From 436ec57f830ec3ac8319b368e4b8fc375d3931b9 Mon Sep 17 00:00:00 2001
From: Andreas Rumpf <rumpf_a@web.de>
Date: Wed, 13 Mar 2019 10:05:05 +0100
Subject: [PATCH] seqs.nim: added a remark about C++ optimizers

---
 lib/core/seqs.nim | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/lib/core/seqs.nim b/lib/core/seqs.nim
index be2cdcf1ac..fd46b96eb7 100644
--- a/lib/core/seqs.nim
+++ b/lib/core/seqs.nim
@@ -36,6 +36,14 @@ when false:
   proc `=trace`[T](s: NimSeqV2[T]) =
     for i in 0 ..< s.len: `=trace`(s.data[i])
 
+#[
+Keep in mind that C optimizers are bad at detecting the connection
+between ``s.p != nil`` and ``s.len != 0`` and that these are intermingled
+with user-level code that accesses ``s.len`` only, never ``s.p`` directly.
+This means the check for whether ``s.p`` needs to be freed should
+be ``s.len == 0`` even though that feels slightly more awkward.
+]#
+
 proc `=destroy`[T](s: var seq[T]) =
   var x = cast[ptr NimSeqV2[T]](addr s)
   var p = x.p