gcc 11.x fixesneedlefish_13.93.0needlefish/13.93.0marlin_12.93.0marlin_12.92.0marlin_12.91.0marlin_12.90.1marlin/12.93.0marlin/12.92.0marlin/12.91.0marlin/12.90.113.93.012.93.012.92.012.91.012.90.1

Changes for compiling with gcc 11.x:
- g++ now seems to instantiate duplicate entries for the set member
  function in the contextclass template class in binding-wrap.hpp.
  The use of a closure as a default argument value seems to be the
  culprit, as it seems there are longstanding issues with respect to
  using closures like that and resulting symbol names (i.e. the use
  of a closure isn't necessarily recognized as generating unique
  instantiations).  In theory, C++17 should explicitly allow this
  when the closure has no captures, but bumping up to -std=gnu++17
  did not fix the issue.  To avoid it, replace the closure usage
  with a private static member function.
- In afb-hook.c, tweaked the ignoring of the writev return code to
  make the stricter checking in gcc 11 happy.
- In decode_base64 in wrap-json.c, initialize u16 as gcc now seems
  to miss that it will be initialized on the first loop iteration.

Bug-AGL: SPEC-3819

Signed-off-by: Scott Murray <scott.murray@konsulko.com>
Change-Id: I8876db196b46cc35ecd2798e20d5ec3425df865e

1 files changed, 2 insertions, 1 deletions

diff --git a/include/afb/c++/binding-wrap.hpp b/include/afb/c++/binding-wrap.hpp
index f94d1bc5..b798cf11 100644
--- a/include/afb/c++/binding-wrap.hpp
+++ b/include/afb/c++/binding-wrap.hpp
@@ -300,6 +300,7 @@ public:
 		friend class req;
 		afb_req_t req_;
 		contextclass(afb_req_t r) : req_(r) {}
+		static void default_destroyer(T*t) { delete t; }
 
 	public:
 		inline operator T *() const { return get(); }
@@ -312,7 +313,7 @@ public:
 					nullptr));
 		}
 
-		inline void set(T *value, void (*destroyer)(T*) = [](T*t){delete t;}) const {
+		inline void set(T *value, void (*destroyer)(T*) = default_destroyer) const {
 			afb_req_context(req_, 1,
 				nullptr,
 				reinterpret_cast<void(*)(void*)>(destroyer),