diff mbox

Package overriding

Message ID CAE21AQraBRp3_94Td_Gw+D9NML+-X0e+2U=cwR979ww99_aJ=w@mail.gmail.com
State Rejected
Headers show

Commit Message

Charles Manning March 18, 2013, 2:33 a.m. UTC
Hello Buildrooters

It seems to me that the package override feature is misnamed and has
some shortcomings.

First off, the name is, IMHO, poor. Package Override is applied before
tha packages are parsed. It does not override them. Instead it is
really overriding configurations before the packages are read. Perhaps
it should really be called CONFIG_OVERRIDE?

There is also a deficiency in that Package Override cannot override
anything set up in the packages.

For example, I recently needed to add two more configuration lines to
building bluez_utils.

Instead of:

BLUEZ_UTILS_CONF_OPT = --enable-test --enable-tools
I wanted
BLUEZ_UTILS_CONF_OPT = --enable-test --enable-tools --enable-bccmd --enable-dund


A logical way I can see to do this would be to have a mechanism
similar to Package Override but that is applied after the packages are
parsed.

Does that sound useful?

Is there a better way?

Thanks

Charles

Comments

Thomas Petazzoni March 18, 2013, 9:08 a.m. UTC | #1
Dear Charles Manning,

On Mon, 18 Mar 2013 15:33:27 +1300, Charles Manning wrote:

> It seems to me that the package override feature is misnamed and has
> some shortcomings.
> 
> First off, the name is, IMHO, poor. Package Override is applied before
> tha packages are parsed. It does not override them. Instead it is
> really overriding configurations before the packages are read. Perhaps
> it should really be called CONFIG_OVERRIDE?
> 
> There is also a deficiency in that Package Override cannot override
> anything set up in the packages.

Yes. The idea is that the "package override file" use special variables
when it needs to override something from a package recipe. For now, the
only special variable available is <pkg>_OVERRIDE_SRCDIR. It actually
does more than just overriding the source directory: it also changes
the behavior of the package infrastructure: instead of
download/extract/patch, it switches to just use the provided source
directory.

> For example, I recently needed to add two more configuration lines to
> building bluez_utils.
> 
> Instead of:
> 
> BLUEZ_UTILS_CONF_OPT = --enable-test --enable-tools
> I wanted
> BLUEZ_UTILS_CONF_OPT = --enable-test --enable-tools --enable-bccmd --enable-dund

Yes, the current override mechanism doesn't support this.

> A logical way I can see to do this would be to have a mechanism
> similar to Package Override but that is applied after the packages are
> parsed.
> 
> --- Makefile	(revision 20385)
> +++ Makefile	(working copy)

Using Subversion ? :-)

> @@ -319,11 +319,20 @@
>  -include $(PACKAGE_OVERRIDE_FILE)
>  endif
> 
>  include package/*/*.mk
>  include hj/package/*/*.mk
>  include boot/common.mk
>  include linux/linux.mk
> 
> +POST_PACKAGE_OVERRIDE_FILE=$(call qstrip,$(BR2_POST_PACKAGE_OVERRIDE_FILE))
> +ifneq ($(POST_PACKAGE_OVERRIDE_FILE),)
> +-include $(POST_PACKAGE_OVERRIDE_FILE)
> +endif
> +
> +
> 
> Then have POST_PACKAGE_OVERRIDE_FILE point to a file which has
> 
> # Add bluez_utils options we need
> BLUEZ_UTILS_CONF_OPT +=  --enable-bccmd --enable-dund
> 
> Does that sound useful?

Does this actually work?

The potential problem I see is that the package variables
(<pkg>_BLABLA) are defined in the package .mk file and then used when
the generic-package, autotools-package or cmake-package macros are
expanded. I'm a bit lost as to whether those <pkg>_BLABLA variables are
actually used when the package macros are expanded, or when the
functions defined by the package macros are executed. This would have
to be verified, or confirmed by someone who is more capable than me in
understanding how many dollar signs are needed to make this work :)

Also, will this work to override the source directory? We definitely
don't want to have two different mechanisms to override stuff.

Best regards,

Thomas
Charles Manning March 18, 2013, 8:28 p.m. UTC | #2
On Mon, Mar 18, 2013 at 10:08 PM, Thomas Petazzoni
<thomas.petazzoni@free-electrons.com> wrote:
> Dear Charles Manning,
>
> On Mon, 18 Mar 2013 15:33:27 +1300, Charles Manning wrote:
>
>> It seems to me that the package override feature is misnamed and has
>> some shortcomings.
>>
>> First off, the name is, IMHO, poor. Package Override is applied before
>> tha packages are parsed. It does not override them. Instead it is
>> really overriding configurations before the packages are read. Perhaps
>> it should really be called CONFIG_OVERRIDE?
>>
>> There is also a deficiency in that Package Override cannot override
>> anything set up in the packages.
>
> Yes. The idea is that the "package override file" use special variables
> when it needs to override something from a package recipe. For now, the
> only special variable available is <pkg>_OVERRIDE_SRCDIR. It actually
> does more than just overriding the source directory: it also changes
> the behavior of the package infrastructure: instead of
> download/extract/patch, it switches to just use the provided source
> directory.
>
>> For example, I recently needed to add two more configuration lines to
>> building bluez_utils.
>>
>> Instead of:
>>
>> BLUEZ_UTILS_CONF_OPT = --enable-test --enable-tools
>> I wanted
>> BLUEZ_UTILS_CONF_OPT = --enable-test --enable-tools --enable-bccmd --enable-dund
>
> Yes, the current override mechanism doesn't support this.
>
>> A logical way I can see to do this would be to have a mechanism
>> similar to Package Override but that is applied after the packages are
>> parsed.
>>
>> --- Makefile  (revision 20385)
>> +++ Makefile  (working copy)
>
> Using Subversion ? :-)
>
>> @@ -319,11 +319,20 @@
>>  -include $(PACKAGE_OVERRIDE_FILE)
>>  endif
>>
>>  include package/*/*.mk
>>  include boot/common.mk
>>  include linux/linux.mk
>>
>> +POST_PACKAGE_OVERRIDE_FILE=$(call qstrip,$(BR2_POST_PACKAGE_OVERRIDE_FILE))
>> +ifneq ($(POST_PACKAGE_OVERRIDE_FILE),)
>> +-include $(POST_PACKAGE_OVERRIDE_FILE)
>> +endif
>> +
>> +
>>
>> Then have POST_PACKAGE_OVERRIDE_FILE point to a file which has
>>
>> # Add bluez_utils options we need
>> BLUEZ_UTILS_CONF_OPT +=  --enable-bccmd --enable-dund
>>
>> Does that sound useful?
>
> Does this actually work?

