Lix 2.90 "Vanilla Ice Cream" (2024-07-10)

Lix 2.90.0 (2024-07-10)

Breaking Changes

Deprecate the online flake registries and vendor the default registry fj#183 fj#110 fj#116 #8953 #9087 cl/1127

The online flake registry https://channels.nixos.org/flake-registry.json is not pinned in any way, and the targets of the indirections can both update or change entirely at any point. Furthermore, it is refetched on every use of a flake reference, even if there is a local flake reference, and even if you are offline (which breaks).

For now, we deprecate the (any) online flake registry, and vendor a copy of the current online flake registry. This makes it work offline, and ensures that it won't change in the future.

Many thanks to julia for this. - Enforce syscall filtering and no-new-privileges on Linux cl/1063

In order to improve consistency of the build environment, system call filtering and no-new-privileges are now unconditionally enabled on Linux. The filter-syscalls and allow-new-privileges options which could be used to disable these features under some circumstances have been removed.

In order to support building on architectures without libseccomp support, the option to disable syscall filtering at build time remains. However, other uses of this option are heavily discouraged, since it would reduce the security of the sandbox substantially.

Many thanks to alois31 for this. - Overhaul nix flake update and nix flake lock UX #8817

The interface for creating and updating lock files has been overhauled:

nix flake lock only creates lock files and adds missing inputs now. It will never update existing inputs.
nix flake update does the same, but will update inputs.
Passing no arguments will update all inputs of the current flake, just like it already did.
Passing input names as arguments will ensure only those are updated. This replaces the functionality of nix flake lock --update-input
To operate on a flake outside the current directory, you must now pass --flake path/to/flake.
The flake-specific flags --recreate-lock-file and --update-input have been removed from all commands operating on installables. They are superceded by nix flake update.

Many thanks to iFreilicht, Lunaphied, and Théophane Hufschmitt for this. - nix profile now allows referring to elements by human-readable name, and no longer accepts indices #8678 cl/978 cl/980

nix profile now uses names to refer to installed packages when running list, remove or upgrade as opposed to indices. Indices have been removed. Profile element names are generated when a package is installed and remain the same until the package is removed.

Warning: The manifest.nix file used to record the contents of profiles has changed. Lix will automatically upgrade profiles to the new version when you modify the profile. After that, the profile can no longer be used by older versions of Lix.

Many thanks to iFreilicht, Qyriad, and Eelco Dolstra for this. - builtins.nixVersion and builtins.langVersion return fixed values cl/558 cl/1144

builtins.nixVersion now returns a fixed value "2.18.3-lix".

builtins.langVersion returns a fixed value 6, matching CppNix 2.18.

This prevents feature detection assuming that features that exist in Nix post-Lix-branch-off might exist, even though the Lix version is greater than the Nix version.

In the future, check for builtins for feature detection. If a feature cannot be detected by those means, please file a Lix bug.

Many thanks to jade for this. - Rename all the libraries nixexpr, nixstore, etc to lixexpr, lixstore, etc

The Lix C++ API libraries have had the following changes: - Includes moved from include/nix/ to include/lix/ - pkg-config files renamed from nix-expr to lix-expr and so on. - Libraries renamed from libnixexpr.so to liblixexpr.so and so on.

There are other changes between Nix 2.18 and Lix, since these APIs are not stable. However, this change in particular is a deliberate compatibility break to force downstreams linking to Lix to specifically handle Lix and avoid Lix accidentally getting ensnared in compatibility code for newer CppNix.

Migration path:

expr.hh -> lix/libexpr/expr.hh
nix/config.h -> lix/config.h

To apply this migration automatically, remove all <nix/> from includes, so #include <nix/expr.hh> -> #include <expr.hh>. Then, the correct paths will be resolved from the tangled mess, and the clang-tidy automated fix will work.

