[libvirt] [PATCH] Fix dependancies for remote generated files

[Eric Blake]

On 08/24/2010 07:21 AM, Daniel P. Berrange wrote:
> Very occasionally during a parallel make, dispatch.c would
> be compiled before the generated remote headers had been
> fully written. This would cause it to compile an empty
> union, and result in really wierd runtime bugs that are

s/wierd/weird/

> near impossible to diagnose.
>
> * daemon/Makefile.am: Fix remote build deps
> ---
>   daemon/Makefile.am |   19 +++++++++++--------
>   1 files changed, 11 insertions(+), 8 deletions(-)
>
> diff --git a/daemon/Makefile.am b/daemon/Makefile.am
> index dbf0ac3..035c149 100644
> --- a/daemon/Makefile.am
> +++ b/daemon/Makefile.am
> @@ -189,14 +189,17 @@ uninstall-data-polkit::
>   endif
>
>
> -remote.c: remote_dispatch_prototypes.h \
> -	  remote_dispatch_table.h \
> -	  remote_dispatch_args.h \
> -	  remote_dispatch_ret.h \
> -	  qemu_dispatch_prototypes.h \
> -	  qemu_dispatch_table.h \
> -	  qemu_dispatch_args.h \
> -	  qemu_dispatch_ret.h
> +remote.c: \
> +	remote_dispatch_prototypes.h \
> +	remote_dispatch_table.h \
> +	qemu_dispatch_prototypes.h \
> +	qemu_dispatch_table.h
> +
> +remote.h: \
> +	remote_dispatch_args.h \
> +	remote_dispatch_ret.h \
> +	qemu_dispatch_args.h \
> +	qemu_dispatch_ret.h

I agree that the explicit remote.h dependencies are needed, since 
remote.h is in turn included by dispatch.c.

I wonder if you're still missing something.  Since both dispatch.c and 
remote.c do #include's remote.h, shouldn't they both also list remote.h 
(or remote.h's generated headers) as dependencies?  On the other hand, 
maybe you already get this for free from automake (which uses gcc's -M 
options to pick up automatic dependency tracking based on #include 
contents) - I do see remote.h listed in both 
daemon/.deps/libvirtd-{remote,dispatch}.Po.

On the other hand, where is the action that updates the timestamp of 
remote.[ch] to be newer than their generated .h counterparts?  Without 
that rule, then every run of make will see the generated files with a 
newer timestamp than the version-controlled files, and insist on 
regenerating the headers; whereas if you touched remote.[ch] after all 
the generated headers are up-to-date, then successive make runs will see 
that there is nothing left to do.

-- 
Eric Blake   eblake at redhat.com    +1-801-349-2682
Libvirt virtualization library http://libvirt.org