Hangout 31 August 2017
- gnn@ is going to pick a modern BNFish to specify the D language in the OpenDTrace spec
domagoj@ has been writing a bunch of tests, including LLVM FileCheck tests for the DTrace compiler. Has determined that the compiler generates unoptimized code, with an overhead of 3 instructions per statement. This looks like a good target for optimization. Has found that shift logical right never works except for unsigned values and that arithmetic shift right only works with signed values. Has updated the assembler to spit out code that might wind up in the specification. Has been trying to write down his thoughts on how to rework DTrace to be more portable, for instance to CHERI.
- samuel@ says that they are trying to spec out a way to make provider removal, which happens at each process exist, more efficient. The current system uses a linear search on removal and that has a measurable effects on systesm.
- arun@ has been hacking on a Rust dtrace(1)
- arun@ asks if FreeBSD would accept a patch that allowed WITH_CDDL_HEADERS and gnn@ says he'll see why the headers do not get installed by default
- We still need to find the right person at Apple to talk to about the DTrace audit provider. Samuel says he'll try to followup.
- gnn@ will look at Tatsiana's reliability patches and send the link to Samuel. Samuel has the link but wants to measure the actual effect on the system.
- samuel@ is looking at moving the xnu DTrace into a KEXT
- Someone at Apple is looking into a CTF to DWARF shim. We'll know more about this at the next hangout or the one after.