Improve traceback

[hadley]

Fixes #214

[hadley]

At least we now get the actual error traceback. Is there some obvious way to trim off some of the earlier frames?

[lionel-]

We can no longer use frame environments as boundaries (or top_env) because we no longer evaluate in an eval() stack frame. We might have to introduce a new way of setting a boundary in the call stack...

[hadley]

Should we create an evaluate_ wrapper that we can call from evaluate() so we still get one regular frame on the call stack?

[hadley]

Hmmm that doesn't make any difference.

[lionel-]

Did you set the rlang_trace_top_env option (from memory) in that wrapper?

[hadley]

I tried this in both evaluate() and evaluate_():

old <- options(rlang_trace_top_env = environment())
defer(options(old))

[hadley]

I also wonder if this is the root cause of #219

[lionel-]

We can no longer use frame environments as boundaries (or top_env) because we no longer evaluate in an eval() stack frame. We might have to introduce a new way of setting a boundary in the call stack...

[lionel-]

@hadley Locally with this fix I don't see any changes to the backtraces of unexpected errors.

However with this change I do see the same backtrace as you do:

@@ -31,10 +31,11 @@ condition_handlers <- function(watcher, on_error, on_warning, on_message) {
       cnd <- sanitize_call(cnd)
       watcher$push(cnd)
 
-      switch(on_error,
+      switch(
+        on_error,
         continue = invokeRestart("eval_continue"),
         stop = invokeRestart("eval_stop"),
-        error = invokeRestart("eval_error", cnd)
+        error = NULL  # Decline to handle
       )
     }
   )

[lionel-]

Worth noting that having the evaluate machinery as part of the backtrace is consistent with the current CRAN behaviour.

[lionel-]

#218 (comment)