Patchwork base.bbclass: Ensure PREFERRED_PROVIDER and PREFERRED_VERSION values are set for multilibs

When running multilib builds its not intuitive for the specificed
versions/providers to be build in one case and the higher version in the
other. This patch mirrors the options unless they're already set.

There shouldn't be any effect if the MULTILIBS variable isn't set.

[YOCTO #1339]

Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
---

Richard Purdie wrote on 2011-08-05:
> When running multilib builds its not intuitive for the specificed
> versions/providers to be build in one case and the higher version in the other.
> This patch mirrors the options unless they're already set.
> 
> There shouldn't be any effect if the MULTILIBS variable isn't set.
> 
> [YOCTO #1339]
> 
> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org> ---
> diff --git a/meta/classes/base.bbclass b/meta/classes/base.bbclass index
> ff960cd..4aee847 100644 --- a/meta/classes/base.bbclass +++
> b/meta/classes/base.bbclass @@ -139,6 +139,46 @@ def pkgarch_mapping(d):
>          if d.getVar("TUNE_PKGARCH", True) == "armv7a-vfp-neon":
>              d.setVar("TUNE_PKGARCH", "armv7a")
> +def preferred_ml_updates(d):
> +    # If any PREFERRED_PROVIDER or PREFERRED_VERSIONS are set,
> +    # we need to mirror these variables in the multilib case
> +    multilibs = d.getVar('MULTILIBS', True) or ""
> +    if not multilibs:
> +        return
> +
> +    prefixes = []
> +    for ext in multilibs.split():
> +        eext = ext.split(':')
> +        if len(eext) > 1 and eext[0] == 'multilib':
> +            prefixes.append(eext[1])
> +
> +    versions = []
> +    providers = []
> +    for v in d.keys():
> +        if v.startswith("PREFERRED_VERSION_"):
> +            versions.append(v)
> +        if v.startswith("PREFERRED_PROVIDER_"):
> +            providers.append(v)
> +
> +    for v in versions:
> +        val = d.getVar(v, False)
> +        pkg = v.replace("PREFERRED_VERSION_", "")
> +        for p in prefixes:
> +            newname = "PREFERRED_VERSION_" + p + "-" + pkg
> +            if not d.getVar(newname, False):
> +                d.setVar(newname, val)
> +
> +    for prov in providers:
> +        val = d.getVar(prov, False)
> +        pkg = prov.replace("PREFERRED_PROVIDER_", "")
> +        virt = ""
> +        if pkg.startswith("virtual/"):
> +             pkg = pkg.replace("virtual/", "")
> +             virt = "virtual/"
> +        for p in prefixes:
> +            newname = "PREFERRED_PROVIDER_" + virt + p + "-" + pkg
> +            if not d.getVar(newname, False):
> +                d.setVar(newname, p + "-" + val)

How about the native and nativesdk packages? I think they should not be prefixed.

>  addhandler base_eventhandler
>  python base_eventhandler() {
> @@ -211,6 +251,7 @@ python base_eventhandler() {
>          if name == "ConfigParsed":
>                  generate_git_config(e)
>                  pkgarch_mapping(e.data)
> +                preferred_ml_updates(e.data)
> 
>  	if not data in e.__dict__:
>  		return
> 

Best Regards,
Lianhao

On Mon, 2011-08-08 at 16:57 +0800, Lu, Lianhao wrote:
> Richard Purdie wrote on 2011-08-05:
> > When running multilib builds its not intuitive for the specificed
> > versions/providers to be build in one case and the higher version in the other.
> > This patch mirrors the options unless they're already set.
> > 
> > There shouldn't be any effect if the MULTILIBS variable isn't set.
> > 
> > [YOCTO #1339]
> > 
> > Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org> ---
> > diff --git a/meta/classes/base.bbclass b/meta/classes/base.bbclass index
> > ff960cd..4aee847 100644 --- a/meta/classes/base.bbclass +++
> > b/meta/classes/base.bbclass @@ -139,6 +139,46 @@ def pkgarch_mapping(d):
> >          if d.getVar("TUNE_PKGARCH", True) == "armv7a-vfp-neon":
> >              d.setVar("TUNE_PKGARCH", "armv7a")
> > +def preferred_ml_updates(d):
> > +    # If any PREFERRED_PROVIDER or PREFERRED_VERSIONS are set,
> > +    # we need to mirror these variables in the multilib case
> > +    multilibs = d.getVar('MULTILIBS', True) or ""
> > +    if not multilibs:
> > +        return
> > +
> > +    prefixes = []
> > +    for ext in multilibs.split():
> > +        eext = ext.split(':')
> > +        if len(eext) > 1 and eext[0] == 'multilib':
> > +            prefixes.append(eext[1])
> > +
> > +    versions = []
> > +    providers = []
> > +    for v in d.keys():
> > +        if v.startswith("PREFERRED_VERSION_"):
> > +            versions.append(v)
> > +        if v.startswith("PREFERRED_PROVIDER_"):
> > +            providers.append(v)
> > +
> > +    for v in versions:
> > +        val = d.getVar(v, False)
> > +        pkg = v.replace("PREFERRED_VERSION_", "")
> > +        for p in prefixes:
> > +            newname = "PREFERRED_VERSION_" + p + "-" + pkg
> > +            if not d.getVar(newname, False):
> > +                d.setVar(newname, val)
> > +
> > +    for prov in providers:
> > +        val = d.getVar(prov, False)
> > +        pkg = prov.replace("PREFERRED_PROVIDER_", "")
> > +        virt = ""
> > +        if pkg.startswith("virtual/"):
> > +             pkg = pkg.replace("virtual/", "")
> > +             virt = "virtual/"
> > +        for p in prefixes:
> > +            newname = "PREFERRED_PROVIDER_" + virt + p + "-" + pkg
> > +            if not d.getVar(newname, False):
> > +                d.setVar(newname, p + "-" + val)
> 
> How about the native and nativesdk packages? I think they should not be prefixed.

Agreed. I've merged this with a tweak to stop the preferred
version/provider code taking effect for native/nativesdk packages.

Cheers,

Richard