[haiku-development] hrev45320: Yet another non-obvious effect of -ftree-vrp "optimization"
- From: "Siarzhuk Zharski" <zharik@xxxxxx>
- To: "Haiku Development" <haiku-development@xxxxxxxxxxxxx>
- Date: Thu, 14 Mar 2013 14:12:02 +0100 (CET)
Dear Colleagues,
during last week I have observed that my gcc4 hybrid builds of Haiku cannot
start the application server. The app server has just failed into gdb and left
me no other option as pressing the Reboot key.
The backtrace of this fail was following:
strlen + 0x33
strdup + 0x19
Desktop::Desktop(unsigned int, char const*) + 0x55
AppServer::_CreateDesktop(unsigned int, char const*) + 0x48
AppServer::_DispatchMessage(long, BPrivate::LinkReceiver&) + 0x170
MessageLooper::_MessageLooper() + 0x26
MessageLooper::_message_thread(void*) + 0xc
AppServer::RunLooper() + 0x2e
main + 0x59
_start + 0x4e
runtime_loader + 0xde
Corresponding code snippet:
407 Desktop::Desktop(uid_t userID, const char* targetScreen)
408 :
409 MessageLooper("desktop"),
410
411 fUserID(userID),
412 fTargetScreen(strdup(targetScreen)),
413 fSettings(NULL),
If you check our implementation of strdup() you see the following:
17 // unlike the standard strdup() function, the BeOS
implementation
18 // handles NULL strings gracefully
19 if (string == NULL)
20 return NULL;
21
22 length = strlen(string) + 1;
in other words we should be on the safe side providing any valid string pointer
without checking it for NULL, right? ;-) But not for Haiku GCC4 4.6.3 with
-ftree-vrp otion.
The situation is looking like this check is removed for some non-obvious
reasons. I suspect, that because NULL-input case is not specified in standard
for strdup the compiler make some assumptions that strdup never be called with
NULL parameter so this check is useless. Are there other ideas?
So, just for your information, Colleagues. ;-)
---
Kind Regards,
S.Zharski
Other related posts:
