So the rumor is that xpinstall is one of the components among others gets loaded at start-up. :-) Due to the fact we are trying to improve start-up speed, is it possible to delay the loading of xpinstall, with some lazy initialization, till later?

There are two things in the way of this. One is changing the startup routine (post-install replace) which is a bug almost checked in. But the other reason is that we need to register and support the InstallTrigger object, and the only way to do that is to register a name set at startup. Vidur at one point said there was a plan to drive these from categories or something, so that the component only needs to get instantiated when used. But I can't find a bug that sounds like that so I'm not hopeful. I'll nominate for the hell of it -- sure would be nice since 99% of the time people don't need xpinstall functionality.

adding perf keyword. vidur, do you have any info on supporting init a component only at the time it's requested?

Provisionally plussing, but can't do it unless jst does bug 71982

I have bug 71982 partly fixed on my DOM/XPConnect branch, once that lands this will be fixed, I might ask for some help with the xpinstall changes required on the branch if that's ok?

Sure, I'll be happy to help (or have you help me) with the XPInstall change once the DOM change is in.

The XPCDom landed, is this really fixed now? Re-nominating since clearing away DLL loads is one of our best bets for real startup perf improvements.

This is also dependent on the post-install replace bug, but should be doable soon.

It's time to move this off m.9.1 milestone.

Ack, I've got the fixes, just waiting for Don's PIRU bug to land (he's seeking a third round of reviews now)

reassigning

actually, reassigning back since dan has fixes. This will have to go in for RTM or next release.

Dan, what's the status of your patches for this?

moving to 0.9.3 because Dan is on sabbatical.

This has been fixed for a while. Big thanks to dbragg -- for "post-install replaces" jst -- XPCDOM made this possible, and he even changed our code to start using the new mechanism for us cathleen/waterson -- cleaning up after the "autoreg every time" hack I put in right before my sabbatical because I ran out of time to fix this one the right way.

Marking Verified. Those have been in for quite awhile now.