Then run the following for out of tree projects (header filter is set to only fix instances in headers in ../src relative to the compiler's working directory, as would be the case in nix-eval-jobs or other things built with meson, e.g.):

lix_root=$HOME/lix
(cd $lix_root/clang-tidy && nix develop -c 'meson setup build && ninja -C build')
run-clang-tidy -checks='-*,lix-fixincludes' -load=$lix_root/clang-tidy/build/liblix-clang-tidy.so -p build/ -header-filter '../src/.*.h' -fix src

Many thanks to jade for this.

Features

Experimental REPL support for documentation comments using :doc cl/564

Using :doc in the REPL now supports showing documentation comments when defined on a function.

Previously this was only able to document builtins, however it now will show comments defined on a lambda as well.

This support is experimental and relies on an embedded version of nix-doc.

The logic also supports limited Markdown formatting of doccomments and should easily support any RFC 145 compatible documentation comments in addition to simple commented documentation.

Many thanks to Lunaphied and jade for this. - Add repl-overlays option #10203 cl/504

A repl-overlays option has been added, which specifies files that can overlay and modify the top-level bindings in nix repl. For example, with the following contents in ~/.config/nix/repl.nix:

info: final: prev: let
  optionalAttrs = predicate: attrs:
    if predicate
    then attrs
    else {};
in
  optionalAttrs (prev ? legacyPackages && prev.legacyPackages ? ${info.currentSystem})
  {
    pkgs = prev.legacyPackages.${info.currentSystem};
  }

We can run nix repl and use pkgs to refer to legacyPackages.${currentSystem}:

$ nix repl --repl-overlays ~/.config/nix/repl.nix nixpkgs
Lix 2.90.0
Type :? for help.
Loading installable 'flake:nixpkgs#'...
Added 5 variables.
Loading 'repl-overlays'...
Added 6 variables.
nix-repl> pkgs.bash
«derivation /nix/store/g08b5vkwwh0j8ic9rkmd8mpj878rk62z-bash-5.2p26.drv»

Many thanks to wiggles for this. - Add a builtin addDrvOutputDependencies #7910 #9216

This builtin allows taking a drvPath-like string and turning it into a string with context such that, when it lands in a derivation, it will create dependencies on all the outputs in its closure (!). Although drvPath does this today, this builtin starts forming a path to migrate to making drvPath have a more normal and less surprising string context behaviour (see linked issue and PR for more details).

Many thanks to John Ericson and eldritch horrors for this. - Enter the --debugger when builtins.trace is called if debugger-on-trace is set #9914

If the debugger-on-trace option is set and --debugger is given, builtins.trace calls will behave similarly to builtins.break and will enter the debug REPL. This is useful for determining where warnings are being emitted from.

Many thanks to wiggles for this. - Add an option enable-core-dumps that enables core dumps from builds cl/1088

In the past, Lix disabled core dumps by setting the soft RLIMIT_CORE to 0 unconditionally. Although this rlimit could be altered from the builder since it is just the soft limit, this was kind of annoying to do. By passing --option enable-core-dumps true to an offending build, one can now cause the core dumps to be handled by the system in the normal way (winding up in coredumpctl, say, on Linux).

Many thanks to julia for this. - Add new eval-system setting #4093

Add a new eval-system option. Unlike system, it just overrides the value of builtins.currentSystem. This is more useful than overriding system, because you can build these derivations on remote builders which can work on the given system. In contrast, system also effects scheduling which will cause Lix to build those derivations locally even if that doesn't make sense.

eval-system only takes effect if it is non-empty. If empty (the default) system is used as before, so there is no breakage.

Many thanks to matthewbauer and eldritch horrors for this. - add --store-path argument to nix upgrade-nix, to manually specify the Nix to upgrade to cl/953

nix upgrade-nix by default downloads a manifest to find the new Nix version to upgrade to, but now you can specify --store-path to upgrade Nix to an arbitrary version from the Nix store.

Many thanks to Qyriad for this.

