Rubinius hangs when setting a key of a `Concurrent::Map`

[philr]

Rubinius hangs when setting a key of a `Concurrent::Map`

[brixen]

Hey @philr thanks for opening the issue. Have you been able to debug this at all? Would you be interested in digging into it if I can help get you started?

[philr]

Other than running strace to find it was stuck waiting on a futex, I've not done any debugging. I'm happy to run commands to capture some stack traces, process dumps, etc, but I don't really have the time to dig into this right now.

[brixen]

@philr thanks again for opening the issue. A bit more explanation of the issue: recently I added the ability to tag nil values so that locating where a nil value originated (which is often distant from where a NoMethodError may be raised) would be possible. Unfortunately, this requires handling all equality checks for nil in a special way. This was one of the checks that was missed.

Another change that I've been slowly making is to replace all "primitive" functions with proper instructions so that the semantics of a program can be fully understood by only looking at how source code is compiled to the instruction set. If this had already been completed, this sort of defect (ie missing a special check for nil) wouldn't occur.

One other thing, what you observed in the strace was likely not be "stuck" in a futex, but repeatedly entering a futex because the naive implementation of CAS (compare and swap) simply loops optimistically expecting eventually to achieve the update operation, but that would never happen in the case of a tagged nil. If a debugger had been available, it would likely have been very easy for you to identify this looping behavior. I've started work on a new console interface (ie replacement for IRB) that includes a debugging mode to make this easy.