Nixpkgs Stdenv - Public Room Timeline

	Nixpkgs Stdenv	222 Members
		71 Servers

Load older messages

Sender	Message	Time
7 Jan 2023
vcunat	I don't know. If it's significantly less controversial, we could just upgrade gcc version in aarch64-linux bootstrap tools (say, to 11 or 12). To get around the worst effects for now, probably even for years. (Though I'm not sure about practical availability of the very few people with permissions to upload tarballs; IIRC sometimes it took weeks or more.) There's quite a general dilemma, too. Having shorter and simpler bootstrap process makes it easier to deal with any stdenv-changing stuff. So we could rotate bootstrap tools faster instead, e.g. yearly - at least for the two main Linux platforms. Some people would instead prefer to have much smaller and basically fixed bootstrap tools (GuixSD does that IIRC).	10:26:49
vcunat	* I don't know. If it's significantly less controversial, we could just upgrade gcc version in aarch64-linux bootstrap tools (say, to 11 or 12). To get around the worst effects for now, probably even for years. (Though I'm not sure about practical availability of the very few people with permissions to upload tarballs; IIRC sometimes it took weeks or more.) There's quite a general dilemma, too. Having shorter and simpler bootstrap process makes it easier to deal with any stdenv-changing stuff. So we could rotate bootstrap tools faster instead, e.g. yearly - at least for the two main Linux platforms. Some people would instead prefer to have much smaller and basically fixed bootstrap tools (GuixSD does that IIRC), even though that makes bootstrapping harder.	10:27:55
@trofi:matrix.org	Heh, Yeah. Looking at a change history of gcc/libgcc it looks like unwinder and new primitives get added all the time: https://gcc.gnu.org/git/?p=gcc.git;a=history;f=libgcc;hb=HEAD. It depends so much on the rest of the ecosystem: new CPUs, new C types (like float decimals, int128/256), atomics against them. Could be lucky not to break for years, or could get new symbols in veery single release.	11:06:37
@trofi:matrix.org	WRT bootstrap defined as a sequence of stdenv/pkgs instantiations it feels like recursive definition does not encourage you to keep it lean. If somebody remembers to check if bootstrap got 10 new packages due to fresh automake in coreutils then lucky them. Otherwise it silently gets expanded with a bunch of seemingly auxiliary things.	11:09:07
8 Jan 2023
@trofi:matrix.org	* Heh, Yeah. Looking at a change history of gcc/libgcc it looks like unwinder and new primitives get added all the time: https://gcc.gnu.org/git/?p=gcc.git;a=history;f=libgcc;hb=HEAD. It depends so much on the rest of the ecosystem: new CPUs, new C types (like float decimals, int128/256), atomics against them. Could be lucky not to break for years, or could get new symbols in every single release.	11:50:33
@trofi:matrix.org	Also drew a few pictures and arrows around bootstrap for general amusement if you are into that kind of thing: http://trofi.github.io/posts/275-nixpkgs-bootstrap-deep-dive.html	12:40:35
@andreas.schraegle:helsinki-systems.de	In reply to @trofi:matrix.org Also drew a few pictures and arrows around bootstrap for general amusement if you are into that kind of thing: http://trofi.github.io/posts/275-nixpkgs-bootstrap-deep-dive.html just started reading, but the fact that libgcc_s.so comes from the bootstrap tarball was reported as an issue a long time ago https://github.com/NixOS/nixpkgs/issues/36947	12:48:07
@trofi:matrix.org	Yeah. I'm a nixpkgs noob. Was way before i popped up around :)	12:51:01
@andreas.schraegle:helsinki-systems.de	Don't worry, I've just been tangentially aware of a lot of these issues but never had the time&energy to properly try to fix them. Thanks for your work on this stuff.	13:08:37
9 Jan 2023
Artturin	https://github.com/NixOS/nixpkgs/pull/209816	04:07:46
Artturin	John Ericson: a retry at your https://github.com/NixOS/nixpkgs/pull/46059	04:08:21
Artturin	Download image.png	04:12:04
Artturin	https://github.com/NixOS/nixpkgs/pull/45403/files#r211386108	04:12:05
Artturin	volth was here :(	04:12:09
Artturin	thanks github	04:12:11
Artturin	* volth (probably) was here :(	04:12:24
Artturin	* volth was here :(	04:13:22
	@adis:blad.is left the room.	06:26:45
15 Jan 2023
Artturin	master `$ time \nix build ".#stdenv" real 26:00.89 user 2.416 sys 1.399 $ nix path-info --derivation -r ".#stdenv" \| wc -l 260` staging `$ time \nix build ".#stdenv" real 21:30.37 user 2.045 sys 1.177 $ nix path-info --derivation -r ".#stdenv" \| wc -l 245` trofi after your and amjosephs work	02:45:45
Winter (she/her)	In reply to @artturin:matrix.org master `$ time \nix build ".#stdenv" real 26:00.89 user 2.416 sys 1.399 $ nix path-info --derivation -r ".#stdenv" \| wc -l 260` staging `$ time \nix build ".#stdenv" real 21:30.37 user 2.045 sys 1.177 $ nix path-info --derivation -r ".#stdenv" \| wc -l 245` trofi after your and amjosephs work was this mainly the libtool stuff?	02:57:17
Artturin	In reply to @winterqt:nixos.dev was this mainly the libtool stuff? nah	03:36:41
Artturin	https://github.com/NixOS/nixpkgs/pull/209870	03:36:44
Artturin	check incorporates section	03:36:47
Artturin	stdenv: build gettext only once #210109	03:36:55
Artturin	binutils: avoid texinfo dependency during bootstrap #210132	03:37:00
Artturin	libtool: drop unused auto* dependencies #210118	03:37:07
Winter (she/her)	whoa	03:39:54
Winter (she/her)	that's very cool	03:40:20
Artturin	yea	03:40:30
@trofi:matrix.org	In reply to @artturin:matrix.org nah It's a bit of everything. There is also recent `lzip` cleanup: https://github.com/NixOS/nixpkgs/pull/210727. And ideally we could lean `bzip2` as well at some point: https://sourceware.org/PR30001. I think it's a last stand for autoconf. The caveat is that I'd like to add an extra `gcc` build instead: https://github.com/NixOS/nixpkgs/pull/209063 :)	07:23:00

Show newer messages

Back to Room ListRoom Version: 9