26 May 2021 |
andi- | * ah nice so we got at last Go 1.12.2 | 10:00:16 |
andi- | * ah nice so we got at least Go 1.12.2 | 10:00:22 |
andi- | Given how fast Go usually compiles that should be a bootstrap that is done in less than 30min? | 10:00:48 |
Gytis Ivaskevicius | I have not looked into dependencies of go, but it should not take long | 10:01:19 |
andi- | My assumption was that all the deps are statitified. Purely the Go -> Go' -> Go'' -> Go''' ... dance | 10:02:19 |
Gytis Ivaskevicius | I was helping out Dram fixing his super minimal gcc derivation https://github.com/dramforever/scratchix/blob/main/pkgs/stage1/gcc.nix - for me it compiles under 2min (tho i got a powerful pc) | 10:02:20 |
andi- | IMHO it is fair to assume that hydra builders have 16 threads in 2021... If it takes one of those boxes less than a few hours to do the full bootstrap it should be fine. On my dated 2020 workstation I can build bootstrap -> Qemu in about 10 minutes. | 10:03:33 |
andi- | Or was it 15? Anyway, reasonably fast for debugging.. not fun for watching | 10:04:03 |
Gytis Ivaskevicius | i thought they bought 64 core monster at some point 🤔 | 10:04:15 |
Gytis Ivaskevicius | might be confusing it with something | 10:04:21 |
andi- | The hydra build farm is a mixture that is mostly made up of Packet.net computers but there are some older "on-premise" beasts. | 10:04:49 |
Gytis Ivaskevicius | Ah, in that case i probably just saw someones htop, maybe zimbat? not sure | 10:08:04 |
Gytis Ivaskevicius | it was a while back | 10:08:10 |
andi- | We have gccgo6 in nixpkgs and I'm right now trying to compile it for gcc9 | 10:09:03 |
Jonas Chevalier | you probably saw john ringer's beefy machine | 10:09:37 |
Gytis Ivaskevicius | quite possilby | 10:10:09 |
Jonas Chevalier | ^ digging into scratchix. it's a cool idea | 10:10:38 |
Gytis Ivaskevicius | yep, very cool | 10:11:00 |
Jonas Chevalier | even just as a means of re-thinking how the bootstrap would work | 10:11:13 |
andi- | How far did it go? I see it is still missing anything like make to actually build something without the bootstrap tarball | 10:11:19 |
| immae (he/him) changed their display name from immae to immae (he/him). | 10:13:07 |
andi- | When I did my bootstrap experiments I was going crazy with the amount of CC/LD/CXX/... wrapper that we have in nixpkgs. It is all over the place for probably good reasons but hard to grasp. | 10:13:51 |
Gytis Ivaskevicius | In reply to @andi:kack.it How far did it go? I see it is still missing anything like make to actually build something without the bootstrap tarball super early stages and nobody is working on it. There is not much done. It basically uses same boostrap as Nixpkgs and got all the way to compiling gcc | 10:13:58 |
Gytis Ivaskevicius | antoher gcc compilation is needed with glibc to start doing anything useful | 10:14:24 |
andi- | My woes started when I tried to do a 2nd stage build out of the previously bootstraped python, glibc, gcc, ... things would randomly fail due to rpaths | 10:14:39 |
raboof | https://reproducible-builds.org/contribute/nixos/ - feel free to improve further via https://salsa.debian.org/reproducible-builds/reproducible-website :) | 10:15:00 |
andi- | I could either get glibc to build with the GCC or glibc but never both :D | 10:15:05 |
andi- | * I could either get glibc to build with the GCC or GCC but never both :D | 10:15:23 |
andi- | At least it allowed me to test building using dash and oil shell which showed that you can't build oilshell with oilshell just yet :D | 10:16:06 |
Gytis Ivaskevicius | Also, check out ion shell, its pretty cool | 10:17:04 |