[oe] [PATCH] x264_r2245: fix DEPENDS for x86

[Phil Blundell]

On Sun, 2011-05-08 at 11:16 +0200, Paul Menzel wrote:
> Thank you for the clarification. Is for example `RDEPENDS_${PN}` also an
> override?

Not in the sense I was meaning here, no.  (It is true that the ${PN} bit
is technically an OVERRIDE, but in most/all cases RDEPENDS_${PN} is
effectively a primary variable, i.e. there is no underlying RDEPENDS
which it overrides.)

The issue with the "DEPENDS_x86" thing is that, as far as the initial
metadata processing is concerned, DEPENDS and DEPENDS_x86 are distinct
variables.  You might hope that "DEPENDS_x86 += ..." would do a
conditional append to DEPENDS, but it doesn't: what happens is that it
creates a new variable "DEPENDS_x86" which then replaces the contents of
DEPENDS during override processing.  So the initial value of ${DEPENDS}
is lost rather than being appended to.

This whole issue is basically the whole reason that the "_append" syntax
hasn't been deprecated, since we don't currently have any other way to
do conditional appending (apart from resorting to python).

>  If yes, then looking at
> 
> 	$ git grep 'DEPENDS_'
> 
> returns a lot of incorrect usages.

>From a quick glance at that output nothing jumped out at me as being
incorrect, but there might well be a few wrong ones in there.  This one,
for example:

classes/vala.bbclass:DEPENDS_virtclass-native += "vala-native"

looks a bit dubious to me (for the same reasons as above) but without
checking the recipe context it's impossible to say whether it's actually
wrong or not.

p.