On Tue, 2024-04-16 at 09:48 +0800, Changqing Li wrote:
> On 4/15/24 18:05, Richard Purdie wrote:
> > CAUTION: This email comes from a non Wind River email account!
> > Do not click links or open attachments unless you recognize the
> > sender and know the content is safe.
> > 
> > On Mon, 2024-04-15 at 16:30 +0800, Changqing Li via
> > lists.openembedded.org wrote:
> > > From: Changqing Li <[email protected]>
> > > 
> > > When there are tarballs in local mirror file://, do_fetch will
> > > create symlink under DL_DIR, eg:
> > > pseudo-prebuilt-2.33.tar.xz ->
> > > /project/path/to/localmirrorA/pseudo-prebuilt-2.33.tar.xz
> > > after user update path from localmirrorA to localmirrorB, because
> > > do_fetch don't depends on PREMIRRORS/MIRRORS, do_fetch will not
> > > be rerun, so the symlink will not update to:
> > > pseudo-prebuilt-2.33.tar.xz ->
> > > /project/path/to/localmirrorB/pseudo-prebuilt-2.33.tar.xz
> > > then, do_upack will failed since the tarball cannot be found.
> > > 
> > > Signed-off-by: Changqing Li <[email protected]>
> > > ---
> > >   meta/classes-global/base.bbclass | 1 +
> > >   1 file changed, 1 insertion(+)
> > > 
> > > diff --git a/meta/classes-global/base.bbclass b/meta/classes-
> > > global/base.bbclass
> > > index 0999b42daa..5094a1540e 100644
> > > --- a/meta/classes-global/base.bbclass
> > > +++ b/meta/classes-global/base.bbclass
> > > @@ -139,6 +139,7 @@ do_fetch[file-checksums] =
> > > "${@bb.fetch.get_checksum_file_list(d)}"
> > >   do_fetch[file-checksums] += "
> > > ${@get_lic_checksum_file_list(d)}"
> > >   do_fetch[prefuncs] += "fetcher_hashes_dummyfunc"
> > >   do_fetch[network] = "1"
> > > +do_fetch[vardeps] += "PREMIRRORS MIRRORS"
> > >   python base_do_fetch() {
> > > 
> > >       src_uri = (d.getVar('SRC_URI') or "").split()
> > > 
> > Changing PREMIRRORS or MIRRORS should not change the output of the
> > task, ever, so do_fetch shouldn't depend upon them.
> 
> The failure case I met is like this:
> 
> I have a project with PREMIRRORS setting, and the tarballs are under
> it.
> 
> PREMIRRORS:append = "file:///project/path/to/localmirrorA"
> 
> after first bitbake pseudo-native, there will a symlink under DL_DIR:
> 
> pseudo-prebuilt-2.33.tar.xz -> 
> /project/path/to/localmirrorA/pseudo-prebuilt-2.33.tar.xz
> 
> Then, Since all the tarballs are moved to
> /project/path/to/localmirrorB
> 
> I changed PREMIRRORS:append = "file:///project/path/to/localmirrorB"
> 
> Then bitbake pseudo-native again,   do_fetch  will not rerun since no
> signature change.
> 
> do_unpack rerun since one dependency  change, it will try to unpack 
> ${DL_DIR}/pseudo-prebuilt-2.33.tar.xz,
> 
> but since do_fetch not rerun, ${DL_DIR}/pseudo-prebuilt-2.33.tar.xz 
> still point to localmirrorA, but localmirrorA not exist
> 
> since localmirrorB replace it.  Then will cause do_unpck failure:
> 
> NOTE: Unpacking /build/downloads/pseudo-prebuilt-2.33.tar.xz to 
> /build/tmp-glibc/work/x86_64-linux/pseudo-native/1.9.0+git xz: 
> /build/downloads/pseudo-prebuilt-2.33.tar.xz: No such file or
> directory
> 
> So this is why I want to add this depends,  and want to make do_fetch
> rerun and resymlink to /project/path/to/localmirrorB

The question is if the fetch reruns, does it correct the path?

do_fetch should definitely not depend on PREMIRRORS or MIRRORS as those
can be set by local users for local setups. The idea is that result
would be the same regardless, else the fetch checksums (or git
revision) would not match. I will therefore not take any patch changing
that.

Also, despite what Alex said, the fetcher does only make symlinks for
local files, it does not copy every file into DL_DIR. It does this for
efficiency, both time and disk usage.

If you start moving files around locally, you're expected to fix up the
links too. I believe that if you manually run the fetch task, it will
update correctly so the only issue is it not automatically triggering
but I don't believe it should trigger.

So whilst I can see how you have an unfortunate build failure, it was
of your own making by moving mirror directories around.

Cheers,

Richard



-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#198461): 
https://lists.openembedded.org/g/openembedded-core/message/198461
Mute This Topic: https://lists.openembedded.org/mt/105532159/21656
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub 
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to