Improvements

nix flake check logs the checks #8882 #8893 cl/259 cl/260 cl/261 cl/262

nix flake check now logs the checks it runs and the derivations it evaluates:

$ nix flake check -v
evaluating flake...
checking flake output 'checks'...
checking derivation 'checks.aarch64-darwin.ghciwatch-tests'...
derivation evaluated to /nix/store/nh7dlvsrhds4cxl91mvgj4h5cbq6skmq-ghciwatch-test-0.3.0.drv
checking derivation 'checks.aarch64-darwin.ghciwatch-clippy'...
derivation evaluated to /nix/store/9cb5a6wmp6kf6hidqw9wphidvb8bshym-ghciwatch-clippy-0.3.0.drv
checking derivation 'checks.aarch64-darwin.ghciwatch-doc'...
derivation evaluated to /nix/store/8brdd3jbawfszpbs7vdpsrhy80as1il8-ghciwatch-doc-0.3.0.drv
checking derivation 'checks.aarch64-darwin.ghciwatch-fmt'...
derivation evaluated to /nix/store/wjhs0l1njl5pyji53xlmfjrlya0wmz8p-ghciwatch-fmt-0.3.0.drv
checking derivation 'checks.aarch64-darwin.ghciwatch-audit'...
derivation evaluated to /nix/store/z0mps8dyj2ds7c0fn0819y5h5611033z-ghciwatch-audit-0.3.0.drv
checking flake output 'packages'...
checking derivation 'packages.aarch64-darwin.default'...
derivation evaluated to /nix/store/41abbdyglw5x9vcsvd89xan3ydjf8d7r-ghciwatch-0.3.0.drv
checking flake output 'apps'...
checking flake output 'devShells'...
checking derivation 'devShells.aarch64-darwin.default'...
derivation evaluated to /nix/store/bc935gz7dylzmcpdb5cczr8gngv8pmdb-nix-shell.drv
running 5 flake checks...
warning: The check omitted these incompatible systems: aarch64-linux, x86_64-darwin, x86_64-linux
Use '--all-systems' to check all.

Many thanks to wiggles, Raito Bezarius, and eldritch horrors for this. - Add an option always-allow-substitutes to ignore allowSubstitutes in derivations #8047

You can set this setting to force a system to always allow substituting even trivial derivations like pkgs.writeText. This is useful for nix-fast-build --skip-cached and similar to be able to also ignore trivial derivations.

Many thanks to lovesegfault and eldritch horrors for this. - Concise error printing in nix repl #9928 cl/811

Previously, if an element of a list or attribute set threw an error while evaluating, nix repl would print the entire error (including source location information) inline. This output was clumsy and difficult to parse:

nix-repl> { err = builtins.throw "uh oh!"; }
{ err = «error:
       … while calling the 'throw' builtin
         at «string»:1:9:
            1| { err = builtins.throw "uh oh!"; }
             |         ^

       error: uh oh!»; }

Now, only the error message is displayed, making the output much more readable.

nix-repl> { err = builtins.throw "uh oh!"; }
{ err = «error: uh oh!»; }

However, if the whole expression being evaluated throws an error, source locations and (if applicable) a stack trace are printed, just like you'd expect:

nix-repl> builtins.throw "uh oh!"
error:
       … while calling the 'throw' builtin
         at «string»:1:1:
            1| builtins.throw "uh oh!"
             | ^

       error: uh oh!

Many thanks to wiggles for this. - Show all FOD errors with nix build --keep-going cl/1108

nix build --keep-going now behaves consistently with nix-build --keep-going. This means that if e.g. multiple FODs fail to build, all hash mismatches are displayed.

Many thanks to ma27 for this. - Duplicate attribute reports are more accurate cl/557

Duplicate attribute errors are now more accurate, showing the path at which an error was detected rather than the full, possibly longer, path that caused the error. Error reports are now

$ nix eval --expr '{ a.b = 1; a.b.c.d = 1; }'
error: attribute 'a.b' already defined at «string»:1:3
       at «string»:1:12:
            1| { a.b = 1; a.b.c.d = 1;
             |            ^

instead of

$ nix eval --expr '{ a.b = 1; a.b.c.d = 1; }'
error: attribute 'a.b.c.d' already defined at «string»:1:3
       at «string»:1:12:
            1| { a.b = 1; a.b.c.d = 1;
             |            ^

Many thanks to eldritch horrors for this. - Reduce eval memory usage and wall time #9658 cl/207

Reduce the size of the Env struct used in the evaluator by a pointer, or 8 bytes on most modern machines. This reduces memory usage during eval by around 2% and wall time by around 3%.

