Re: [Qemu-devel] [PATCH 2/6] Move paio_cancel() to new infrastructure.

[Stefan Hajnoczi]

On Thu, Nov 18, 2010 at 6:06 PM, Arun R Bharadwaj
<address@hidden> wrote:
> Move paio_cancel() to new infrastructure and introduce
> the necessary APIs for this.
>
> Signed-off-by: Arun R Bharadwaj <address@hidden>
> ---
>  posix-aio-compat.c |   92 
> ++++++++++++++++++++++++++++++++++++++++++----------
>  1 files changed, 74 insertions(+), 18 deletions(-)

This commit is not a safe step to make.  paio_cancel() is going to
cause uninitialized threadlet structures to be accessed.

You split the patches up into smaller commits for easier review.
Unfortunately this particular commit will result in a QEMU which
builds successfully but has undefined behavior at runtime - it may
crash or do random things.

That's a problem for git-bisect(1) and in general for anyone who
assumes they can build QEMU at an arbitrary point in the commit
history.  It's really important to preserve bisectability, because
once commits get introduced that build bad QEMUs we can no longer have
confidence in bisect!

Please split up commits so they introduce a new feature incrementally
but work properly at each step in the series.

Stefan