Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
In ARROW-9235 we have a problem where we need to evaluate code (may be C++ that calls the R API in a way that may allocate R memory or literal R code) on the R main thread. As discussed in the pull request ( https://github.com/apache/arrow/pull/12323#issuecomment-1054726309 ), there are other times that this has come up and will come up in the future (e.g., evaluating user-defined R functions from the query engine, implementing extension types from R).
After discussing with westonpace, it seems like it is worth investing in a SafeCallFromR() C++ function that lives in the R bindings. The initial approach (that may need to be altered as we go) will be to
- have a global object that is initialized at the beginning of a C++ call from R to specific functions (and released on exit) that maintains some list of tasks that need to run on the R main thread and an event loop to periodically evaluate them while background tasks are running.
- define a SafeCallFromR() C++ function that adds a task to the global object and waits for the result to come back.
(with apologies to Weston if I mischaracterized our conversation!)
Attachments
Issue Links
- links to