Re: [Qemu-devel] [PATCH v6 1/2] block/vxhs.c: Add support for a new block device type called "vxhs"

From: Eric Blake <eblake@redhat.com>
To: Stefan Hajnoczi <stefanha@gmail.com>, Fam Zheng <famz@redhat.com>
Cc: kwolf@redhat.com, Venkatesha.Mg@veritas.com,
	ashish.mittal@veritas.com, jcody@redhat.com,
	qemu-devel@nongnu.org, Rakesh.Ranjan@veritas.com,
	armbru@redhat.com, Ketan.Nilangekar@veritas.com,
	Abhijit.Dey@veritas.com, pbonzini@redhat.com,
	Buddhi.Madhav@veritas.com, Ashish Mittal <ashmit602@gmail.com>
Subject: Re: [Qemu-devel] [PATCH v6 1/2] block/vxhs.c: Add support for a new block device type called "vxhs"
Date: Tue, 15 Nov 2016 09:52:20 -0600	[thread overview]
Message-ID: <bee01cd5-b560-6113-052f-fc96cc9e60d3@redhat.com> (raw)
In-Reply-To: <20161115101850.GA4836@stefanha-x1.localdomain>

[-- Attachment #1: Type: text/plain, Size: 2175 bytes --]

On 11/15/2016 04:18 AM, Stefan Hajnoczi wrote:
> On Tue, Nov 15, 2016 at 10:04:05AM +0800, Fam Zheng wrote:
>> On Mon, 11/14 13:06, Eric Blake wrote:
>>> So I guess you have to determine if libqnio is something that should
>>> compile completely independent from qemu, or whether it is so closely
>>> tied to the rest of qemu that it should follow qemu conventions.
>>
>> The question is on include directives in block/vxhs.c, not libnqio library
>> header, so qemu conventions apply.
> 
> Eric: The libqnio library header is not copied into the QEMU source
> tree.  It is an external library dependency like libnfs or libglfs.
> 
> Fam, Markus: Unfortunately neither the clean-includes script nor its
> patch series cover letter explains *why* osdep.h should be included
> before system headers.

For the same reason that <config.h> should be included first before
system headers in an autoconf'd project. Many platforms ship
(semi-)broken system headers, where configure detects AND works around
the problems, but only if the workaround CONSISTENTLY happens before any
system header is included.  It is a LOT harder to track down compilation
or link or even subtle runtime failures if you don't consistently use
your workaround all the time, such that some .c files got the workaround
but others did not.

> The libqnio header is self-contained (i.e. you can #include it and it
> has no dependencies) and only used by vxhs.c.  Why is it a good idea to
> include qemu/osdep.h first?
> 
> Seems like a bad idea to me because it masks missing dependencies in the
> libqnio header.

If the libqnio header is completely independent, then it should not be
part of the qemu source tree (with "" inclusion), but should instead be
externally included (with <> inclusion), just like any other third-party
library we link against.  But even then, we STILL want our osdep.h to
occur before ANY third-party library, so that we have a CONSISTENT
environment, which is why we mandate that osdep.h be included first in
all qemu .c files.

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 604 bytes --]