[ Resent from correct account, so MacPorts won't bounce it back again ]

On Jan 2, 2008, at 12:22 PM, Tabitha McNerney wrote:

Has anyone else noticed incredible MacPorts speed up improvements when installing various ports?

Here's an example: The same hardware (Xserve Intel 2 x 2 GhZ Dual Core Intel). Once this machine was running Tiger Server with MacPorts 1.5.0 and I installed the gcc41 port which (if my notes are accurate) took about 88 minutes to install from beginning to end (some of the time could have been eaten up in the "fetching" of the resource source files). Anyway, the same port on Leopard Server on the same machine with MacPort 1.6.0 took me roughly 20 minutes to install today. From 88 to 20 minutes is almost unheard of. I checked the port files too, to make sure the version of the gcc41 port I was installing was the same:

Tiger Server:

gcc41 4.1.2, Revision 2, lang/gcc41 (Variants: universal, darwin_8, darwin_i386, powerpc, odcctools)

Leopard Server:

gcc41 4.1.2, Revision 2, lang/gcc41 (Variants: universal, darwin_8, darwin_i386, powerpc, odcctools)

I presume this is because Leopard and Leopard Server are now leveraging the Apple Developer tools with a complete 64-bit wide architecture (both the hardware and the software)?

No, I think it's because Leopard's Tcl no longer searches an insane search path (including network volumes) as part of its startup.  When we made this change in Leopard (largely because of pathological performance problems I noticed with MacPorts and fs_usage(1)), a full minute was shaved off the most basic port operations, including build/install (it's more observable with something like "port search").

- Jordan