Would hard-linking be any better? Will need to recursively hard-link all the
files in a specific directory, but may be faster and less IO-intensive than
copying or tarring/untarring all the files?

--
Denys


On Mon, Nov 25, 2019 at 11:24:12AM -0800, Khem Raj wrote:
> Ideally I agree it’s best to have per component git repos
> But traditional SDKs have been monolithic
>
> On Mon, Nov 25, 2019 at 10:50 AM Jacob Stiffler <j-stiffler@ti.com> wrote:
>
> >
> > On 11/25/2019 1:16 PM, Khem Raj wrote:
> >
> > On Mon, Nov 25, 2019 at 10:12 AM Jacob Stiffler <j-stiffler@ti.com> <j-stiffler@ti.com> wrote:
> >
> > On 11/25/2019 12:42 PM, Denys Dmytriyenko wrote:
> >
> > On Fri, Nov 15, 2019 at 09:14:15AM -0800, Khem Raj wrote:
> >
> > On Fri, 2019-11-15 at 10:14 -0500, Jacob Stiffler wrote:
> >
> > Recently individual components and LLD sources have been combined
> > into a single PDK repo. Use this class to specify the common source.
> > Also use this class to keep the sources separate from each other when
> > building. This keeps the build identical to previous recipes while
> > keeping control on interdependencies.
> >
> > Signed-off-by: Jacob Stiffler <j-stiffler@ti.com> <j-stiffler@ti.com>
> > ---
> >
> > ...
> >
> > +# Extract only required sources from PDK
> > +python do_unpack_append() {
> > +    if len(d.getVar('TI_PDK_COMP') or '') > 0:
> > +        import subprocess
> > +
> > +        src =
> > os.path.join(d.getVar('TI_PDK_SRC'),'packages',d.getVar('TI_PDK_COMP_
> > PATH'))
> > +
> > +        # Package up only the required sources
> > +        cmd = 'tar -cf - -C %s -p -S . | tar -xf - -C %s' % (src,
> > d.getVar('S'))
> > +        subprocess.check_output(cmd, shell=True,
> > stderr=subprocess.STDOUT)
> > +
> > +        bb.utils.remove(d.getVar('TI_PDK_SRC'), recurse=True)
> > +}
> >
> > perhaps looking in work-shared idea from kernel or gcc or clang
> > packages could make it better where, you only untar sources once
> >
> > Jake,
> >
> > Any updates on this - will you be looking into the above suggestion?
> >
> > I had originally considered this approach, but decided against it so
> > that there would be no way a recipe could access other sources in the
> > pdk.git.
> >
> > It may be a bit paranoid, but the work-shared approach would make the
> > entire pdk sources available through a ti-pdk-fetch variable. I wanted
> > it so that the entire pdk.git will never be unpacked in its entirety.
> >
> > disk operations are not cheap, so you could save a lot by
> > having a monorepo. Is it due to access perms ? then its a single
> > tarballs you cant have ACLs. I am still not convinced if monorepo
> > approach has downsides, plus its a known approach elsewhere so
> > will be easy on users.
> >
> > We wanted to have this safety net to make sure that this pdk.git does not
> > end up as spaghetti code. So this method was devised to ensure that only a
> > single component's sources and its DEPENDS are available during
> > compilation.
> >
> > It may be acceptable to clone to the work-shared path, and then symlink
> > the specific directory into the WORKDIR. But we will need to remain
> > vigilant so that work-shared path is never referenced in the recipe.
> >
> >
> > -=-=-=-=-=-=-=-=-=-=-=-
> > Links: You receive all messages sent to this group.
> >
> > View/Reply Online (#12528): https://lists.yoctoproject.org/g/meta-ti/message/12528
> > Mute This Topic: https://lists.yoctoproject.org/mt/61321552/3619770
> > Group Owner: meta-ti+owner@lists.yoctoproject.org
> > Unsubscribe: https://lists.yoctoproject.org/g/meta-ti/unsub  [j-stiffler@ti.com]
> > -=-=-=-=-=-=-=-=-=-=-=-
> >
> >
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.

View/Reply Online (#12531): https://lists.yoctoproject.org/g/meta-ti/message/12531
Mute This Topic: https://lists.yoctoproject.org/mt/61945973/1997914
Group Owner: meta-ti+owner@lists.yoctoproject.org
Unsubscribe: https://lists.yoctoproject.org/g/meta-ti/unsub  [raj.khem@gmail.com]
-=-=-=-=-=-=-=-=-=-=-=-