extend conclusions

Author: mattip

pycon2017_cffi.ipynb

@@ -5498,15 +5498,24 @@
    "source": [
     "Things to think about, besides speed:\n",
     "\n",
-    "* Maintainability  \n",
-    "* Compiler dependency  \n",
-    "* Suseptability to bugs (object lifetimes, signature mismatches)  \n",
+    "* Maintainability\n",
+    "  - What happens when the C code changes?\n",
+    "* Compiler dependency\n",
+    "  - ctypes needs none, CFFI has two modes, Cython requires one\n",
+    "* Suseptability to bugs (object lifetimes, signature mismatches)\n",
+    "  - All use a minilanguage for interfacting, only CFFI's is standard C\n",
+    "  - Cython will handle most transformations automatically\n",
+    "  - CFFI can be tricky for C-level pointers\n",
+    "* Speed and productivity\n",
+    "  - Cython is heavily optimized, tightly integrated to the C-API\n",
+    "  - If the headers are pure C, CFFI should be simple\n",
+    "  - Projects exist to generate wrappers for all three\n",
     "* Which technology is actively maintained (ctypes went into the stdlib to die?)\n",
     "\n",
     "And now the pop-quiz. If we run the pure python version in PyPy what time will we get?:\n",
     "\n",
     "* Around a 2X speed up\n",
-    "* About like Cython or CFFI calling mandel 1.5e6 times)\n",
+    "* About like Cython or CFFI calling mandel 1.5e6 times\n",
     "* About like C compiled -O3"
    ]
   },