Dan Luu
Dan Luu
This goes away if I limit the expression length so that the arena allocator doesn't run out of memory. This seems to indicate that we have non-deterministic behavior when we...
I checked this with valgrind because it's pretty good about finding uses of uninitialized memory and it didn't find anything even in the case where we blow up the allocator....
See also #49
``` // TODO: cases for (signed, unsigned) x (1, 2, 4, 8), float ```
I keep cleaning out the analogous warnings from BitFunnel, but now that we've integrated NativeJIT with BitFunnel it's going to be hard to stay warning clean without fixing these.
The next obvious warning to clean out is the precision loss / sign extension stuff. The last time I went through those I made some change that caused a subtle...
Part of this is that we should have a version of the build that runs clang's asan, msan, etc., which we're not doing now.
For clang asan, msan, etc., [this link looks helpful](https://github.com/arsenm/sanitizers-cmake/).
The following options work (independently) to enable clang's sanitizers: ``` add_compile_options("-fsanitize=memory") add_compile_options("-lubsan -fsanitize=undefined") add_compile_options("-lasan -fsanitize=address") add_compile_options("-ltsan -fsanitize=thread") ``` I'll add these at some point, but before I make that change...
Those options work for `NativeJIT`, but the linker stuff fails with `BitFunnel`. Apparently, we're doing something different for the two projects? When trying to build with thread sanitizer, `BitFunnel` fails...