Misc Meson progress

[Ericson2314]

Motivation

Meson-ify a few things, scripts, completions, etc. Should make our Meson build complete except for docs.

Context

depends on #11301

Priorities and Process

Add 👍 to pull requests you find important.

The Nix maintainer team uses a GitHub project board to schedule and track reviews.

[dpulls]

🎉 All dependencies have been resolved !

[Ericson2314]

@eli-schwartz Do you know what I might do about this macos failure? Seems like the issue is that the is_dir check in https://github.com/mesonbuild/meson/blob/1e2e30bbcfea876e05ce5b863579dc7eb71b5fb3/mesonbuild/interpreter/interpreter.py#L704-L715 is failing because of sandboxing.

[eli-schwartz]

python pathlib has an alarming tendency to raise nonsense exceptions when it should not. Maybe this is another one...

[eli-schwartz]

And here you call it localstatedir! :D

Which makes me wonder why you don't use meson's own --localstatedir option (https://www.gnu.org/prep/standards/standards.html#Directory-Variables, even) and just set that value to /nix/var, perhaps in default_options.

nix/src/libstore/meson.build

Lines 1 to 13 in 31f3f23

	project('nix-store', 'cpp',
	version : files('.version'),
	default_options : [
	'cpp_std=c++2a',
	# TODO(Qyriad): increase the warning level
	'warning_level=1',
	'debug=true',
	'optimization=2',
	'errorlogs=true', # Please print logs for tests that fail
	],
	meson_version : '>= 1.1',
	license : 'LGPL-2.1-or-later',
	)

[Ericson2314]

Oh hmm, I just took this from Lix's Meson, but maybe the Lix devs didn't realize one could override a built-in option default value. CC @Qyriad

[tomberek]

Perhaps:

diff --git a/src/libstore/meson.build b/src/libstore/meson.build
index d0cd7d3cb..b829de3f7 100644
--- a/src/libstore/meson.build
+++ b/src/libstore/meson.build
@@ -421,9 +421,9 @@ install_headers(headers, subdir : 'nix', preserve_path : true)
 
 libraries_private = []
 
-extra_pkg_config_variables = {
-  'localstatedir' : get_option('state-dir'),
-  'storedir' : get_option('store-dir'),
-}
+extra_pkg_config_variables = [
+  f'localstatedir=' + get_option('state-dir'),
+  f'storedir=' + get_option('store-dir'),
+]
 
 subdir('build-utils-meson/export')

We do the dict to string conversion ourselves and it doesn't look like a Path, thus bypassing that check.

[Ericson2314]

Trying @eli-schwartz's localstatedir=/nix/var suggestion.

[Ericson2314]

Still a problem. I minimized it and opened mesonbuild/meson#13584 Meson options are a red herring -- it's declared dependency variables instead.

[Ericson2314]

@eli-schwartz Well, I am open to alternatives! Here's basically what I am thinking:

1. The paths are a property of the libstore library, and configurable, and needs to be conveyed downstream.

2. We can't just put it in a header, because we need it at configure time downstream, patting localstatedir and storedir in the pkg-config file seems more appropriate.

3. When we are doing the combined subprojects build, we don't use a pkg-config dep, and instead use an internal dep. So we get the variables from the internal dep instead.

Where does it go off the rails to you, conceptually? Each step seems fine to me, even if the end result is a bit funky.

[eli-schwartz]

When you're doing the combined subprojects build, do different subprojects get built with different prefix/libdir/localstatedir?

[Ericson2314]

Fair question: Nope. Each project should either use the same localstatedir, or doesn't care about the variable at all.

[eli-schwartz]

So perhaps for the purpose of this PR we can dodge the question of what declare_dependency is doing, and simply have projects that want this value look it up via get_option() instead of subp_dep.get_variable()...

[Ericson2314]

I did something along that lines. I still need the pkg-config variable for storedir, but I now don't set the localstatedir one on macOS, and I have an assert that it lines up with get_option('localstatedir').

[Ericson2314]

python pathlib has an alarming tendency to raise nonsense exceptions when it should not. Maybe this is another one...

I think the (at least longer term) solution might be a Meson-side try-catch, as permission error is indeed different from knowing for sure whether something is a directory or not, but I'm not quite sure what the meson logic is doing there so I'm not sure I'd be patching it right.

[nixos-discourse]

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/2024-08-26-nix-team-meeting-minutes-172/51300/1

[cole-h]

This appears to produce a busted systemd unit.

$ cat result/lib/systemd/system/nix-daemon.service
[Unit]
Description=Nix Daemon
Documentation=man:nix-daemon https://nixos.org/manual
RequiresMountsFor=/nix/store
RequiresMountsFor=/nix/var
RequiresMountsFor=/nix/var/nix/db
ConditionPathIsReadWrite=/nix/var/nix/daemon-socket

[Service]
ExecStart=@share/nix-daemon nix-daemon --daemon
KillMode=process
LimitNOFILE=1048576
TasksMax=1048576

[Install]
WantedBy=multi-user.target

Notice how ExecStart is @share/nix-daemon, which is unlikely to exist on any machine.

Fixed in #11413.