|
From: Crispin F. <val...@fl...> - 2003-12-09 10:42:18
|
Hi, At the moment when valgrind hits a system call that it doesn't understand it prints out a message and quits. I was wondering if it make sense to change this so that it prints out a message and continues returning ENOSYS. This would be useful as I have a program that can make use of epoll() if it exists. When it starts up, it calls one of the epoll() functions and if that returns ENOSYS it continues on and reverts to poll(). Obviously at the moment valgrind just aborts when it encounters the epoll() system call. Crispin |
|
From: Dan K. <da...@ke...> - 2003-12-09 16:07:31
|
Crispin Flowerday wrote: > At the moment when valgrind hits a system call that it doesn't > understand it prints out a message and quits. I was wondering if it make > sense to change this so that it prints out a message and continues > returning ENOSYS. > > This would be useful as I have a program that can make use of epoll() if > it exists. When it starts up, it calls one of the epoll() functions and > if that returns ENOSYS it continues on and reverts to poll(). Obviously > at the moment valgrind just aborts when it encounters the epoll() system > call. Y'know, that's probably a good idea. My wrapper around epoll() will fall back to an older technique, too. (See http://kegel.com/rn/current/rn.c) It'd be nice if epoll_create() returned -1 on failure. - Dan |
|
From: Jeremy F. <je...@go...> - 2003-12-09 18:37:14
|
On Tue, 2003-12-09 at 08:05, Dan Kegel wrote: > Crispin Flowerday wrote: > > At the moment when valgrind hits a system call that it doesn't > > understand it prints out a message and quits. I was wondering if it make > > sense to change this so that it prints out a message and continues > > returning ENOSYS. > > > > This would be useful as I have a program that can make use of epoll() if > > it exists. When it starts up, it calls one of the epoll() functions and > > if that returns ENOSYS it continues on and reverts to poll(). Obviously > > at the moment valgrind just aborts when it encounters the epoll() system > > call. > > Y'know, that's probably a good idea. My wrapper around epoll() > will fall back to an older technique, too. > (See http://kegel.com/rn/current/rn.c) > It'd be nice if epoll_create() returned -1 on failure. CVS head does this now. It shouldn't be too hard to fix 2.0. J |