Many thanks to eldritch horrors for this. - Warn on unknown settings anywhere in the command line #10701

All nix commands will now properly warn when an unknown option is specified anywhere in the command line.

Before:

$ nix-instantiate --option foobar baz --expr '{}'
warning: unknown setting 'foobar'
$ nix-instantiate '{}' --option foobar baz --expr
$ nix eval --expr '{}' --option foobar baz
{ }

After:

$ nix-instantiate --option foobar baz --expr '{}'
warning: unknown setting 'foobar'
$ nix-instantiate '{}' --option foobar baz --expr
warning: unknown setting 'foobar'
$ nix eval --expr '{}' --option foobar baz
warning: unknown setting 'foobar'
{ }

Many thanks to Cole Helbling for this. - Nested debuggers are no longer supported #9920

Previously, evaluating an expression that throws an error in the debugger would enter a second, nested debugger:

nix-repl> builtins.throw "what"
error: what


Starting REPL to allow you to inspect the current state of the evaluator.

Welcome to Nix 2.18.1. Type :? for help.

nix-repl>

Now, it just prints the error message like nix repl:

nix-repl> builtins.throw "what"
error:
       … while calling the 'throw' builtin
         at «string»:1:1:
            1| builtins.throw "what"
             | ^

       error: what

Many thanks to wiggles for this. - Find GC roots using libproc on Darwin cl/723

Previously, the garbage collector found runtime roots on Darwin by shelling out to lsof -n -w -F n then parsing the result. The version of lsof packaged in Nixpkgs is very slow on Darwin, so Lix now uses libproc directly to speed up GC root discovery, in some tests taking 250ms now instead of 40s.

Many thanks to Artemis Tosini for this. - Increase default stack size on macOS #9860

Increase the default stack size on macOS to the same value as on Linux, subject to system restrictions to maximum stack size. This should reduce the number of stack overflow crashes on macOS when evaluating Nix code with deep call stacks.

Many thanks to wiggles for this. - Show more log context for failed builds #9670

Show 25 lines of log tail instead of 10 for failed builds. This increases the chances of having useful information in the shown logs.

Many thanks to DavHau for this. - rename 'nix show-config' to 'nix config show' #7672 #9477 cl/993

nix show-config was renamed to nix config show to be more consistent with the rest of the command-line interface.

Running nix show-config will now print a deprecation warning saying to use nix config show instead.

Many thanks to Théophane Hufschmitt and ma27 for this. - Print derivation paths in nix eval cl/446

nix eval previously printed derivations as attribute sets, so commands that print derivations (e.g. nix eval nixpkgs#bash) would infinitely loop and segfault. It now prints the .drv path the derivation generates instead.

Many thanks to wiggles for this. - Add an option --unpack to unpack archives in nix store prefetch-file #9805 cl/224

It is now possible to fetch an archive then NAR-hash it (as in, hash it in the same manner as builtins.fetchTarball or fixed-output derivations with recursive hash type) in one command.

Example:

~ » nix store prefetch-file --name source --unpack https://git.lix.systems/lix-project/lix/archive/2.90-beta.1.tar.gz
Downloaded 'https://git.lix.systems/lix-project/lix/archive/2.90-beta.1.tar.gz' to '/nix/store/yvfqnq52ryjc3janw02ziv7kr6gd0cs1-source' (hash 'sha256-REWlo2RYHfJkxnmZTEJu3Cd/2VM+wjjpPy7Xi4BdDTQ=').

