[news.eclipse.tools] Re: Possible to recover from corrupt workspace (.tr

[Eric Rizzo <eric.rizzo@xxxxxxxxxxx>]

Bob Foster wrote:
> > IMHO, workbench shutdown should not be allowed to
> > fail in such a harsh manner.
>
>
> This (running out of heap memory) is only one of several cases where the JVM
> can decide to die without possibility of recovery. You should direct your
> ire to Sun.

I should clarify: naturally, I understand Eclipse is at the mercy of the 
VM. What I think needs addressing in Eclipse is the fact that it has no 
fallback mechanism in case of VM failure while shutting down. I'm 
thinking a couple of changes might help:

1) Making shutdown as tight as possible, not requiring persisting of so 
much critical data. I don't know the internals, but perhaps more of the 
info can be persisted "as you go" instead of waiting for workbench 
shutdown. Possibly in the background. The goal should be to get shutdown 
as small as possible, so VM failure at that point is not so tragic.

2) Giving startup some fallback rules if it finds corrupt workspace 
data. Maintaining the "last known good" configuration and using that if 
all else fails seems like a decent strategy (even windoze has that :-). 
Based on my eventual relative success in getting back up without losing 
any uncommitted work and very little of my Eclipse setup, I can't 
imagine it is impossible to do that automatically or at least 
semi-automatically.
At the VERY least, a little more helpful logging pointing to the exact 
file(s) that are corrupt is needed.

I reiterate, I am not at all familiar with the internals of Eclipse, so 
I don't know how difficult these things would be to implement. I just 
think it was too hard for the user (me) to figure out how to restore 
things, especially when it turned out to be a pretty simple process once 
it was outlined.

I can hear the responses now: "File a request in Bugzilla."  Perhaps I 
will...

	Eric
--
Eric Rizzo
Software Architect
Jibe, Inc.
http://www.jibeinc.com