refactor(api): use a unpacker based on libmpack instead of msgpack-c

Currently this is more or less a straight off reimplementation,
but this allow further optimizations down the line, especially
for avoiding memory allocations of rpc objects.

Current score for "make functionaltest; make oldtest" on a -DEXITFREE build:

is 117 055 352 xfree(ptr != NULL) calls (that's NUMBERWANG!).

1 files changed, 17 insertions, 0 deletions

diff --git a/src/nvim/rbuffer.c b/src/nvim/rbuffer.c
index d280e08c03..6407ac172e 100644
--- a/src/nvim/rbuffer.c
+++ b/src/nvim/rbuffer.c
@@ -154,6 +154,23 @@ void rbuffer_consumed(RBuffer *buf, size_t count)
   }
 }
 
+/// Use instead of rbuffer_consumed to use rbuffer in a linear, non-cyclic fashion.
+///
+/// This is generally usefull if we can guarantee to parse all input
+/// except some small incomplete token, like when parsing msgpack.
+void rbuffer_consumed_compact(RBuffer *buf, size_t count)
+  FUNC_ATTR_NONNULL_ALL
+{
+  assert(buf->read_ptr <= buf->write_ptr);
+  rbuffer_consumed(buf, count);
+  if (buf->read_ptr > buf->start_ptr) {
+    assert((size_t)(buf->read_ptr - buf->write_ptr) == buf->size);
+    memmove(buf->start_ptr, buf->read_ptr, buf->size);
+    buf->read_ptr = buf->start_ptr;
+    buf->write_ptr = buf->read_ptr + buf->size;
+  }
+}
+
 // Higher level functions for copying from/to RBuffer instances and data
 // pointers
 size_t rbuffer_write(RBuffer *buf, const char *src, size_t src_size)