Many thanks to yshui and eldritch horrors for this. - REPL printing improvements #9931 #10208 cl/375 cl/492

The REPL printer has been improved to do the following: - If a string is passed to :print, it is printed literally to the screen - Structures will be printed as multiple lines when necessary

Before:

nix-repl> { attrs = { a = { b = { c = { }; }; }; }; list = [ 1 ]; list' = [ 1 2 3 ]; }
{ attrs = { ... }; list = [ ... ]; list' = [ ... ]; }

nix-repl> :p { attrs = { a = { b = { c = { }; }; }; }; list = [ 1 ]; list' = [ 1 2 3 ]; }
{ attrs = { a = { b = { c = { }; }; }; }; list = [ 1 ]; list' = [ 1 2 3 ]; }

nix-repl> :p "meow"
"meow"

After:

nix-repl> { attrs = { a = { b = { c = { }; }; }; }; list = [ 1 ]; list' = [ 1 2 3 ]; }
{
  attrs = { ... };
  list = [ ... ];
  list' = [ ... ];
}

nix-repl> :p { attrs = { a = { b = { c = { }; }; }; }; list = [ 1 ]; list' = [ 1 2 3 ]; }
{
  attrs = {
    a = {
      b = {
        c = { };
      };
    };
  };
  list = [ 1 ];
  list' = [
    1
    2
    3
  ];
}

nix-repl> :p "meow"
meow

Many thanks to wiggles and eldritch horrors for this. - Coercion errors include the failing value #561 #9754

The error: cannot coerce a <TYPE> to a string message now includes the value which caused the error.

Before:

       error: cannot coerce a set to a string

After:

       error: cannot coerce a set to a string: { aesSupport = «thunk»;
       avx2Support = «thunk»; avx512Support = «thunk»; avxSupport = «thunk»;
       canExecute = «thunk»; config = «thunk»; darwinArch = «thunk»; darwinMinVersion
       = «thunk»; darwinMinVersionVariable = «thunk»; darwinPlatform = «thunk»; «84
       attributes elided»}

Many thanks to wiggles and eldritch horrors for this. - New-cli flake commands that expect derivations now print the failing value and its type cl/1177

In errors like flake output attribute 'legacyPackages.x86_64-linux.lib' is not a derivation or path, the message now includes the failing value and type.

Before:

