Re: [Qemu-devel] [PATCH v3 03/10] tests: Add vm test lib

[Philippe Mathieu-Daudé]

On 08/22/2017 11:00 AM, Fam Zheng wrote:
> On Tue, 08/22 08:19, Philippe Mathieu-Daudé wrote:
> > > +    def add_source_dir(self, data_dir):
> > > +        name = "data-" + hashlib.sha1(data_dir).hexdigest()[:5]
> > > +        tarfile = os.path.join(self._tmpdir, name + ".tar")
> > > +        logging.debug("Creating archive %s for data dir: %s", tarfile, 
> > > data_dir)
> > > +        subprocess.check_call(["tar", "--exclude-vcs",
> >
> > --exclude-vcs-ignores will process .gitignore,
>
> Sounds useful, thanks!
>
> > you can also list with 'git ls-files',
> > or generate tarball with scripts/make-release
>
> They won't handle submodules, will they?

"git ls-files --recurse-submodules"

This way you can avoid the --exclude-vcs-ignores which is too recent for 
NetBSD as remarked Kamil, and avoid the tar --exclude=

> > > +                               "--exclude=tests/vm/*.img",
> > > +                               "--exclude=tests/vm/*.img.*",
> > > +                               "--exclude=*.d",
> > > +                               "--exclude=*.o",
> > > +                               "--exclude=docker-src.*",
> > > +                               "-cf", tarfile, '.'], cwd=data_dir,
> > > +                              stdin=self._devnull, stdout=self._stdout)
> > > +        self._data_args += ["-drive",
> > > +                            "file=%s,if=none,id=%s,cache=writeback,format=raw" 
> > > % \
> > > +                                    (tarfile, name),
> > > +                            "-device",
> > > +                            "virtio-blk,drive=%s,serial=%s,bootindex=1" % 
> > > (name, name)]
> >
> > what about mounting readonly via '-device virtio-9p-pci' (not using
> > tarball)?
>
> It doesn't work. That way an out-of-tree build is required in the guest, but if
> an in-tree build has been done in host side, out-of-tree builds fail.  Also the
> performance will be much worse, I believe: untar is sequential read, and after
> that everything is in the guest page cache; with virtio-9p, there will be a
> large amount of "random" requests between guest and host.

Indeed you are correct, I didn't think of it, thank to remember me that :)