[oe] [PATCH 1/1] libbonobo.inc: add libbonobo-bin package

[Koen Kooi]

> Op 19 nov. 2014, om 12:17 heeft Paul Eggleton <paul.eggleton at linux.intel.com> het volgende geschreven:
> 
> On Wednesday 19 November 2014 08:53:11 Koen Kooi wrote:
>>> Op 19 nov. 2014, om 01:36 heeft Paul Eggleton
>>> <paul.eggleton at linux.intel.com> het volgende geschreven:> 
>>> On Tuesday 18 November 2014 19:32:19 Koen Kooi wrote:
>>>>> Op 18 nov. 2014, om 16:27 heeft Paul Eggleton
>>>>> <paul.eggleton at linux.intel.com> het volgende geschreven:
>>>>> 
>>>>> Hi Koen,
>>>>> 
>>>>> On Tuesday 18 November 2014 13:48:15 Koen Kooi wrote:
>>>>>> Hongxu Jia schreef op 13-11-14 12:42:
>>>>>>> On 11/13/2014 07:02 PM, Koen Kooi wrote: Hongxu Jia schreef op
>>>>>>> 12-11-14
>>>>>>> 
>>>>>>> 03:13:
>>>>>>>>>> The previous libbonobo package contains perl scripts, while
>>>>>>>>>> installing libbonobo, package management detected perl required,
>>>>>>>>>> but it actually not needed as lib. So add libbonobo-bin package to
>>>>>>>>>> contain these scripts. It refers debian:
>>>>>>>>>> https://packages.debian.org/sid/libbonobo2-bin
>>>>>>>>>> 
>>>>>>>>>> Signed-off-by: Hongxu Jia <hongxu.jia at windriver.com> ---
>>>>>>>>>> meta-gnome/recipes-gnome/bonobo/libbonobo.inc | 3 ++- 1 file
>>>>>>>>>> changed, 2 insertions(+), 1 deletion(-)
>>>>>>>>>> 
>>>>>>>>>> diff --git a/meta-gnome/recipes-gnome/bonobo/libbonobo.inc
>>>>>>>>>> b/meta-gnome/recipes-gnome/bonobo/libbonobo.inc index
>>>>>>>>>> 8b6007e..e0f6168 100644 ---
>>>>>>>>>> a/meta-gnome/recipes-gnome/bonobo/libbonobo.inc +++
>>>>>>>>>> b/meta-gnome/recipes-gnome/bonobo/libbonobo.inc @@ -5,10 +5,11 @@
>>>>>>>>>> LIC_FILES_CHKSUM =
>>>>>>>>>> "file://COPYING;md5=94d55d512a9ba36caa9b7df079bae19f \ SECTION =
>>>>>>>>>> "x11/gnome/libs" DEPENDS = "glib-2.0 orbit2 intltool-native libxml2
>>>>>>>>>> dbus dbus-glib"
>>>>>>>>>> 
>>>>>>>>>> -inherit gnomebase gtk-doc +inherit gnomebase gtk-doc lib_package
>>>>>>>>>> 
>>>>>>>>>> ORBIT_IDL_SRC = "${STAGING_BINDIR_NATIVE}/orbit-idl-2"
>>>>>>>>>> 
>>>>>>>>>> +RDEPENDS_${PN}-bin = "${PN} perl" FILES_${PN} +=
>>>>>>>>>> "${libdir}/orbit-2.0/*.so ${libdir}/bonobo/monikers/*.so"
>>>>>>>>>> FILES_${PN}-dbg += "${libdir}/bonobo/monikers/.debug \
>>>>>>>>>> ${libdir}/bonobo-2.0/samples/.debug ${libdir}/orbit-2.0/.debug"
>>>>>>> 
>>>>>>> This breaks the upgrade path
>>>>>>> 
>>>>>>>> I am sorry I do not understand what you mean, do you mean running
>>>>>>>> 'smart upgrade' failed?
>>>>>> 
>>>>>> You are moving files between packages, which makes 'opkg update ; opkg
>>>>>> upgrade' fail with conflicts unless you carefully craft
>>>>>> RREPLACES/RCONFLICTS or enforce ordering in upgrade order (which isn't
>>>>>> currently possible).>
>>>>> 
>>>>> So what would you suggest - never make any packaging changes?
>>>> 
>>>> No. I suggest thinking about what you're doing and minimizing the damage.
>>>> So far the patches I've seen are misguided (recipe==package assumption)
>>>> or plain weird. And in all cases break upgrade paths. Since arguing the
>>>> use case seems to fall on deaf ears I can only argue the technical
>>>> problems with them.
>>> 
>>> What do you mean by "arguing the use case" ?
>> 
>> Making recipe name and package name match, see the avahi patch for OE-core
> 
> Ah, except not only did those patches not get merged, the submitter even asked 
> for the patches to be dropped after the discussion. That hardly counts as 
> "falling on deaf ears".

Great, I can never be sure if the lack of comments from the powers that be mean approval or dismissal.Apart from Martin no one commented on the recipe==package assumptionm which usually means the patch will get merged.

> 
>>>>> It's not as if this is the first time we've done this kind of thing...
>>>> 
>>>> And people still sound surprised when I say upgrade paths are broken
>>>> between releases. You yourself told me I should speak up when seeing
>>>> such patches and so I did.
>>> 
>>> I did yes, and I hadn't forgotten. However we do need to make packaging
>>> changes from time to time, so we need to work out how to do them properly
>>> -
>>> and I'll bang the same drum as I always do - if we want to ensure these
>>> changes are properly considered and done in the correct manner, we need to
>>> have those things documented. If we do not, you shouldn't be surprised
>>> when people miss things. This is clearly an area you are interested in so
>>> it's an area you could help document.
>>> 
>>> Getting back to this specific problem, when splitting out a set of files
>>> into a different package as is being done by this patch, as far as I am
>>> aware there is no means to ensure that the new package gets installed on
>>> upgrade without introducing a hard dependency which would defeat the
>>> purpose of making the change. So that means you either make the change or
>>> you don't.
>> 
>> Not quite, but let's assume so for this case.
> 
> Well if that's not accurate then let's be clear - why not ... ?

"either you make the change or you don't" is creating something like a false dilemma. For the rest of your analysis, see below.

> 
>>> Are any of
>>> these files that are being split out actually of interest on a typical
>>> system that OE would build for? If not, do we care that those files will
>>> disappear?
>> 
>> If the package were the file gets moved to is not installed by default this
>> change will cause no problems, but for all others (e.g. people with -dev
>> and/or -dbg installed) it will.
> 
> I'm not following. Won't the corresponding -dev/-dbg packages be updated at 
> the same time?

In the opkg case they will get upgraded sequentially in reverse dependency order (or something close to it, opkg has seen a lot of fixes in that area). If the package where the file moved to gets upgraded before the package where it used to be the upgrade will fail.

In this specific case, I don't know what the perl script does, I'd be tempted to move it into its own package and have the -dev package RRECOMMEND it if it's for development. If it's not for development, just put it in its own package and see if RRECOMMENDS/RDEPENDS are needed. And make sure someone actually runtime tests the results.