On Sat, May 14, 2011 at 06:56:19PM +0200, Peter Mazinger wrote:
> Hi,
> 
> I see a problem in the fallback (fake) code, I decided there to use 
> __X_nocancel for lseek*, read, write...
> The comment below describes this situation in 6.a., but for master branch I 
> think it would be better to use X instead, haven't checked, if all needed 
> function's nocancel variants are implemented...

Why does this fallback code even still exist? It's hopelessly
incorrect. The whole point of pread/pwrite is atomicity with respect
to multiple threads or processes working with the same open file
description. Emulating it with a hack that can't provide atomicity is
worse than not supporting pread/pwrite at all since it will introduce
race conditions in correct code that's expecting them to be atomic if
they exist. For the same reason, pselect should not be emulated.

I vote for removing the emulation completely.

Rich
_______________________________________________
uClibc mailing list
[email protected]
http://lists.busybox.net/mailman/listinfo/uclibc

Reply via email to