freenode/#clasp - IRC Chatlog
Search
9:17:52
beach
minion: memo for Bike: I fixed some problems in the figures, improved the layout, and added three more. So now they are fig1.pdf ... fig7.pdf and they show a complete inlining. Though I cheat a bit with the call to ERROR to simplify things a bit.
12:45:04
beach
CST-to-AST is shaping up. I am working on writing test cases for the remaining special operators. In doing that, I will no doubt discover defects and omissions, but I am definitely making good progress.
13:32:16
drmeister
My silence on this work you are doing is not because of disinterest. I'm biting my tongue and looking forward to the future.
13:34:14
drmeister
I'm shaking out some bugs in the build system - interactions between fastgf and the new backtrace system. fastgf generates lots of jitted symbols and they have to look right to the backtrace code.
13:36:17
Bike
re: error: generate-ast currently checks for null return type annotations. might not be the best place for it, but it's difficult to get all the type stuff in one place
13:36:17
minion
Bike, memo from beach: I fixed some problems in the figures, improved the layout, and added three more. So now they are fig1.pdf ... fig7.pdf and they show a complete inlining. Though I cheat a bit with the call to ERROR to simplify things a bit.
13:40:21
Bike
yeah, i recognize xfig. I understood the ERRO block as just being shorthand for the ERROR call, since you could just replace that with a subgraph for an actual call no problem.
13:49:27
beach
Bike: I am pretty sure that the inlining technique could not handle assignments to existing local variables. If I am right, this is another argument in favor of something SSA-like.
13:51:17
beach
If the setq were to be copied to the caller, it would be a violation of the semantics of a function call.
13:52:52
beach
Anyway, I am going to work on several more examples like that, but it is reassuring to know that the technique seems to work for simple system functions such as CAR and CDR.