hacks.pod

Prodding by Matt Diephouse to generate documentation for two things:

1. Things that partcl has done that might be considered hacks - things that will likely impair our ability in the distant future to cleanly do language interoperability.

2. Things that partcl hasn't done yet at all because parrot makes them hard.

It is, of course, quite likely that said feature already exists with a nice interface, and was just not found by the partcl developers - If you find something along these lines, let them know.

WORKAROUNDS

subroutine meta information.

Tcl provides the ability to query subroutines about their arguments and their code. For example:

% proc sum {{a 2} {b 3}} {
return [expr $a + $b]
}
% sum 
5
% info args sum
a b
% info body sum

 return [expr $a + $b]

% info default sum a the_argument
1
% puts $the_argument
2
%

Right now, partcl is using globals to store the argument list and the the body information. It would seem that we'd want to have a builtin mechanism for querying the subs directly rather than storing this information separately.

user defined subs; parameter handling.

Tcl provides nice exceptions for dealing with argument handling - to continue our example:

% sum a b c
wrong # args: should be "sum ?a? ?b?"

Right now, this is handled inside the function definition - it is defined to take an arbitary number of args, and then is checked for correctness internally. It would be nice to have invoke() automatically figure this out and generate an exception that Tcl can use. (This also starts to drag in "how to do Tcl exceptions cleanly from parrot")

[error], [catch], [break], continue...

Tcl kind of conflates normal returns and exceptional returns, and uses the same mechanism for [break]ing and [continue]'ing out of loops.

Right now, this is being handled by returning *two* values from each publicly exposed function. (And from several internal ones, as well.) The first is the code, that is, an integer, which is a faked up enum of:

TCL_OK, TCL_ERROR, TCL_BREAK, TCL_CONTINUE, TCL_RETURN

Then, the actual interesting value is returned as as a PMC. How do we tie in the two happy return cases, the two cases that interact with loops, and the error case into parrot's normal return/exception handler in a way that works for tcl and will allow interoperability

Note: The reason we're not using parrot exceptions is that we're unable at the moment to redefine them ourselves - I don't see any easy way to get the nice stacktraces that tcl gives you using the parrot exception handling mechanism. So we could potentially use exceptions for TCL_ERROR in the long term. But then [catch] will need to be re-written to cope with exceptions or not (as opposed to now, where all codes are equal).

stack depth

Cheating and keeping a global around right now, so we can figure out if we should be using a global or a lexical (and if so, how far down or up).

[trace]'ing

There are two ways we can go about the tracing - either we can keep the information about the traces in a global, and check that global every time the appropriate action is taken; or, we can tie that information into the actual commands and variables. I think the latter would be somewhat cleaner, and, if any other languages support this feature, this would give us a chance to interoperate.