[covr.record_tests] segfault when tracing R6 active bindings to parquet data

[dgkf]

Note

This is in code I introduced and only affects very specific scenarios when options(covr.record_tests) is enabled.
I plan to fix this

Example at dgkf/covr-r6-active-binding-test

git clone https://github.com/dgkf/covr-r6-active-binding-test
cd covr-r6-active-binding-test
R -e 'options(covr.record_tests = TRUE)' -e 'covr::package_coverage()'

Details

From what I can tell, there are a few key features of this package that culminate in this bug:

• An R6 active binding is included in the test call stack
  • For reasons that are still unclear to me, the sys.calls() captured in update_current_test(), when capturing an active binding, includes the R function, not just the call. Critically, this function brings its calling environment.
• When the R6 object environment includes a parquet dataset (I'm assuming due to the way arrow works with the in-memory representation of the data), this environment segfaults when serialized to disk.

Proposed solution

• Will first try to drill into why this is grabbing a function. There might be a more holistic way to avoiding this altogether.
• If there's no better way to avoid functions in the call stack, then before recording the call stack, if a frame contains a function, set its environment to emptyenv() so it can be serialized.

[dgkf]

Just following up to say that I have a safeguard implemented in dgkf/covr@573-record-tests-arrow-segfault.

I can open a PR whenever you'd like, but I'd prefer to test it in some more real-world applications first to make sure there aren't any severe performance implications and that no necessary data gets thrown out.