On 01/16/2013 03:13 AM, Daniel P. Berrange wrote:
On Wed, Jan 16, 2013 at 10:53:05AM +0800, Hu Tao wrote:
> required by virSetError.
> ---
> src/util/virthread.h | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/src/util/virthread.h b/src/util/virthread.h
> index b11a251..c209440 100644
> --- a/src/util/virthread.h
> +++ b/src/util/virthread.h
> @@ -23,6 +23,7 @@
> # define __THREADS_H_
>
> # include "internal.h"
> +# include "virerror.h"
>
> typedef struct virMutex virMutex;
> typedef virMutex *virMutexPtr;
AFAICT, the header file doesn't need this addition - the .c file
needs it.
I disagree. The definition of the VIR_ONCE_GLOBAL_INIT macro uses
virErrorPtr, which means if we don't include virerror.h here, _every_ .c
file that uses one-shot initialization would then have to pull in
virerror.h. It turns out that all existing clients already do, or we
would have hit a compilation error sooner; thus this patch is really
only mandatory for the new client later in this series - but that commit
has already been nack'd on design grounds. Still, requiring a client .c
file to do extra includes just to use the macro makes it harder than
necessary, so I'm still in favor of making virthread.h self-contained.
ACK, after updating the commit message to explain why. I've pushed 1-3
in this series. Like Dan, I agree that the remainder of the series
needs a step back to figure out what the real design goal is, before
doing any more refactoring.
--
Eric Blake eblake redhat com +1-919-301-3266
Libvirt virtualization library
http://libvirt.org