[Mac] Allow NSExceptions in certain cases.

printing/printing_context_mac.mm

Index: printing/printing_context_mac.mm
diff --git a/printing/printing_context_mac.mm b/printing/printing_context_mac.mm
index a8e4fa048d4a1bfdde5908762788d363e08258b7..6b09a94c28520e8cfe0dcb3746dc3b7ffea6608b 100644
--- a/printing/printing_context_mac.mm
+++ b/printing/printing_context_mac.mm
@@ -9,6 +9,7 @@
 
 #include "base/logging.h"
 #include "base/mac/scoped_cftyperef.h"
+#include "base/mac/scoped_nsexception_enabler.h"
 #include "base/sys_string_conversions.h"
 #include "base/values.h"
 #include "printing/print_settings_initializer_mac.h"
@@ -36,6 +37,15 @@ void PrintingContextMac::AskUserForSettings(gfx::NativeView parent_view,
                                             int max_pages,
                                             bool has_selection,
                                             PrintSettingsCallback* callback) {
+  // Third-party print drivers seem to be an area prone to raising exceptions.
+  // This will allow exceptions to be raised, but does not handle them.

[stuartmorgan, 2011/05/17 18:52:14]

I don't understand. Why would we *deliberately* allow exceptions to blow through
C++ stacks, when we know that will cause weird crashes we won't understand? This
seems like the worst possible behavior.

[Scott Hess - ex-Googler, 2011/05/17 19:34:40]

The primary benefit of making NSExceptions fatal is that it gives us fast
turnaround when _we_ write bad code.  My short-term goal is to get as many
exceptions fatal as possible for as much of the code as possible.  As my TODO
indicates, it's not entirely clear to me how we can generically deal with the
issue in this case, but it is pretty clear that the printing subsystem is one
area where we see exceptions that we aren't (apparently) at fault for.

If the sub-system is handling them correctly, then this change is a reasonable
change.  If not, we'll still need to track down why they are being raised and
what a reasonable way to address it is (I'm sure that will be FUN!).

+  // TODO(shess): Some experimentation indicates that uncaught exceptions can
+  // leave the print panel up, and any use after the exception will almost
+  // certainly crash on access to zombie objects, but simply _catching_ the
+  // exception doesn't resolve the problem, because you cannot easily dismiss
+  // the print panel from code.  This situation doesn't seem right, to me.
+  base::mac::ScopedNSExceptionEnabler enabler(true);
+
   DCHECK([NSThread isMainThread]);
 
   // We deliberately don't feed max_pages into the dialog, because setting