Another nice date.
I needed to increment the spillover region (the free space needed in between combinator rewrites) to 32kB, which was minor. With, I guess, an average of five cells, each eight bytes, you get forty bytes for each average allocated construct, which means now eight hundred somewhat constructs can be allocated. Not sure I am just fixing a heap growth bug.
Still stuck with some kind of escape bug, which I am clueless where it comes from. I, again, patched some C code by hand.
Patched, it does the full semantical analysis of the system file. Some stuff is really fast but the prover is way to slow to be usuable. Even when kind checking, unifying small constraints like v0 -> # = * -> v1, it proceeds with about one rule each 300ms or something. It fails somewhere after the type checker.
Type checking is faster than kind checking??? I guess I pass to many constraints during kindification?
It fails during the AST emitting phase with a corrupted heap...
I extended the tests on escaped pieces of texts. I now print an AST node before serializing to track where it goes wrong. For the rest, just waiting for slow compiles. I should use my printer again.
(God, I guess I figured out the escape bug. It doesn't make sense to patch a compiler, and forget to recompile relevant individual modules before linking...)
Another long wait, implemented a new performance test from the computer languages shootout.
No comments:
Post a Comment