    error: flake output attribute 'nixosConfigurations.yuki.config' is not a derivation or path
````

After:

error: expected flake output attribute 'nixosConfigurations.yuki.config' to be a derivation or path but found a set: { appstream = «thunk»; assertions = «thunk»; boot = { bcache = «thunk»; binfmt = «thunk»; binfmtMiscRegistrations = «thunk»; blacklistedKernelModules = «thunk»; bootMount = «thunk»; bootspec = «thunk»; cleanTmpDir = «thunk»; consoleLogLevel = «thunk»; «43 attributes elided» }; «48 attributes elided» } ```

Many thanks to Qyriad for this. - Type errors include the failing value #561 #9753

In errors like value is an integer while a list was expected, the message now includes the failing value.

Before:

       error: value is a set while a string was expected

After:

       error: expected a string but found a set: { ghc810 = «thunk»;
       ghc8102Binary = «thunk»; ghc8107 = «thunk»; ghc8107Binary = «thunk»;
       ghc865Binary = «thunk»; ghc90 = «thunk»; ghc902 = «thunk»; ghc92 = «thunk»;
       ghc924Binary = «thunk»; ghc925 = «thunk»;  «17 attributes elided»}

Many thanks to wiggles and eldritch horrors for this. - Visual clutter in --debugger is reduced #9919

Before:

info: breakpoint reached


Starting REPL to allow you to inspect the current state of the evaluator.

Welcome to Nix 2.20.0pre20231222_dirty. Type :? for help.

nix-repl> :continue
error: uh oh


Starting REPL to allow you to inspect the current state of the evaluator.

Welcome to Nix 2.20.0pre20231222_dirty. Type :? for help.

nix-repl>

After:

info: breakpoint reached

Nix 2.20.0pre20231222_dirty debugger
Type :? for help.
nix-repl> :continue
error: uh oh

nix-repl>

Many thanks to wiggles and eldritch horrors for this. - REPL now supports CTRL+Z to suspend

Editline is now built with SIGTSTP support, so now typing CTRL+Z in the REPL will suspend the REPL and allow it to be resumed later or backgrounded.

Many thanks to Qyriad for this. - Allow single quotes in nix-shell shebangs #8470

Example:

#! /usr/bin/env nix-shell
#! nix-shell -i bash --packages 'terraform.withPlugins (plugins: [ plugins.openstack ])'

Many thanks to ncfavier and eldritch horrors for this. - reintroduce shortened -E form for --expr to new CLI cl/605

In the old CLI, it was possible to supply a shorter -E flag instead of fully specifying --expr every time you wished to provide an expression that would be evaluated to produce the given command's input. This was retained for the --file flag when the new CLI utilities were written with -f, but -E was dropped.

We now restore the -E short form for better UX. This is most useful for nix eval but most any command that takes an Installable argument should benefit from it as well.

Many thanks to Lunaphied for this. - Source locations are printed more consistently in errors #561 #9555

Source location information is now included in error messages more consistently. Given this code:

let
  attr = {foo = "bar";};
  key = {};
in
  attr.${key}

Previously, Nix would show this unhelpful message when attempting to evaluate it:

error:
       … while evaluating an attribute name

       error: value is a set while a string was expected

Now, the error message displays where the problematic value was found:

error:
       … while evaluating an attribute name

         at bad.nix:4:11:

            3|   key = {};
            4| in attr.${key}
             |           ^
            5|

       error: expected a string but found a set: { }

Many thanks to wiggles and eldritch horrors for this. - Some stack overflow segfaults are fixed #9616 #9617 cl/205

The number of nested function calls has been restricted, to detect and report infinite function call recursions. The default maximum call depth is 10,000 and can be set with the max-call-depth option.

This fixes segfaults or the following unhelpful error message in many cases:

  error: stack overflow (possible infinite recursion)

Before:

$ nix-instantiate --eval --expr '(x: x x) (x: x x)'
Segmentation fault: 11

After:

$ nix-instantiate --eval --expr '(x: x x) (x: x x)'
error: stack overflow

       at «string»:1:14:
            1| (x: x x) (x: x x)
             |              ^

Many thanks to wiggles and eldritch horrors for this. - Warn about ignored client settings cl/1026

Emit a warning for every client-provided setting the daemon ignores because the requesting client is not run by a trusted user. Previously this was only a debug message.

Many thanks to jade for this. - Better error reporting for with expressions #9658 cl/207

with expressions using non-attrset values to resolve variables are now reported with proper positions.

Previously an incorrect with expression would report no position at all, making it hard to determine where the error originated:

nix-repl> with 1; a
error:
       … <borked>

         at «none»:0: (source not available)

       error: value is an integer while a set was expected

Now position information is preserved and reported as with most other errors:

nix-repl> with 1; a
error:
