Just committed the patch to a branch, you can see it here https://github.com/xwiki/xwiki-commons/commit/32fb77049fe9d8f758e776d26bfe5… Since it contains an API break in EnvironmentConfiguration, I wanted to run it by the list before merging it into the master. I'll push later on today if nobody objects.

Caleb On 04/23/2012 09:43 AM, Caleb James DeLisle wrote: > Hi, > > In order to fix XWIKI-7748 and XWIKI-7749, we need a plan for deleting temporary file. > XWikiAttachmentContent uses File.deleteOnExit(), but in Sun's implementation, deleteOnExit() > only deletes on a *clean* exit, in a crash it can't delete the files but it doesn't tag them > so they can be deleted in the next run either. Since temp files are usually pseudo-randomly > named, they are impossible to find and delete later. > > I propose that we create a subdirectory, `java.io.tmpdir`/xwiki/ which will be removed on > application exit. If the jvm crashes, the files will be removed next time around. Then alter > ApplicationContext.getTemporaryDirectory() to yield this directory. > > WDYT? > > Caleb

On 05/02/2012 09:20 AM, Vincent Massol wrote: > Hi Caleb, > > On May 2, 2012, at 3:12 PM, Caleb James DeLisle wrote: > >> Just committed the patch to a branch, you can see it here >> https://github.com/xwiki/xwiki-commons/commit/32fb77049fe9d8f758e776d26bfe5… >> >> Since it contains an API break in EnvironmentConfiguration, I wanted to run it by the list before merging it into the master. >> I'll push later on today if nobody objects. > > We should no longer break APIs. So this will need to go through deprecation and legacy. > > Why is File an issue? > > Big -1 to the code I've seen. We should never ever use System.exit(). How do you envision the wiki behaving as a whole if there is no writable temp dir? I agree that killing the jvm is very bad if there is any way to recover but this seems like the edge case where there is no way out.

> Also I've seen a lot of "final" used in the code. Please don't use them since they're unnecessary and bad in most cases and we don't have any best practice to use them in methods for ex (this prevents extensibility). As for variables I don't see the point at all. I can remove the final annotation on the methods, I think it's important however to differentiate between clean extensibility and using "extends" to monkey patch code in ways that weren't intended. As for variables and fields, I am very fond of making those final because they are inline documentation and verification. I declare my intent not to change a variable with the final keyword, the reader/maintainer knows what I meant and the compiler checks that I didn't make a mistake. If you want me to strip those, I can but think it's like stripping comments. Caleb > > Re the fail fast I was expecting some event listener listening on Application started event to check if the directories were writable for example and throw some exception if not. > > Thanks > -Vincent > >> Caleb >> >> >> On 04/23/2012 09:43 AM, Caleb James DeLisle wrote: >>> Hi, >>> >>> In order to fix XWIKI-7748 and XWIKI-7749, we need a plan for deleting temporary file. >>> XWikiAttachmentContent uses File.deleteOnExit(), but in Sun's implementation, deleteOnExit() >>> only deletes on a *clean* exit, in a crash it can't delete the files but it doesn't tag them >>> so they can be deleted in the next run either. Since temp files are usually pseudo-randomly >>> named, they are impossible to find and delete later. >>> >>> I propose that we create a subdirectory, `java.io.tmpdir`/xwiki/ which will be removed on >>> application exit. If the jvm crashes, the files will be removed next time around. Then alter >>> ApplicationContext.getTemporaryDirectory() to yield this directory. >>> >>> WDYT? >>> >>> Caleb > _______________________________________________ > devs mailing list > devs(a)xwiki.org > http://lists.xwiki.org/mailman/listinfo/devs > _______________________________________________ devs mailing list devs(a)xwiki.org http://lists.xwiki.org/mailman/listinfo/devs