Changeset 332

Timestamp:

05/23/07 07:04:58 (2 years ago)

Author:

blackhedd

Message:

Moved network initialization from EventMachine_t::Run to EventMachine_t::EventMachine_t.
This fixes a bug. We were initializing loop breakers before initializing the network
library. The bug was masked because Ruby initializes the network library itself.
Thanks to Bill Kelly (billk . . cts.org) for pointing this out.

Files:

• version_0/ext/em.cpp (modified) (3 diffs)

version_0/ext/em.cpp

@@ -53 +53 @@
-        // objects before we start running.
-        gCurrentLoopTime = time(NULL);
+
+        /* We initialize the network library here (only on Windows of course)
+         * and initialize "loop breakers." Our destructor also does some network-level
+         * cleanup. There's thus an implicit assumption that any given instance of EventMachine_t
+         * will only call ::Run once. Is that a good assumption? Should we move some of these
+         * inits and de-inits into ::Run?
+         */
+        #ifdef OS_WIN32
+        WSADATA w;
+        WSAStartup (MAKEWORD (1, 1), &w);
+        #endif
-
-        _InitializeLoopBreaker();
@@ -128 +139 @@
-     * to that user. There is no meaningful implementation (and no error)
-     * on Windows. On Unix, a failure to setuid the caller-supplied string
-us
+here
-     * in order to fulfill a security requirement. If we fail silently,
-     * the user may continue to run with too much privilege.
+     *
+     * TODO, we need to decide on and document a way of generating C++ level errors
+     * that can be wrapped in documented Ruby exceptions, so users can catch
+     * and handle them. And distinguish it from errors that we WON'T let the Ruby
+     * user catch (like security-violations and resource-overallocation).
+     * A setuid failure here would be in the latter category.
-     */
-
@@ -220 +237 @@
-{
-        #ifdef OS_WIN32
-        WSADATA w;
-        WSAStartup (MAKEWORD (1, 1), &w);
-        HookControlC (true);
-        #endif