Unifying PostgreSQL Ebuilds

After an excruciating wait and years of learning PostgreSQL, it’s time to unify the PostgreSQL ebuilds.I’m not sure what the original motivation was to split the ebuilds, but, from the history I’ve seen on Gentoo, it has always been that way. That’s a piss-poor reason for continuing to do things a certain way. Especially when that way is wrong and makes things more tedious and difficult than they ought to be.

I’m to blame for pressing forward with the splitting the ebuilds to -docs, -base, and -server when I first got started in Gentoo. I knew from the outset that having them split was not a good idea. I just didn’t know as much as I do now to defend one way or the other. To be fair, Patrick (bonsaikitten) would have gone with whatever I decided to do, but I thought I understood the advantages. Now I look at it and just see disadvantages.

Let’s first look at the build times for building the split ebuilds:

1961.35user 319.42system 33:59.44elapsed 111%CPU (0avgtext+0avgdata 682896maxresident)k
46696inputs+2000640outputs (34major+34350937minor)pagefaults 0swaps
1955.12user 325.01system 33:39.86elapsed 112%CPU (0avgtext+0avgdata 682896maxresident)k
7176inputs+1984960outputs (33major+34349678minor)pagefaults 0swaps
1942.90user 318.89system 33:53.70elapsed 111%CPU (0avgtext+0avgdata 682928maxresident)k
28496inputs+1999688outputs (124major+34343901minor)pagefaults 0swaps

And now the unified ebuild:

1823.57user 278.96system 30:29.20elapsed 114%CPU (0avgtext+0avgdata 683024maxresident)k
32520inputs+1455688outputs (100major+30199771minor)pagefaults 0swaps
1795.63user 282.55system 30:35.92elapsed 113%CPU (0avgtext+0avgdata 683024maxresident)k
9848inputs+1456056outputs (30major+30225195minor)pagefaults 0swaps
1802.47user 275.66system 30:08.30elapsed 114%CPU (0avgtext+0avgdata 683056maxresident)k
13800inputs+1454880outputs (49major+30193986minor)pagefaults 0swaps

So, the unified ebuild is about 10% faster than the split ebuilds.

There are also a few bugs open that will be resolved by moving to a unified ebuild. Whenever someone changes anything in their flags, Portage tends to only pick up dev-db/postgresql-server as needing to be recompiled rather than the appropriate dev-db/postgresql-base, which results in broken setups and failures to even build. I’ve even been accused of pulling the rug out from under people. I swear, it’s not me…it’s Portage…who I lied to. Kind of.

There should be little interruption, though, to the end user. I’ll be keeping all the features that splitting brought us. Okay, feature. There’s really just one feature: Proper slotting. Portage will be informed of the package moves, and everything should be hunky-dory with one caveat: A new ‘server’ USE flag is being introduced to control whether to build everything or just the clients and libraries.

No, I don’t want to do a lib-only flag. I don’t want to work on another hack.

You can check out the progress on my overlay. I’ll be working on the updating the dependent packages as well so they’re all ready to go in one shot.

Leave a Reply