---------- Forwarded message ---------
From:
Mickael Istria <mistria@xxxxxxxxxx>Date: Thu, Nov 29, 2018 at 9:57 AM
Subject: Re: MNG-6530 - Please fix or revert important regression from MNG-6311
To: <
info@xxxxxxxxx>
The idea was to introduce a system property in the worst case but not as
a solution...
As mentioned on the ticket, I'm pessimistic about existence of a solution. What was introduced in MNG-6311 is not a code issue, it's a design flaw caused by the fact that author was only considering a single use-case of this API and ignored other ones. Actually, in an IDE, we may really want to avoid an application lifetime cache as it was introduced with MNG-6311 for reasons explained on the bug (memory retained for no good reason), forcing all non-Maven batch adopters to create the own cache management -if a good cache API can be found first, which is far from being trivial- is really overkill for them.
IMO, we're already in the worst case. I'll prepare a patch for such system property so at least there is hope we can one day adopt 3.6.x in m2e.
> I'm against the system property cause we would open the door for system properties for each change we make which will be a horror.
It's not each change that's made, it's any regression that isn't reverted and that cause bugs in many projects downstream.
> We should try to find a better solution...and do not walk the short path...
The short path was taken on Oct 10th when ignoring Fred's comment and being fine introducing a regression.
But I'm 100% in favor of a good fix, I just happen to be pessimistic about such fix actually exists.
--
