evan_tech -- inventing your own language

03:42 pm, 17 Nov 03

inventing your own language

Discussion on the ll1 list about the dangers of creating application-specific languages. make, of course, is the classic example of something that wasn't really fully planned out, then gradually extended, and now is a mess. Graydon's talk is relevant, too. But I liked this bit, from Dan Sugalski:

What I find most effective (modulo heckling from the LL audience here) is "Larry and Bjarne were both language design pros, and look what they ended up with. And you think you can do better?"

brad	Heh.
reply to this

evan	the s1->s2 transition is a pretty good example of this: s1 was too limited, so s2 is a "real" language. i'm thankful you didn't try to hack more functionality into s1...
reply to this

mart
The S2 language is still kinda wacky, though! :)

reply to this

billemon	Third time's the charm ... ;) Actually S2 isn't that bad, but ... somehow kinda limited at the same time, for someone used to "real" languages like perl etc. Is it likely there will be a S3 anytime, or will S2 maybe evolve a bit over time?
reply to this

mart

Lots of S2's limitations are intentional, for security/efficiency reasons. These will likely never be changed.

Is there any specific limitation that's bothering you? It's possible that it could be fixed, assuming it wasn't deliberate and that fixing it wouldn't create any issues.

reply to this

billemon

Lots of odd things that are hard if you end up trying to write "real code" .. I ended up writing a replacement colour class (to do HSV instead of HSL) and found a few odd limitations. "constructor" functions don't appear to actually work, or be very useful. Otherwise, just odd things that were really hard, like I also ended up writing code for "#rrggbb" -> rgb triple and back. Sometimes when I tried to paste two strings together the ones with leading 0's seemed to get treated as integers and lose the 0 (makes concatenating a hex string hard e.g. "F00F0F" was becoming F0FF :)

reply to this

mart

Yeah, the constructor stuff is kinda wacky. I don't think it actually really does anything. The only time the Color construtor is used is when a Color property gets exposed in the “friendly” interface, and that's a special case in the UI code. I've certainly never used constructors in S2 other than that.

Rather than creating your own color class, it might be nicer to simply extend the existing one with a layer-defined method, which has to be named something starting with lay_ (which is ugly, of course, but avoids namespace conflicts later). This way you get all of the existing behavior for free. I won't ask why you want HSV instead of HSL, since I suspect as a programmer I wouldn't understand such issues!

As for the string concatenation stuff, I would consider that to be a bug. A string in S2 should never be implicitly cast into an integer, and I don't think I've ever seen it do that. If you can produce a minimal test case showing the problem and post about it on Zilla someone will probably fix it.

reply to this

billemon

The reasons I replaced the class in the end were: 1. I wanted to make the HSV the "canonical" representation so that I didn't get strangeness converting rgb->hsv->rgb every time I wanted to change something; 2. the scaling of the hsv values is a bit unintuitive (read: didn't work as expected) and 3. the set_hsv function doesn't work at all, returning an error (something about _v not existing in a hash?).

Reason being, I wanted to take a single colour for the highlights on a page and automatically generate the different accents by shifting hue, saturation or value appropriately. I used to use the "bumblebee" colourscheme, but some of my friends didn't like the yellow, and I got the idea stuck in my head of basically shifting the hue but keeping the overall scheme .. which fell by the wayside a bit, but I liked the idea.

reply to this

jeffr	I thought perl was the classic example of something that wasn't really fully planned out but was gradually extended?
reply to this

evan	Yeah, that's a good point. C++, too, to the extent it's an extension of C (and to the extent that they gradually added features-- I don't know any details of this, but it seems some of the crazy stuff (see also: my post with eight-byte pointers) couldn't have been in there from the beginning).
reply to this

mart	The S2 language is still kinda wacky, though! `:)`
reply to this