Patchwork [meta-webserver,1/1] webmin: drop allarch

login
register
mail settings
Submitter Anders Darander
Date June 24, 2014, 7:54 a.m.
Message ID <1b5c27f39444bed435c8c050b2fb6def74a5adb8.1403596416.git.anders@chargestorm.se>
Download mbox | patch
Permalink /patch/74287/
State New, archived
Headers show

Comments

Anders Darander - June 24, 2014, 7:54 a.m.
* runtime dependencies are TUNE_PKGARCH causing do_package_write_*
  task to have different signature for MACHINEs with different
  TUNE_PKGARCH

Signed-off-by: Anders Darander <anders@chargestorm.se>
---
 meta-webserver/recipes-webadmin/webmin/webmin_1.620.bb | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
Paul Eggleton - June 25, 2014, 7:45 a.m.
On Tuesday 24 June 2014 09:54:12 Anders Darander wrote:
> * runtime dependencies are TUNE_PKGARCH causing do_package_write_*
>   task to have different signature for MACHINEs with different
>   TUNE_PKGARCH
> 
> Signed-off-by: Anders Darander <anders@chargestorm.se>
> ---
>  meta-webserver/recipes-webadmin/webmin/webmin_1.620.bb | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/meta-webserver/recipes-webadmin/webmin/webmin_1.620.bb
> b/meta-webserver/recipes-webadmin/webmin/webmin_1.620.bb index
> 7eb1d63..0f8781f 100644
> --- a/meta-webserver/recipes-webadmin/webmin/webmin_1.620.bb
> +++ b/meta-webserver/recipes-webadmin/webmin/webmin_1.620.bb
> @@ -20,7 +20,7 @@ SRC_URI =
> "${SOURCEFORGE_MIRROR}/webadmin/webmin-${PV}.tar.gz \ SRC_URI[md5sum] =
> "e93bc5057a5b7d7e31ecb2bb228d7044"
>  SRC_URI[sha256sum] =
> "3d36153406d8e5d3dcaeadba34dfb5cdbc4060b75c38339174ac97b2277f284b"
> 
> -inherit allarch perlnative update-rc.d
> +inherit perlnative update-rc.d
> 
>  do_configure() {
>      # Remove binaries and plugins for other platforms

So apart from me being reticent to make this change at all (for reasons that 
we've discussed before), there is another issue - allarch.bbclass does more 
than just setting PACKAGE_ARCH, it also disables certain other functionality 
such as debug symbol splitting which is superfluous for these kinds of recipes; 
in my testing when I originally wrote the webmin recipe this significantly 
reduced the time required to build the recipe. We can't just drop this inherit 
without regard to that. (I'm sure the same applies elsewhere, it's just that 
the size of webmin makes the impact significant.)

Cheers,
Paul
Anders Darander - June 25, 2014, 9:28 a.m.
* Paul Eggleton <paul.eggleton@linux.intel.com> [140625 09:46]:

> On Tuesday 24 June 2014 09:54:12 Anders Darander wrote:
> > * runtime dependencies are TUNE_PKGARCH causing do_package_write_*
> >   task to have different signature for MACHINEs with different
> >   TUNE_PKGARCH

> > -inherit allarch perlnative update-rc.d
> > +inherit perlnative update-rc.d

> So apart from me being reticent to make this change at all (for reasons that 
> we've discussed before), 

Yes, I'm well aware of those discussions. I was actually thinking about
CC:ing you on my reply to Martin, after he suggested something like this
change.

> there is another issue - allarch.bbclass does more than just setting
> PACKAGE_ARCH, it also disables certain other functionality such as
> debug symbol splitting which is superfluous for these kinds of
> recipes; in my testing when I originally wrote the webmin recipe this
> significantly reduced the time required to build the recipe. We can't
> just drop this inherit without regard to that. (I'm sure the same
> applies elsewhere, it's just that the size of webmin makes the impact
> significant.)

I'm all for saving time! (And I'm not personally concerned by this
issue, as I'm currently only using webmin in a build tree for a signle
machine).

Though, to get back to Martin Jansa's issue; if there isn't already an
enhancment request file to the Yocto bugzilla, maybe it's time to file
one regarding the possibillity of having an allarch package
RDEPENDS/RRECOMMNEND a package with an ARCH or MACHINE tune...

Paul, what do you think about the feasibility of finding such a
solution?

Cheers,
Anders
Martin Jansa - June 25, 2014, 9:45 a.m.
On Wed, Jun 25, 2014 at 11:28:07AM +0200, Anders Darander wrote:
> * Paul Eggleton <paul.eggleton@linux.intel.com> [140625 09:46]:
> 
> > On Tuesday 24 June 2014 09:54:12 Anders Darander wrote:
> > > * runtime dependencies are TUNE_PKGARCH causing do_package_write_*
> > >   task to have different signature for MACHINEs with different
> > >   TUNE_PKGARCH
> 
> > > -inherit allarch perlnative update-rc.d
> > > +inherit perlnative update-rc.d
> 
> > So apart from me being reticent to make this change at all (for reasons that 
> > we've discussed before), 
> 
> Yes, I'm well aware of those discussions. I was actually thinking about
> CC:ing you on my reply to Martin, after he suggested something like this
> change.
> 
> > there is another issue - allarch.bbclass does more than just setting
> > PACKAGE_ARCH, it also disables certain other functionality such as
> > debug symbol splitting which is superfluous for these kinds of
> > recipes; in my testing when I originally wrote the webmin recipe this
> > significantly reduced the time required to build the recipe. We can't
> > just drop this inherit without regard to that. (I'm sure the same
> > applies elsewhere, it's just that the size of webmin makes the impact
> > significant.)
> 
> I'm all for saving time! (And I'm not personally concerned by this
> issue, as I'm currently only using webmin in a build tree for a signle
> machine).

you can probably use allarch and assign
PACKAGE_ARCH = "${TUNE_PKGARCH}"
I don't remember how many savings will be disabled by this combination.

For multimachine builds these savings could be worse, because better to
spend a bit more time building everything just once per TUNE_PKGARCH and
don't touch it again (until the signature is changed) than keeping it
allarch, but still building it once per TUNE_PKGARCH (because of different
signature and then unpacking it from sstate *every-single-time* you're
building image for different MACHINE (so even without any change to
webmin recipe).

> Though, to get back to Martin Jansa's issue; if there isn't already an
> enhancment request file to the Yocto bugzilla, maybe it's time to file
> one regarding the possibillity of having an allarch package
> RDEPENDS/RRECOMMNEND a package with an ARCH or MACHINE tune...
 
There is one already:
https://bugzilla.yoctoproject.org/show_bug.cgi?id=5970

Patch

diff --git a/meta-webserver/recipes-webadmin/webmin/webmin_1.620.bb b/meta-webserver/recipes-webadmin/webmin/webmin_1.620.bb
index 7eb1d63..0f8781f 100644
--- a/meta-webserver/recipes-webadmin/webmin/webmin_1.620.bb
+++ b/meta-webserver/recipes-webadmin/webmin/webmin_1.620.bb
@@ -20,7 +20,7 @@  SRC_URI = "${SOURCEFORGE_MIRROR}/webadmin/webmin-${PV}.tar.gz \
 SRC_URI[md5sum] = "e93bc5057a5b7d7e31ecb2bb228d7044"
 SRC_URI[sha256sum] = "3d36153406d8e5d3dcaeadba34dfb5cdbc4060b75c38339174ac97b2277f284b"
 
-inherit allarch perlnative update-rc.d
+inherit perlnative update-rc.d
 
 do_configure() {
     # Remove binaries and plugins for other platforms