wip support different types of time_step/time_type

R/slide.R

@@ -567,16 +567,32 @@ epi_slide_mean = function(x, col_name, ..., before, after, ref_time_values,
   #   after <- time_step(after)
   # }
 
-  # time_step can be any of `c("days", "weeks", "months", "quarters", "years")`
-  all_dates <- seq(min(x$time_value), max(x$time_value), by = time_step)
+  all_dates <- as.Date(c())
+  pad_early_dates <- as.Date(c())
+  pad_late_dates <- as.Date(c())
 
-  pad_early_dates <- c()
-  pad_late_dates <- c()
-  if (before != 0) {
-    pad_early_dates <- all_dates[1L] - before:1
-  }
-  if (after != 0) {
-    pad_late_dates <- all_dates[length(all_dates)] + 1:after
+  # Find `by` corresponding to epi_df `time_type` or `time_step`.
+  by <- ...
+
+  if (!is.null(by)) {
+    # time_step can be any of `c("days", "weeks", "months", "quarters", "years")`
+    # tidyr::full_seq(c(min(x$time_value), max(x$time_value)), period = 1) ?
+    all_dates <- seq(min(x$time_value), max(x$time_value), by = by)
+
+    if (before != 0) {
+      pad_early_dates <- Start(all_dates) - before:1
+    }
+    if (after != 0) {
+      pad_late_dates <- End(all_dates) + 1:after
+    }
+  } else {
+    Warn(
+      c(
+        "`frollmean` requires a full window to compute a result. Add NA values
+        to beginning and end of the group so that we get results for the
+        first `before` and last `after` elements."
+      )
+    )
   }
 
   # `frollmean` is 1-indexed, so create a new window width based on our