| 26 May 2021 |
 Gytis Ivaskevicius | * Could you elaborate on that? Here are few questions that come up:
- How should it be shaped? Is github issue fine or should it be RFC?
- Does it need to include POC?
| 08:24:08 |
 Jonas Chevalier | I'm still not sure what problem it's solving. I get that the previous GCC version is leaking, but why is it a problem. | 08:28:15 |
| Jonas Chevalier | I know I'm being annoying :) | 08:29:34 |
 raboof | RFC seems overkill in this stage, a GitHub issue might not be noticed - perhaps a discourse topic seem most suitable? I agree a clear motivation (or even better, an assessment of pro's and con's?) is important. | 08:30:19 |
| Gytis Ivaskevicius | In reply to @zimbatm:numtide.com
I'm still not sure what problem it's solving. I get that the previous GCC version is leaking, but why is it a problem. Motivation:
Avoid nixpkgs lieing, especially with glibc and to gain latest compiler optimizations. In addition to these two points - this will allow us to pretty much never update bootstrap tools.
Technically nothing major, just super not nice from my point of view
| 08:31:18 |
| Gytis Ivaskevicius | In reply to @raboof:matrix.org
RFC seems overkill in this stage, a GitHub issue might not be noticed - perhaps a discourse topic seem most suitable? I agree a clear motivation (or even better, an assessment of pro's and con's?) is important. Will do! | 08:31:41 |
 andi- | Wait, we didn't bump the bootstrap tools to GCC10 when we did the upgrade? Was it simply skipped or did we run into actual problems? | 09:37:23 |
| andi- | In any case didn't we have the bootstrapTarballs jobs to exactly get rid of the manual tasks with bumping them? We have to put the files on a bucket but that is about it, no? | 09:38:48 |
| Gytis Ivaskevicius | In reply to @andi:kack.it
Wait, we didn't bump the bootstrap tools to GCC10 when we did the upgrade? Was it simply skipped or did we run into actual problems? Currently bootstrap tools uses GCC8. It was not bumped because this is something that is not expected to be done due to NixOS trust issues <3 | 09:38:49 |
| andi- | We are using hydra to build new version of bootstrap tools it should be as trustworthy as any other hydra output | 09:39:21 |
| Gytis Ivaskevicius | https://github.com/NixOS/nixpkgs/issues/123467#issuecomment-843264782
Bootstrapping tools have inherent trust complications, so it makes sense not to update them too often.
And I more or less agree with him
| 09:40:34 |
| Gytis Ivaskevicius | * https://github.com/NixOS/nixpkgs/issues/123467#issuecomment-843264782
Bootstrapping tools have inherent trust complications, so it makes sense not to update them too often.
And I more or less agree with him
| 09:40:39 |
| andi- | I agree that we shouldn't bump them too often but given that we have a clear path how they were obtained and if that solves an issue, why not? | 09:44:04 |
| Gytis Ivaskevicius | it all depends on trust policy that our supreme Nix overlords wish to enforce. Both options make sense but If i'd had to choose - i'd rather never update bootstrap tools | 09:45:57 |
| andi- | For a trust perspective that is the ultimate goal. I already expressed that in the rustc bootstrap thread. I'd rather bootstrap via mrustc + 5 newer releases of rustc than always updating the binaries that we pull from "random places". It is worse there as we aren't even building our own boostrap tools. | 09:47:29 |
| andi- | For the GCC world we at least routinely exercise creating the bootstrap files and can show a trail for how they came to be. | 09:47:54 |
| Gytis Ivaskevicius | In reply to @andi:kack.it
For a trust perspective that is the ultimate goal. I already expressed that in the rustc bootstrap thread. I'd rather bootstrap via mrustc + 5 newer releases of rustc than always updating the binaries that we pull from "random places". It is worse there as we aren't even building our own boostrap tools. Oh, i noticed this too. Havent checked for the thread tho. Imma go search for it :D | 09:48:17 |
| andi- | I think it is dead now.. Last activity about a year or so ago. | 09:48:34 |
| Gytis Ivaskevicius | Good old "it is still important" comment will change the activity date :D | 09:49:05 |
| Gytis Ivaskevicius | Also about rustc - i heard that there are plans to get gcc to compile rust due to linux kernel - that would open few doors 👀 | 09:50:10 |
| Gytis Ivaskevicius | https://github.com/NixOS/nixpkgs/issues/72606 | 09:51:28 |
| Jonas Chevalier | same with Go. We used to bootstrap from 5 versions but switched back to binary bootrstrap | 09:51:29 |
| Gytis Ivaskevicius | what, why? | 09:53:20 |
| Gytis Ivaskevicius | I thought gcc can compile Go 🤔 | 09:53:32 |
| andi- | IIRC only until Go 1.6 | 09:55:02 |
| Gytis Ivaskevicius | ahh | 09:55:16 |
| andi- | and perhaps the bootstrap was getting too slow for some.. | 09:55:18 |
| andi- | and wasn't GccGO being removed in GCC9 or something? | 09:55:30 |
| Gytis Ivaskevicius | https://golang.org/doc/install/gccgo | 09:56:04 |
| andi- | ah nice so we got at last Go 1.12.1 | 10:00:12 |
