write(2)/writev(2) atomicity?
write(2)/writev(2) atomicity?
- Subject: write(2)/writev(2) atomicity?
- From: email@hidden
- Date: Tue, 8 Nov 2005 14:11:05 +0100
Hi All,
in a multi-threaded terminal application I use writev(2) to write status
messages to stderr (for testing purposes an xterm), however although the
complete message is written with a single call to writev (and now I also
tested write, with the same effect), and in the tested case the length of
the message was well below PIPE_BUF, the output of the different threads
is sometimes interleaved (always starting at 40 bytes into a message,
i.e. 40 bytes of the first message, and then another message, then the
rest of the first message).
I am pretty sure that all standards require at least write(2) to be
async-signal-safe, and to write messages smaller than PIPE_BUF
atomically ... does anyone know more about this on Darwin?
(Including pointers to more documentation about thread safety of the
kernel and libc functions; only very few like getaddrinfo(3) seem to
have notes on this subject (and in that case stating only that it is
_not_ thread safe, contrary to POSIX) in their man-page.)
Thanks, Colin
PS: 10.4.3 on a Mac mini.
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden