Daniel Xu

https://github.com/iovisor/bpftrace/pull/2964 fixes the issue with the minimized repro. But turns out the issue w/ the original script in this issue is something else entirely. It's still not quite clear to...

Ok, I have a better idea now. I added some printfs: https://pastes.dxuuu.xyz/b1ze81 This is the bad output: ``` DEBUG: not pointee: struct task_struct DEBUG: is pointee: struct files_struct * DEBUG:...

@jordalgo if you're interested, this project is something we could fairly trivially parallelize. AOT has been a long requested feature -- many folks have told me they don't use bpftrace...

Awesome! I'm a bit tight on time this week (in bay, actually) but next week should be good. I think after the first item is done, all further work is...

I'm not sure that we considered it but my gut feeling is that generating LLVM IR is better. It gives us much more control. You'll see stuff in the kernel...

Agreed that feature probing is more robust. Distros do backports all the time without necessarily bumping the version code. Perhaps something like weak symbols for probes might make sense. In...

Mind sharing the full fd list? From that truncated list it looks both reasonable and difficult to avoid: we need a FD for each prog as well as a perf...

> > also, a benefit of cat being so simple is that you avoid potentially weird behaviour around moved files. If we open at program loading we have to check...

I don't have anything authoritative to say but here are some of my thoughts: * I think it depends on if the verifier is smart enough to figure out the...

Seems reasonable. Will need a design before committing to it -- this could complicate the runtime/codegen bits quite a bit if done poorly