Yes this actually works.

If you set this up in PACKAGE_OVERRIDE the " BLUEZ_UTILS_CONF_OPT =
..." line in bluez_utils wipes out the settings.


>
> The potential problem I see is that the package variables
> (<pkg>_BLABLA) are defined in the package .mk file and then used when
> the generic-package, autotools-package or cmake-package macros are
> expanded. I'm a bit lost as to whether those <pkg>_BLABLA variables are
> actually used when the package macros are expanded, or when the
> functions defined by the package macros are executed. This would have
> to be verified, or confirmed by someone who is more capable than me in
> understanding how many dollar signs are needed to make this work :)
>
> Also, will this work to override the source directory? We definitely
> don't want to have two different mechanisms to override stuff.

It really depends on what you're overriding.

If you're overriding config stuff (before package/* gets parsed) then
the current PACKAGE_OVERRIDE works file.

For instance, I sent the .config to  use HEAD as the revision for
various source fetches but then use a revison override file to set
them to the desired revisions.

Unfortunately it would seem that there is not just one place where you
can do all overriding, hence my proposal for adding another mechanism
**after** the packages to tweak them.

Even this will not do everything for all packages as make immediately
uses some values and defers the evaluation of others.

-- Charles
Arnout Vandecappelle March 22, 2013, 5:26 p.m. UTC | #3
On 18/03/13 03:33, Charles Manning wrote:
[snip]
> For example, I recently needed to add two more configuration lines to
> building bluez_utils.
>
> Instead of:
>
> BLUEZ_UTILS_CONF_OPT = --enable-test --enable-tools
> I wanted
> BLUEZ_UTILS_CONF_OPT = --enable-test --enable-tools --enable-bccmd --enable-dund
>
>
> A logical way I can see to do this would be to have a mechanism
> similar to Package Override but that is applied after the packages are
> parsed.
>
> --- Makefile	(revision 20385)
> +++ Makefile	(working copy)
> @@ -319,11 +319,20 @@
>   -include $(PACKAGE_OVERRIDE_FILE)
>   endif
>
>   include package/*/*.mk
>   include hj/package/*/*.mk
>   include boot/common.mk
>   include linux/linux.mk
>
> +POST_PACKAGE_OVERRIDE_FILE=$(call qstrip,$(BR2_POST_PACKAGE_OVERRIDE_FILE))
> +ifneq ($(POST_PACKAGE_OVERRIDE_FILE),)
> +-include $(POST_PACKAGE_OVERRIDE_FILE)
> +endif
> +
> +
>
> Then have POST_PACKAGE_OVERRIDE_FILE point to a file which has
>
> # Add bluez_utils options we need
> BLUEZ_UTILS_CONF_OPT +=  --enable-bccmd --enable-dund
>
> Does that sound useful?

  The problem with this approach is that it is not reliable. It _happens_ 
to work fine for _CONF_OPT. But it would not work for _DEPENDENCIES, for 
example, because the rule that consumes that variable comes before it. 
And probably there are some variables where it doesn't work at all 
because of some weird double-dollar stuff.

  That's why we just don't want to deal with this kind of changes. The 
override file is meant to be a development tool that allows you to hack 
the source code for a package. For other stuff, you should modify the 
package's .mk file directly.


> Is there a better way?

  In this particular case, the better way is to create new Config.in 
options for bccmd and dund. And send us the patch, of course.


  Regards,
  Arnout
diff mbox

Patch

--- Makefile	(revision 20385)
+++ Makefile	(working copy)
@@ -319,11 +319,20 @@ 
 -include $(PACKAGE_OVERRIDE_FILE)
 endif

 include package/*/*.mk
 include hj/package/*/*.mk
 include boot/common.mk
 include linux/linux.mk

+POST_PACKAGE_OVERRIDE_FILE=$(call qstrip,$(BR2_POST_PACKAGE_OVERRIDE_FILE))
+ifneq ($(POST_PACKAGE_OVERRIDE_FILE),)
+-include $(POST_PACKAGE_OVERRIDE_FILE)
+endif
+
+

Then have POST_PACKAGE_OVERRIDE_FILE point to a file which has

# Add bluez_utils options we need
BLUEZ_UTILS_CONF_OPT +=  --enable-bccmd --enable-dund