On Mon, 29 Aug 2011, Stéphane Ducasse wrote:
Roel typer does inference using method bodies and I do not think that it takes the receiver. Now the problem is that we could take the expression (here 100) and evaluate and get its class (and catch error in case you wrote jlkkj
Now what if the expression should not be evaluated twice.
So type inferencer
May be if we had a message:
expr isLiteral we could guess a bit better.
So if someone want to experiment please do.
ECompletion already does that. See ECContext >> #guessTypeOf:. Levente
Stef
Cool! That would make a big difference. Where are they/what are they called?
I know of Roel Wuyts' RoelTyper (http://decomp.ulb.ac.be/roelwuyts/smalltalk/roeltyper/) and Lex Spoon's Chuck (http://www.lexspoon.org/ti/).
forget it. It was nice but you need 80 mb or more of cache and I'm not sure the code is somewhere and working
Matt Might has also written about k-CFA (Control Flow Analysis) for inferring types in a dynamic language (Scheme, in particular) - http://matt.might.net/articles/implementation-of-kcfa-and-0cfa/ - which we might be able to apply to Smalltalk.
well this is not something you do in 3 months.
frank
Sean
-- View this message in context: http://forum.world.st/Pharo-1-4-tp3759684p3774973.html Sent from the Pharo Smalltalk mailing list archive at Nabble.com.