Shouldn't packages for all PACKAGE_CLASSES be generated?

Shouldn't packages for all PACKAGE_CLASSES be generated?

[Michael Opdenacker]

Greetings,

I'm surprised by this behavior, which I didn't expect... I'm running 
"bitbake core-image-cmdline-full; bitbake generate-index" from Poky master.

with PACKAGE_CLASSES = "package_ipk package_rpm package_deb"...

find tmp/deploy/deb -name *.deb | wc; find tmp/deploy/rpm -name *.rpm | 
wc; find tmp/deploy/ipk -name *.ipk | wc
    1918    1918  129927
    1918    1918  135794
    5404    5404  393252

with PACKAGE_CLASSES = "package_rpm package_deb package_ipk"...

find tmp/deploy/deb -name *.deb | wc; find tmp/deploy/rpm -name *.rpm | 
wc; find tmp/deploy/ipk -name *.ipk | wc
    1918    1918  129927
    5404    5404  393606
    1918    1918  135584

with PACKAGE_CLASSES = "package_deb package_ipk package_rpm"...

find tmp/deploy/deb -name *.deb | wc; find tmp/deploy/rpm -name *.rpm | 
wc; find tmp/deploy/ipk -name *.ipk | wc
    5404    5404  376187
    1918    1918  135794
    1918    1918  135584

So, more packages are generated to the first item in PACKAGE_CLASSES.

It's expected to have the first item in PACKAGE_CLASSES that is used to 
generate the image, but I expected all packages to be generated for all 
package types anyway. Isn't this the expected behavior? Otherwise, how 
would you generate 3 types of feeds (rpm, deb and ipk) at the same time?

Cheers
Michaell

-- 
Michael Opdenacker, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

Re: [OE-core] Shouldn't packages for all PACKAGE_CLASSES be generated?

[Richard Purdie]

On Thu, 2024-02-01 at 13:47 +0100, Michael Opdenacker via
lists.openembedded.org wrote:
> Greetings,
> 
> I'm surprised by this behavior, which I didn't expect... I'm running 
> "bitbake core-image-cmdline-full; bitbake generate-index" from Poky master.
> 
> with PACKAGE_CLASSES = "package_ipk package_rpm package_deb"...
> 
> find tmp/deploy/deb -name *.deb | wc; find tmp/deploy/rpm -name *.rpm | 
> wc; find tmp/deploy/ipk -name *.ipk | wc
>     1918    1918  129927
>     1918    1918  135794
>     5404    5404  393252
> 
> with PACKAGE_CLASSES = "package_rpm package_deb package_ipk"...
> 
> find tmp/deploy/deb -name *.deb | wc; find tmp/deploy/rpm -name *.rpm | 
> wc; find tmp/deploy/ipk -name *.ipk | wc
>     1918    1918  129927
>     5404    5404  393606
>     1918    1918  135584
> 
> with PACKAGE_CLASSES = "package_deb package_ipk package_rpm"...
> 
> find tmp/deploy/deb -name *.deb | wc; find tmp/deploy/rpm -name *.rpm | 
> wc; find tmp/deploy/ipk -name *.ipk | wc
>     5404    5404  376187
>     1918    1918  135794
>     1918    1918  135584
> 
> So, more packages are generated to the first item in PACKAGE_CLASSES.
> 
> It's expected to have the first item in PACKAGE_CLASSES that is used to 
> generate the image, but I expected all packages to be generated for all 
> package types anyway. Isn't this the expected behavior? Otherwise, how 
> would you generate 3 types of feeds (rpm, deb and ipk) at the same time?

This is expected for various reasons, try adding --runall build to the
command.

Cheers,

Richard

Re: [OE-core] Shouldn't packages for all PACKAGE_CLASSES be generated?

[Michael Opdenacker]

Hi Richard,

Thanks for your answer!

On 2/2/24 at 16:57, Richard Purdie wrote:
> On Thu, 2024-02-01 at 13:47 +0100, Michael Opdenacker via
> lists.openembedded.org wrote:
>> Greetings,
>>
>> I'm surprised by this behavior, which I didn't expect... I'm running
>> "bitbake core-image-cmdline-full; bitbake generate-index" from Poky master.
>>
>> with PACKAGE_CLASSES = "package_ipk package_rpm package_deb"...
>>
>> find tmp/deploy/deb -name *.deb | wc; find tmp/deploy/rpm -name *.rpm |
>> wc; find tmp/deploy/ipk -name *.ipk | wc
>>      1918    1918  129927
>>      1918    1918  135794
>>      5404    5404  393252
>>
>> with PACKAGE_CLASSES = "package_rpm package_deb package_ipk"...
>>
>> find tmp/deploy/deb -name *.deb | wc; find tmp/deploy/rpm -name *.rpm |
>> wc; find tmp/deploy/ipk -name *.ipk | wc
>>      1918    1918  129927
>>      5404    5404  393606
>>      1918    1918  135584
>>
>> with PACKAGE_CLASSES = "package_deb package_ipk package_rpm"...
>>
>> find tmp/deploy/deb -name *.deb | wc; find tmp/deploy/rpm -name *.rpm |
>> wc; find tmp/deploy/ipk -name *.ipk | wc
>>      5404    5404  376187
>>      1918    1918  135794
>>      1918    1918  135584
>>
>> So, more packages are generated to the first item in PACKAGE_CLASSES.
>>
>> It's expected to have the first item in PACKAGE_CLASSES that is used to
>> generate the image, but I expected all packages to be generated for all
>> package types anyway. Isn't this the expected behavior? Otherwise, how
>> would you generate 3 types of feeds (rpm, deb and ipk) at the same time?
> This is expected for various reasons, try adding --runall build to the
> command.


Right, it works and I'm now getting the same number of packages for all 
variants.

By the way, another surprise, ipk packages take more space than deb 
ones, at least in the feeds:

$ du -sh tmp/deploy/*
1.4G    tmp/deploy/deb
2.2G    tmp/deploy/ipk
1.6G    tmp/deploy/rpm

What makes ipk more attractive than deb for embedded cases? I can just 
find the statement that opkg is lightweight and optimized for embedded 
applications (https://git.yoctoproject.org/opkg/tree/README.md).

Thanks again
Michael.

-- 
Michael Opdenacker, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com