Larry McVoy's Love/Hate relationship with Tcl gives birth to L

Larry McVoy gave the first afternoon talk. I missed the first couple of minutes of the beginning of it. Again, here are my notes:

Why not use someone elses $SCRIPTING_LANGUAGE. None can be extended from C as easily as Tcl. "Python is the obvious choice but has crappy bindings to Tk." Java slow startup [and something else I didn't write down in time]. Says he tried to convince Sun originally to make Tk be the GUI for Java.

Why not pure Tcl? No C-like structs in Tcl. Associative arrays don't count. No lint like tools in Tcl (dynamic languages make that impossible to get 100% right) No type checking. Optimizing Tcl is harder than L (L is a "weaker" language, easier to optimize).

Why not C & native GUIs? No need to assemble a Tk + Title + ... + distro. GUI functionality—it's still hard to beat the canvas and text widgets. Development costs are at least 6x higher (3x engineering cost, 3x QA cost) for native code. They target a lot of platforms: Windows, MacOS, X11, and maybe java. More variations == more support costs. Goal of BitMover is to drive the company to where they need zero support.

Basic types: int, float, string like C (mostly). variable interpolation requires ${foo} or really just any L expression ${1+2}. :magic_constants such as: text(".t", :bg, "white", :fg, "black");

Locals get initialized to zero, the empty string, whatever is appropriate. Strings are first class objects. No character type ("we are unicode, what the hell is the length of a character?")

Associate arrays (swiped Perl syntax for initializing them). Keys and values are string. Tcl dicts under the covers.

C like arrays (int v[]). They look like C but don't need specified size. They autogrow on assignment.

Structs: currently like C but may get initializers. Tcl lists under the covers. Structure fields are in "::L::struct eg" (for introspection).

Pass by value as expected like Tcl & C.

By reference: automatic for hashes and arrays. Needs & for other types. Implemented as syntactic sugar using upvar. L pointers used where upvar won't work.

New Tcl Object:

struct pointer {
        int    depth; //upvar #depth
        string name;  //tcl var pointed to
        string index; // optional index
    };

Used in places like: checkbutton(".f.v1", :variable, &foo.v);

Control flow: Usual C stuff plus, perl regex statements, iterating over hashes. Some minor sugar: unless {expr} stmt; until {expr} stmt;

Switches cases may be constats, regexps, or globs.

loops for hashes. iterating over an array of whatever. Looking for a better syntax than: foreach (h as key => val) {...}

Top-level compilation. Because there is no L interpreter path.

Changes to the Tcl Parser. #lang(tcl), #lang(L) to switch languages. If filename is not .L (.l ?) then assume Tcl mode. When you see #lang(L), grabs until eof or #lang(tcl) and feeds it to L. Don't need to match curly braces.

Left to do: Type checking isn't done; They haven't written substantial programs in it; Plan to rewrite their [BitMover's] GUIs in L; Check back next year...

Future work: scoping, pre-compiled modules, optimizations, debugging, dynamic type checking.

Last slide: L@bitmover.com. Apparently they have a Wiki somewhere, but it only has one page at the moment, and they didn't mention what the URL was.

Gerald Lester asked about: foreach i in v { ... }. Unless Larry has studied that carefully will he remember that i is an index value or a value.

Richard Hipp: So does the name of the language stand for Larry? Larry: It doesn't stand for "Loser". Earlier names: LORD (Larry Oscar Richard D____), HINT (Hint Is Not Tcl). Open to suggestions.

License is Tcl license, no strings attached.

Guiding principles: if we can't compelling use case we just don't. (Larry has to be able to understand it.) Slammed C++ here (wish I'd got it word for word—very inflammably quotable ;-)

Gerald: You didn't mention catch or anything like it; do you guys just not need it because you don't make errors? Larry: need some experience first. Pointed out that anything that is a Tcl procedure [command] can be called by L, so can use [catch].


—Michael A. Cleverly

Comments:

  1. Pythoneer wrote (at Wed, 11 Oct 2006, 16:17):

"Python is the obvious choice but has crappy bindings to Tk."

Yeah, it uses Tcl for the glue layer. Tcl? How crappy.

  1. Robert wrote (at Fri, 13 Oct 2006, 15:25):

Was he making some kind of point or just using a Tcl forum to push his own cr*p?

The dig at Python is unwaranted. Python is moving to "wx" anyway.

Permanent URL for this post: http://blog.cleverly.com/permalinks/249.html