[Glass] migrate problem
Johan Brichau
johan at yesplan.be
Thu Jan 16 11:41:40 PST 2014
Norbert,
Given that you are searching for a solution, I thought I quickly write down how we deal with these things:
I think #migrateTo: only configures the class for migration but does not execute it. When reviewing some of our past migration scripts, I notice I have always been using #migrateInstancesTo:
But for any code change where no class names are changed (but other class migrations are necessary), I'm loading the new code in topaz using Metacello and launch the GsDeployer:
MCPlatformSupport commitOnAlmostOutOfMemoryDuring: [
[
(Smalltalk at: #ConfigurationOfNextPlan) load.
]
on: Warning
do: [:ex | Transcript show: ex greaseString. ex resume]].
GsDeployer new doBulkMigrate.
An important detail is that the Metacello configuration must be set to #atomic loading. If you leave it to the standard 'linear' loading, you might end up breaking a classhistory in the loading process, requiring you to manually add the new versions to the class history. This is because the linear loading might first load the package where the class was removed, hence removing the class definition from your system. When the other package that now contains the class is loaded afterwards, the monticello load does not detect is a new version of the old class.
Since we use the above way of loading, we have only had to write manual migrations scripts when we do refactorings that move variables around in the class history such that the loading order becomes important.
Johan
On 16 Jan 2014, at 20:24, Johan Brichau <johan at yesplan.be> wrote:
> Norbert,
>
> I had the problem of 'double classes' in the past as well when I did loading via Monticello in gemtools rather than in topaz and using the GsDeployer.
> I retained the following script to fix that (which I believe was whipped up by Dale at that time)
>
> |classesWithOldSuperclasses|
> classesWithOldSuperclasses := Set new.
> Smalltalk allClasses do:[:s |
> Smalltalk allClasses do:[:c |
> (c superclass name = s name and:[(c superclass = s) not]) ifTrue:[classesWithOldSuperclasses add: c]]].
> classesWithOldSuperclasses do:[:c | Transcript show: c name;cr. c definition evaluate]
>
> Hope it helps
> Johan
>
> On 16 Jan 2014, at 20:18, Norbert Hartl <norbert at hartl.name> wrote:
>
>> Dale,
>>
>> that didn’t work. Now I have some of the classes doubled and some tripled. I need to try a few more things. I was just wondering that the migrateTo: does not work as expected. I did this a few times in the past and it always worked like that. Or I just can’t remember correctly.
>>
>> Norbert
>>
>> Am 16.01.2014 um 17:37 schrieb Dale Henrichs <dale.henrichs at gemtalksystems.com>:
>>
>>> Norbert,
>>>
>>> I am not a migration expert and I haven't read the docs (yet), but in reading the code, I think that you are supposed to send the #migrate message to instances of your class ...
>>>
>>> Have you looked at GsDeployer? I think the following will do a pretty good job of migrating the instances for you:
>>>
>>> GsDeployer bulkMigrate: ["add instance variable to class"]
>>>
>>> in fact I think right now you should be able to do:
>>>
>>> GsDeployer bulkMigrate: ["empty block"]
>>>
>>> and get "pending migrations performed"....
>>>
>>> Dale
>>>
>>>
>>> On Thu, Jan 16, 2014 at 7:25 AM, Norbert Hartl <norbert at hartl.name> wrote:
>>> I have one class hierarchy that has between 40 and 50 Mio. instances. Now I needed to add an instVar to the root of the hierarchy. Doing it with gemtools I switched off auto migration and loaded the monticello package containing the changed. But now I have problems migrating the instances manually.
>>>
>>> COEvent allSubclasses size
>>>
>>> gives 5 while
>>>
>>> COEvent classHistory first allSubclasses size
>>>
>>> gives 23. A
>>>
>>> COEvent classHistory first migrateTo: COEvent
>>>
>>> didn’t do the trick. And
>>>
>>> COEvent migrate
>>>
>>> gives me
>>>
>>> a MessageNotUnderstood occurred (error 2010), a Metaclass3 does not understand #’migrationDestination'
>>>
>>> I’m using
>>>
>>> GemStone version ‚3.1.0.2'
>>>
>>> Any hints,
>>>
>>> Norbert
>>>
>>> _______________________________________________
>>> Glass mailing list
>>> Glass at lists.gemtalksystems.com
>>> http://lists.gemtalksystems.com/mailman/listinfo/glass
>>>
>>>
>>
>> _______________________________________________
>> Glass mailing list
>> Glass at lists.gemtalksystems.com
>> http://lists.gemtalksystems.com/mailman/listinfo/glass
>
More information about the Glass
mailing list