BTW, help with doing this is very actively solicited; I expect we could substantially speed up MacPorts if we figure out what's going wrong here, and that would be a good thing for everyone.
On Tue, 15 Oct 2019 09:56:09 -0400 "Perry E. Metzger" <pe...@piermont.com> wrote: > On Tue, 15 Oct 2019 08:53:22 -0400 "Perry E. Metzger" > <pe...@piermont.com> wrote: > > Under Catalina, it's taking a very long time to complete > > installation of texlive-fonts-extra - the install phase seems to > > hang for about 15 minutes. I vaguely suspect something slow is > > happening in the registry part of the operation. > > > > Could someone else who has Catalina installed try installing > > texlive-fonts-extra from source and see if they get the same > > problem? > > Adding additional information: > > "time port -s install texlive-fonts-extra" yields: > > real 23m0.587s > user 18m34.592s > sys 3m26.868s > > Note the "user" time there, which is truly extraordinary. So, I'd still like to know how to profile Tcl code, but: I've done some very crude timing work by inserting print statements into various parts of portinstall::install_main that dump out the time (in seconds) at various stages of processing. This very crude method indicates most of the mysterious CPU spin for install of tex-fonts-extra takes place in the registry::write section of that function -- about 15 minutes worth. So there's definitely weirdness here with slow registry operation, though it might not be database I/O as such. That said, almost everything is surprisingly slow. There's several CPU minutes in Tcl here in the install_main function even before the registry write begins. That's not several wall-clock minutes with most of it being spent in I/O, that's CPU time. That shouldn't be. Generally, I'm pretty surprised at how slow all of this is; I would have expected the operation to be I/O bound. I suspect that after figuring out what is going wrong with the registry::write part, it might be good to profile all of this and figure out why it is taking quite so long. I suspect we could get the 18.5m of user time way, way down. Perry -- Perry E. Metzger pe...@piermont.com
