On Fri, 2005-04-08 at 19:02 +0100, Mike Hearn wrote:
On Thu, 07 Apr 2005 21:09:53 -0400, John Thacker wrote:
Similarly, we expected people to have to do export LD_ASSUME_KERNEL=2.2.5, etc., because their old apps (like acroread4) wouldn't work otherwise. That seems pretty hard for a newbie to figure out; I've had to tell people about it. They don't like their apps suddenly breaking on upgrade.
LD_ASSUME_KERNEL was an absolute usability disaster, it's not something anyone should be pointing to as justification. Threading really was not a bottleneck on most peoples systems, and it would have made much more sense for NPTL to be opt-in, instead of opt-out.
The end result of stuff like this is that all the good work projects like GNOME have done gets thrown away, because the moment Joe User wants to run a cool game or their mission critical app, they have to dick about with the command line setting obscure meaningless variables that you Just Have To Know.
You know, usually the workaround is opt-in instead of default, and for good reasons: if you change things (supposedly for the better), you want applications to be adjusted/fixed if necessary. This works so much better if you need to opt-in for the workaround to let old apps run -- the pressure for application developers to fix their applications is higher. With Windows, you would usually just have to bite the dust if your application doesn't work on the next OS version -- LD_ASSUME_KERNEL being awkward for the end user beats this hands down. Of course it would be even better if there were some kind of infrastructure to handle this better for the easily intimidated ;-).
Nils