freenode/#clasp - IRC Chatlog
Search
15:01:21
drmeister
Hello - clasp with MPS appears pretty stable. I've been working with it for the last day - no crashes that aren't clearly something else.
15:04:07
drmeister
stassats`: I don't want to cut that chord yet. I can build the Boehm version to run the static analyzer to build the MPS version.
15:12:28
Bike
so the way an actual build works is cboehm is built (boehm not requiring any information about the internals of C++ structs), the static analyzer is run in cboehm to analyze the interpreter and maybe llvm source, that information is used to build cmps?
16:06:39
drmeister
Shinmera: That was true until about a month ago - then I enhanced the scraper to assign sorted stamps to exposed C++ classes.
16:07:52
drmeister
So the boehm build doesn't need anything from the static analyzer. The mps build still needs the object layouts/fixable pointer field offsets that can only be obtained from the static analyzer.
16:08:57
drmeister
I was able to simplify the build system by getting rid of the 'boehmdc' build and now the 'boehm' build takes its place.
16:11:09
drmeister
I've wondered this for a long time though - the MPS GC supports (requires) conservative scanning of the stack. Why can't I use this conservative scanner to scan blocks of memory conservatively like the Boehm garbage collector does?
16:12:12
drmeister
The Ravenbrook folks tell me that they have always talked about a "boehm-like mode" for MPS as an easier on-ramp to MPS for programmers - but never implemented it.
16:13:50
drmeister
ACTION is still quite impressed by how robust the Boehm GC has been and how well it has worked for Clasp so far.