       … while evaluating the first subexpression of a with expression
         at «string»:1:1:
            1| with 1; a
             | ^

       error: expected a set but found an integer: 1

Many thanks to eldritch horrors for this.

Fixes

Fix nested flake input follows #6621 cl/994

Previously nested-input overrides were ignored; that is, the following did not override anything, in spite of the nix3-flake manual documenting it working:

{
  inputs = {
    foo.url = "github:bar/foo";
    foo.inputs.bar.inputs.nixpkgs = "nixpkgs";
  };
}

This is useful to avoid the 1000 instances of nixpkgs problem without having each flake in the dependency tree to expose all of its transitive dependencies for modification.

Many thanks to Kha and ma27 for this. - Fix CVE-2024-27297 (GHSA-2ffj-w4mj-pg37) cl/266

Since Lix fixed-output derivations run in the host network namespace (which we wish to change in the future, see lix#285), they may open abstract-namespace Unix sockets to each other and to programs on the host. Lix contained a now-fixed time-of-check/time-of-use vulnerability where one derivation could send writable handles to files in their final location in the store to another over an abstract-namespace Unix socket, exit, then the other derivation could wait for Lix to hash the paths and overwrite them.

The impact of this vulnerability is that two malicious fixed-output derivations could create a poisoned path for the sources to Bash or similarly important software containing a backdoor, leading to local privilege execution.

CppNix advisory: https://github.com/NixOS/nix/security/advisories/GHSA-2ffj-w4mj-pg37

Many thanks to puck, jade, Théophane Hufschmitt, Tom Bereknyei, and Valentin Gagarin for this. - --debugger can now access bindings from let expressions #8827 #9918

Breakpoints and errors in the bindings of a let expression can now access those bindings in the debugger. Previously, only the body of let expressions could access those bindings.

Many thanks to wiggles for this. - Fix handling of truncated .drv files. #9673

Previously a .drv that was truncated in the middle of a string would case nix to enter an infinite loop, eventually exhausting all memory and crashing.

Many thanks to eldritch horrors for this. - The --debugger will start more reliably in let expressions and function calls #6649 #9917

Previously, if you attempted to evaluate this file with the debugger:

let
  a = builtins.trace "before inner break" (
    builtins.break "hello"
  );
  b = builtins.trace "before outer break" (
    builtins.break a
  );
in
  b

Lix would correctly enter the debugger at builtins.break a, but if you asked it to :continue, it would skip over the builtins.break "hello" expression entirely.

Now, Lix will correctly enter the debugger at both breakpoints.

Many thanks to wiggles and eldritch horrors for this. - Creating setuid/setgid binaries with fchmodat2 is now prohibited by the build sandbox #10501

The build sandbox blocks any attempt to create setuid/setgid binaries, but didn't check for the use of the fchmodat2 syscall which was introduced in Linux 6.6 and is used by glibc >=2.39. This is fixed now.

Many thanks to ma27 for this. - consistent order of lambda formals in printed expressions #9874

Always print lambda formals in lexicographic order rather than the internal, creation-time based symbol order. This makes printed formals independent of the context they appear in.

Many thanks to eldritch horrors for this. - fix duplicate attribute error positions for inherit #9874

When an inherit caused a duplicate attribute error, the position of the error was not reported correctly, placing the error with the inherit itself or at the start of the bindings block instead of the offending attribute name.

Many thanks to eldritch horrors for this. - inherit (x) ... evaluates x only once #9847

inherit (x) a b ... now evaluates the expression x only once for all inherited attributes rather than once for each inherited attribute. This does not usually have a measurable impact, but side-effects (such as builtins.trace) would be duplicated and expensive expressions (such as derivations) could cause a measurable slowdown.

Many thanks to eldritch horrors for this. - Store paths are allowed to start with . #912 #9867 #9091 #9095 #9120 #9121 #9122 #9130 #9219 #9224

Leading periods were allowed by accident in Nix 2.4. The Nix team has considered this to be a bug, but this behavior has since been relied on by users, leading to unnecessary difficulties. From now on, leading periods are officially, definitively supported. The names . and .. are disallowed, as well as those starting with .- or ..-.

Nix versions that denied leading periods are documented in the issue.

Many thanks to Robert Hensing and eldritch horrors for this. - Fix nix-env --query --drv-path --json #9257

Fixed a bug where nix-env --query ignored --drv-path when --json was set.

Many thanks to Artturin and eldritch horrors for this. - re-evaluate cached evaluation errors cl/771

