What should the default of `CHPL_RT_UNWIND` be

[jabraham17]

We have been discussing changing the default of CHPL_UNWIND, which is the compile time option for whether to build Chapel with stack trace support. See #27926. However, a parallel issue if we change the default of CHPL_UNWIND to be !=none is what is the default value of CHPL_RT_UNWIND.

CHPL_UNWIND controls the compile time ability to have stack traces, changing it requires rebuilding the runtime. But once the compile time ability to have stack traces is turned on, CHPL_RT_UNWIND controls the runtime decision to display those stack traces.

The current state is

• CHPL_RT_UNWIND defaults to true, meaning stack traces will be printed if possible
• Users can turn it off by setting CHPL_RT_UNWIND to false
• Our test system explicitly sets CHPL_RT_UNWIND to false for ease of testing

I think the no-brainer is that our test system should primarily test the default configuration. But the important question is whats the default.

Should CHPL_RT_UNWIND default to true or false? This would change if we default to printing a stack trace or not.

I will also note that using an environment variable to toggle between CHPL_RT_UNWIND is a bit unsatisfying to me.

[jabraham17]

My feeling is that CHPL_RT_UNWUND should default to true. While its true a user may be overwhelmed by the stack trace or find it hard to read, I think there are other ways to solve that than just turning it off (e.g. formatting, color/text emphasis, etc). And I think there is potential benefit for the following situation

• I am using a long running application with user inputs
• I somehow create a situation where I cause a crash
• If there is a stack trace, I can use that to identify the issue
• If there is no stack trace, I first have to determine what series of steps are needed to reproduce the crash. Then actually rerun the application and reproduce it. This may be non-trivial

[bradcray]

I'm OK with either default as a starting approach if the default prints information about how to switch to the other mode when relevant (e.g., when a stack trace is or would be printed).

[jabraham17]

if the default prints information about how to switch to the other mode when relevant (e.g., when a stack trace is or would be printed).

Regardless of what we decide here, I've opened #28031 to do exactly this. Right now, pending decision on this issue, it tells users how to turn the stack traces off (and does nothing if users explicitly request stack traces)

[jabraham17]

We discussed this in the Chapel meeting today, we are going to keep the default of CHPL_RT_UNWIND=true