[jbosstools-dev] Packaging JBoss Tools for Fedora (issue & potential patches)

Denis Golovin dgolovin at exadel.com
Tue Aug 21 13:05:35 EDT 2012


On 08/20/2012 04:22 AM, Max Rydahl Andersen wrote:
> On 20 Aug 2012, at 12:56, Mickael Istria <mistria at redhat.com> wrote:
>
>> On 08/20/2012 12:39 PM, Max Rydahl Andersen wrote:
>>>>> Ah ok, I wasn't aware why that was happening. I guess it would be
>>>>> 'higher' because of the build date in the qualifier, right?
>>>>>
>>>>>
>>>> That's it, but more generally, it happens be qualifier is higher, not only date.
>>>>
>>> yes, but for us it is the date; which is our fault and another reason why we should not have date be the significant decider.
>> Having date in qualifier makes it very readable and user-friendly. It's quite useful.
> I did not say *remove* the date, I said it should not be the significant decider for p2.
>
> i.e. Release of a version should make any nightly/snapshot builds of that version invalid; today it does not because the data comes *before* the release type flag.
>
> proper versioning that has semantic API meaning instead of just rely on what date the build was made.

In trunk it is now uses release type in front of the date. It was fixed 
a month ago and works fine since.
I uses '${BUILD_ALIAS}-v'yyyyMMdd-HHmm' for local builds and 
'${BUILD_ALIAS}-v'yyyyMMdd-HHmm'-H${BUILD_NUMBER}' for hudson

Denis

>
>>> I believe eclipse.org does this by doing releases like 3.2.0, 3.2.100, 3.2.200 ...giving room for 99 "custom" version builds.
>> Only platform does that. I'm not aware of any other project using this versioning AFAIK.
> WTP does too afaik.
>
>>>>> How about
>>>>> if I force the date to be lower than the your build from the update
>>>>> site? That way, if one downloads from the eclipse marketplace, or
>>>>> somewhere else, then that would be used instead of the Fedora
>>>>> installed version. Am I right in thinking that?
>>>>>
>>>>>
>>>> Qualifier pattern for us is v<...>. If you want your build pattern to be lower, don't deal with dates. You can change simply qualifier to something like: f<...> or fedora-v<...>. Since 'f' < 'v' our bundles will be preferred by p2 and OSGi independently of the date on so on. I like the fedora-v<...> since it answers this issue and a previous one at the same time.
>>>> Be aware that you are dealing with conventions here. You should strongly document them and get other people working on them aware of them. It's easy to have someone changing this pattern for a "better" one without knowing such rules and then breaking it.
>>>>
>>> This doesn't really solve it does it ? not unless all jboss.org updatesites are removed from p2 since when the user does Help > Update it will bump everything.
>> So we want to prevent updates of JBT plugins from inside Eclipse when installed with Fedora? I did not understand it that way.
>> IF we want to block install of JBT ww probably can't rely on p2 as it. p2 compares version, we can't prevent it to update from 3.3.0.fedora-v2012 to 3.4.0.v2012... If we need to block this behavior, we'll probably need to hack p2 to prevent from any installation of org.jboss* IUs.
> no, I'm saying as long as Fedora needs to do changes to the content and not *just* rebuild it then anything like p2 updates and jboss central and m2e connectors (anything that uses p2) will
> be very fragile under the Fedora distribution.
>
>>>>>>> Another aspect is what to do with our JBoss Central feature - which also relies on eclipse p2 to install additional features.
>>>>>>>
>>>> Did you change the JBoss Central feature to rely on yum instead of p2?
>>>> If yes, that's a different feature, and it modified bundles and features probably requires a new name.
>>>> If no, then it means there is nothing to worry about. It will be installed and will refer to our sites, and will work as always.
>>>>
>>> No it won't if the bundles done in fedora doesn't have the same API (i.e. hibernatetools not bundling hibernate 3.5 for example)
>>>
>>> It will *seem* to work, but funky sideeffects will eventually happen the more differences there are.
>>>
>> Yet another use-case I didn't have in mind. This Fedora integration is a real puzzle. Is there a document that sums up the installation scenario to be allowed/forbidden when using Fedora JBT package?
> I don't think we can do an "Allowed/forbidden" one  - we can basically just say "if central installs seem to work great! if it does not and you used the Fedora install please uninstall it via yum and install from marketplace"
>
> /max
> _______________________________________________
> jbosstools-dev mailing list
> jbosstools-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/jbosstools-dev



More information about the jbosstools-dev mailing list