handle case where coverage object has 0 tests

Previously, the $tests element being non-null was used to indicate that
a coverage object was produced without tests. However, if no test were
executed, these coverage traces may still need to be pruned of
allocated, but unused matrix rows. Also migrated default test index from
'0' to 'NA_integer_'. In nearly all cases this will have no effect, but
has been rarely observed for tests which perform interrupts.

Author: dgkf

R/covr.R

@@ -642,7 +642,8 @@ merge_coverage.list <- function(coverage_objs) {
 #   may have been allocated, but not entirely populated.
 #
 clean_coverage_tests <- function(obj) {
-  if (is.null(obj$tests)) return()
+  counter_has_tests_tally <- function(counter) !is.null(counter$tests)
+  if (is.na(Position(counter_has_tests_tally, obj))) return()
 
   for (i in seq_along(obj)) {
     if (is.null(val <- obj[[i]]$value)) next

R/trace_tests.R

@@ -145,7 +145,7 @@ new_test_counter <- function(key) {
   .counters[[key]]$tests$.data <- vector("integer", 3L)
   .counters[[key]]$tests$.value <- integer(1L)
   .counters[[key]]$tests$tally <- matrix(
-    0L,
+    NA_integer_,
     ncol = 3L,
     # initialize with 4 empty rows, only expanded once populated
     nrow = 4L,