"cached failure of [expr]" errors have been removed: expressions already in the eval cache as a failure will now simply be re-evaluated, removing the need to set --no-eval-cache or similar to see the error.

Many thanks to Qyriad for this. - Interrupting builds in the REPL works more than once cl/1097

Builds in the REPL can be interrupted by pressing Ctrl+C. Previously, this only worked once per REPL session; further attempts would be ignored. This issue is now fixed, so that builds can be canceled consistently.

Many thanks to alois31 for this. - In the debugger, while evaluating the attribute errors now include position information #9915

Before:

0: while evaluating the attribute 'python311.pythonForBuild.pkgs'
0x600001522598

After:

0: while evaluating the attribute 'python311.pythonForBuild.pkgs'
/nix/store/hg65h51xnp74ikahns9hyf3py5mlbbqq-source/overrides/default.nix:132:27

   131|
   132|       bootstrappingBase = pkgs.${self.python.pythonAttr}.pythonForBuild.pkgs;
      |                           ^
   133|     in

Many thanks to wiggles for this. - Include phase reporting in log file for ssh-ng builds #9280

Store phase information of remote builds run via ssh-ng remotes in the local log file, matching logging behavior of local builds.

Many thanks to r-vdp for this. - Fix ssh-ng:// remotes not respecting --substitute-on-destination #9600

nix copy ssh-ng:// now respects --substitute-on-destination, as does nix-copy-closure and other commands that operate on remote ssh-ng stores. Previously this was always set by builders-use-substitutes setting.

Many thanks to SharzyL for this. - using nix profile on /nix/var/nix/profiles/default no longer breaks nix upgrade-nix cl/952

On non-NixOS, Nix is conventionally installed into a nix-env style profile at /nix/var/nix/profiles/default. Like any nix-env profile, using nix profile on it automatically migrates it to a nix profile style profile, which is incompatible with nix-env. nix upgrade-nix previously relied solely on nix-env to do the upgrade, but now will work fine with either kind of profile.

Many thanks to Qyriad for this.

Packaging

Lix turns more internal bugs into crashes cl/797 cl/626

Lix now enables build options such as trapping on signed overflow and enabling libstdc++ assertions by default. These may find new bugs in Lix, which will present themselves as Lix processes aborting, potentially without an error message.

If Lix processes abort on your machine, this is a bug. Please file a bug, ideally with the core dump (or information from it).

On Linux, run coredumpctl list, find the crashed process's PID at the bottom of the list, then run coredumpctl info THE-PID. You can then paste the output into a bug report.

On macOS, open the Console app from Applications/Utilities, select Crash Reports, select the crash report in question. Right click on it, select Open In Finder, then include that file in your bug report. See the Apple documentation for more details.

Many thanks to jade for this. - Stop vendoring toml11 cl/675

We don't apply any patches to it, and vendoring it locks users into bugs (it hasn't been updated since its introduction in late 2021).

Many thanks to winter for this. - Lix is built with meson cl/580 cl/627 cl/628 cl/707 cl/711 cl/712 cl/719

Lix is built exclusively with the meson build system thanks to a huge team-wide effort, and the legacy make/autoconf based build system has been removed altogether. This improves maintainability of Lix, enables things like saving 20% of compile times with precompiled headers, and generally makes the build less able to produce obscure incremental compilation bugs.

Non-Nix-based downstream packaging needs rewriting accordingly.

Many thanks to Qyriad, eldritch horrors, jade, wiggles, and winter for this. - Upstart scripts removed cl/574

Upstart scripts have been removed from Lix, since Upstart is obsolete and has not been shipped by any major distributions for many years. If these are necessary to your use case, please back port them to your packaging.

Many thanks to jade for this.

Development

Clang build timing analysis cl/587

We now have Clang build profiling available, which generates Chrome tracing files for each compilation unit. To enable it, run meson configure build -Dprofile-build=enabled in a Clang stdenv (nix develop .#native-clangStdenvPackages) then rerun the compilation.

If you want to make the build go faster, do a clang build with meson, then run maintainers/buildtime_report.sh build, then contemplate how to improve the build time.

You can also look at individual object files' traces in https://ui.perfetto.dev.

See the wiki page for more details on how to do this.

Miscellany

Disallow empty search regex in nix search #9481

nix search now requires a search regex to be passed. To show all packages, use ^.

Many thanks to iFreilicht and eldritch horrors for this. - nix repl history is saved more reliably cl/1164

nix repl now saves its history file after each line, rather than at the end of the session; ensuring that it will remember what you typed even after it crashes.

Many thanks to puck for this.