doc: explain call-stack-depth logic (#436)

Author: Stebalien

fvm/src/call_manager/default.rs

@@ -86,6 +86,19 @@ where
     where
         K: Kernel<CallManager = Self>,
     {
+        // We check _then_ set because we don't count the top call. This effectivly allows a
+        // call-stack depth of `max_call_depth + 1` (or `max_call_depth` sub-calls). While this is
+        // likely a bug, this is how NV15 behaves so we mimic that behavior here.
+        //
+        // By example:
+        //
+        // 1. If the max depth is 0, call_stack_depth will be 1 and the top-level message won't be
+        //    able to make sub-calls (1 > 0).
+        // 2. If the max depth is 1, the call_stack_depth will be 1 in the top-level message, 2 in
+        //    sub-calls, and said sub-calls will not be able to make further subcalls (2 > 1).
+        //
+        // NOTE: Unlike the FVM, Lotus adds _then_ checks. It does this because the
+        // `call_stack_depth` in lotus is 0 for the top-level call, unlike in the FVM where it's 1.
         if self.call_stack_depth > self.machine.config().max_call_depth {
             return Err(
                 syscall_error!(LimitExceeded, "message execution exceeds call depth").into(),