eerily similar to mine.
I did have this working in the past... er, well, "less distant future", but like so many scripts, when I went to use it again, only 6 or so years later, it seems to be a different beast--or I am. The (non)-behavior, for me, with v2.1 is that I set up the white-and-gray-on-black density layer as a density map, and then create an equally sized white or transparent map above it, and select it as the active layer, and then run the random-density-map filter/script on it with a new red cute-little-star brush, and? nada. zippo. dudzo. I'm left with an unblemished new layer, white or transparent as the case may be. (may will have been, in the future: 2016)
In the other, more complicated, xcf--the one I really want it to work for, I was getting a different error: the one Rob reported here, that the active and density-map layers' sizes did not match. That left me crest-fallen, as I had made both as new layers, w/o touching size parameters. At one point, I even duped the map layer to make a new layer and filled it with white, but with similar disappointment as reward.
However, finding mention here of other versions led me to look inside my (GIMP 2.8--yes, it gets THAT far!!!) scripts/ folder where I found I had even earlier[1] found versions 1.0 and 2.0. v2.0 worked for me in the ways described.
1.0 did also. It has a plainer set of options, including a "number of objects to paint", which I prefer but ignorantly set to 5000 as a first test, and a tell-tale user-setting for a "Bailout Threshold" as well--of which one couldn't ask a better hint of original authorship...

).
I do think I detect a "pattern" to the randomness, in this fashion: in order to monitor progress, I chose to draw only 100 objects, and then used Ctrl-F to repeat applications of the same filter. "I thought I saw" a preference for the filter to place new objects near existing ones; that is to say, I think I see arcs and lines forming--but that may just be the brain doin its thing too.
(Like the original Rob, I'm trying to paint forests, but to give a bumpy impression of them in heightmaps carved on CNC.)
[1] "early/late" soon have little meaning for us time-travelers. and hallucinators. ...and forum readers.