(no subject)

[chrisamaphone]

by the way, if you ever want a backtrace in smlnj, here's the ticket:

    CM.make "$smlnj-tdp/back-trace.cm";
    SMLofNJ.Internals.TDP.mode := true;

apparently everyone i've talked to has always been as lazy as i am about doing a google search to discover this, so i thought i'd point it out.

Comments

[ikeepaleopard]

That is good to know. I am surprised I had not heard this before. That said, it seems to collapse recursive calls, which makes it moderately less useful:

- fun f 0 = raise Fail "foo" | f n = f (n - 1);
val f = fn : int -> 'a
- f 10;
*** BACK-TRACE ***
CALL-( stdIn:4.5-4.45: f[2]
          (from: stdIn:1.1-1.11: it)
...
- f 1;
*** BACK-TRACE ***
CALL-( stdIn:4.5-4.45: f[2]
          (from: stdIn:1.1-1.11: it)
...

[gwillen.livejournal.com]

That's almost surely because it's doing tail call elimination. Try a recursive call in a non-tail position.

[ikeepaleopard]

Yeah, if you change it to a standard factorial function it gives the whole trace.

[demarko (from livejournal.com)]

really useful!

[mogrothir.livejournal.com]

I love how you have a line I can read at the beginning, two lines I can read at the end, and then in the middle there's the completely unintelligible moon-language stuff. What the hell.

[chrisamaphone]

the stuff in the middle doesn't need to be intelligible; you just type it into the interpreter and it magically does the right thing :P

[mogrothir.livejournal.com]

...The who what things?

It looks like you're speaking English, and then all this computer-babble comes out.

[jgrafton]

clearly at the next party-thing you hold I need to teach you about sml and type theory :D

[mogrothir.livejournal.com]

Something tells me that next party-thing I hold, I will come away with a headache and this vague feeling that my principles have been violated in the dead of night. o_O

[gwillen.livejournal.com]

Holy shit you can get SMLNJ to give backtraces?? That would have made it infinitely more useful for, like everything ever. Why the hell isn't that documented? Or like, on by default?

[chrisamaphone]

a good question indeed!

[wjl.livejournal.com]

as for "on by default", it probably makes generated code less efficient, and possibly dramatically so.

[ikeepaleopard]

I think what I read said twice as slow. Also it makes for cluttered.

[techstep.livejournal.com]

I've been tossing around the idea of relearning SML (I took and bombed 212 about a decade ago, which only hastened my dropping out). But I'm not sure what it will give me that my current tools (R and Stata for statistical analysis, bits of perl for datafile transformation) don't provide, especially in regards to econometric modeling.

Alternately, I guess I could relearn it for the intellectual thrill. (The downside of that is, of course, I have so much to learn already, especially since I'm prepping for grad school.)

Thoughts?

[chrisamaphone]

well, i'm always a proponent of people learning ml. at least for the perl stuff (R and Stata are domain specific, so they're probably the easiest tools for that stuff), it's just as easy to slurp and tokenize a file in ML as it is in perl, and then once you have the data in your hands, you have a much nicer language for actually manipulating it. if you're thinking "what about regular expressions?" you may be interested to learn about micmatch (for ocaml), which leverages pattern matching in a really pretty way.

i don't really know what econometric modelling entails, but wjl can tell you about this finanical company he worked at a couple summers ago where they code everything in ocaml (mostly analyzing and displaying sales data, afaik). and they love it.

there are just a lot of basic language features that are *so useful* (polymorphism, type inference, sum types, pattern matching, persistent data structures...) for doing even really basic things, but you don't realize you wanted them until you had them... and then you miss them when you go back to other languages.