RDEPENDS_${PN}-dev = ${PN} badness

RDEPENDS_${PN}-dev = ${PN} badness

[John Lee]

Hi,

Take gtk-doc-dev for example:

The gtk-doc package is empty, so it will not be created.  However, if
you install gtk-doc-dev, it RDEPENDS on gtk-doc, which cannot be
found.  The install will fail.

I think we should use

RRECOMMENDS_${PN}-dev = "${PN} (= ${DEBPV})"

instead.

We (Julian Chu and me) found this when digging the cause of
meta-toolchain-openmoko build failure.  task-openmoko-toolchain-target
contains quite a few -dev packages in it, so the opkg-cl install will
fail.  The log looks something like this:

ERROR: Cannot satisfy the following dependencies for
         task-openmoko-toolchain-target: gtk-doc (= 1.0-r1)
         util-macros (= 1:1.1.5-r1) renderproto (= 1:0.9.3-r0) xproto
         (= 1:7.0.10-r1) bigreqsproto (= 1:1.0.2-r1) xextproto (=
         1:7.0.2-r1) xtrans (= 1:1.0.4-r0) xcmiscproto (= 1:1.1.2-r1)
         xf86bigfontproto (= 1:1.1.2-r1) kbproto (= 1:1.0.3-r1)
         inputproto (= 1:1.4.2.1-r0) fixesproto (= 1:4.0-r1)
         xineramaproto (= 1:1.1.2-r1) randrproto (= 1:1.2.1-r1)
         compositeproto (= 1:0.4-r0) damageproto (= 1:1.1.0-r1)

Similiar symptoms can be found in the packages above.

Regards,
John

Re: RDEPENDS_${PN}-dev = ${PN} badness

[Koen Kooi]

John Lee wrote:
> Hi,
>
> Take gtk-doc-dev for example:
>
> The gtk-doc package is empty, so it will not be created.  However, if
> you install gtk-doc-dev, it RDEPENDS on gtk-doc, which cannot be
> found.  The install will fail.
>
> I think we should use
>
> RRECOMMENDS_${PN}-dev = "${PN} (= ${DEBPV})"
>
> instead.

ALLOW_EMPTY_${PN} = "1" would be a better idea, and in line without 
other depchain workarounds.

regards,

Koen

>
> We (Julian Chu and me) found this when digging the cause of
> meta-toolchain-openmoko build failure.  task-openmoko-toolchain-target
> contains quite a few -dev packages in it, so the opkg-cl install will
> fail.  The log looks something like this:
>
> ERROR: Cannot satisfy the following dependencies for
>           task-openmoko-toolchain-target: gtk-doc (= 1.0-r1)
>           util-macros (= 1:1.1.5-r1) renderproto (= 1:0.9.3-r0) xproto
>           (= 1:7.0.10-r1) bigreqsproto (= 1:1.0.2-r1) xextproto (=
>           1:7.0.2-r1) xtrans (= 1:1.0.4-r0) xcmiscproto (= 1:1.1.2-r1)
>           xf86bigfontproto (= 1:1.1.2-r1) kbproto (= 1:1.0.3-r1)
>           inputproto (= 1:1.4.2.1-r0) fixesproto (= 1:4.0-r1)
>           xineramaproto (= 1:1.1.2-r1) randrproto (= 1:1.2.1-r1)
>           compositeproto (= 1:0.4-r0) damageproto (= 1:1.1.0-r1)
>
> Similiar symptoms can be found in the packages above.
>
> Regards,
> John

Re: RDEPENDS_${PN}-dev = ${PN} badness

[Richard Purdie]

On Thu, 2008-04-17 at 22:00 +0200, Koen Kooi wrote:
> John Lee wrote:
> > Take gtk-doc-dev for example:
> >
> > The gtk-doc package is empty, so it will not be created.  However, if
> > you install gtk-doc-dev, it RDEPENDS on gtk-doc, which cannot be
> > found.  The install will fail.
> >
> > I think we should use
> >
> > RRECOMMENDS_${PN}-dev = "${PN} (= ${DEBPV})"
> >
> > instead.
> 
> ALLOW_EMPTY_${PN} = "1" would be a better idea, and in line without 
> other depchain workarounds.

If ${PN} is empty and its existence makes no sense I think its
acceptable to do:

RDEPENDS_${PN}-dev = ""
RRECOMMENDS_${PN}-dbg = "${PN}-dev (= ${DEBPV})"

in the recipe itself. linux-libc-headers.inc does this for example since
${PN} makes no sense there.

There really aren't that many packages like this although there may be
some that have been dealt with in Poky but not in OE.

Cheers,

Richard

Re: RDEPENDS_${PN}-dev = ${PN} badness

[Michael 'Mickey' Lauer]

On Friday 18 April 2008 01:24:46 Richard Purdie wrote:
> On Thu, 2008-04-17 at 22:00 +0200, Koen Kooi wrote:
> > John Lee wrote:
> > > Take gtk-doc-dev for example:
> > >
> > > The gtk-doc package is empty, so it will not be created.  However, if
> > > you install gtk-doc-dev, it RDEPENDS on gtk-doc, which cannot be
> > > found.  The install will fail.
> > >
> > > I think we should use
> > >
> > > RRECOMMENDS_${PN}-dev = "${PN} (= ${DEBPV})"
> > >
> > > instead.
> >
> > ALLOW_EMPTY_${PN} = "1" would be a better idea, and in line without
> > other depchain workarounds.
>
> If ${PN} is empty and its existence makes no sense I think its
> acceptable to do:
>
> RDEPENDS_${PN}-dev = ""
> RRECOMMENDS_${PN}-dbg = "${PN}-dev (= ${DEBPV})"
>
> in the recipe itself. linux-libc-headers.inc does this for example since
> ${PN} makes no sense there.
>
> There really aren't that many packages like this although there may be
> some that have been dealt with in Poky but not in OE.

How much noise would a poky/OE diff contain? I guess more of us should try to 
merge things from Poky.

:M: