build: enable libstdc++ assertions

Benchmarks say that it does not regress performance by more than 1%
(which is where it gets really hard to measure accurately anyhow).

Meson appears to be planning to do this for us without asking us in a
release we will get in the future, and it seems good enough to ship
today:
https://mesonbuild.com/Release-notes-for-1-4-0.html#ndebug-setting-now-controls-c-stdlib-assertions

Benchmarks:

| Command | Mean [ms] | Min [ms] | Max [ms] | Relative |
|:---|---:|---:|---:|---:|
| `result-asserts/bin/nix --extra-experimental-features 'nix-command flakes' eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix` | 418.4 ± 25.0 | 396.9 | 451.2 | 1.01 ± 0.08 |
| `result/bin/nix --extra-experimental-features 'nix-command flakes' eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix` | 416.1 ± 23.9 | 397.1 | 445.4 | 1.00 |

| Command | Mean [s] | Min [s] | Max [s] | Relative |
|:---|---:|---:|---:|---:|
| `GC_INITIAL_HEAP_SIZE=10g result-asserts/bin/nix eval --extra-experimental-features 'nix-command flakes' --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 4.147 ± 0.021 | 4.123 | 4.195 | 1.00 |
| `GC_INITIAL_HEAP_SIZE=10g result/bin/nix eval --extra-experimental-features 'nix-command flakes' --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 4.149 ± 0.027 | 4.126 | 4.215 | 1.00 ± 0.01 |

| Command | Mean [s] | Min [s] | Max [s] | Relative |
|:---|---:|---:|---:|---:|
| `result-asserts/bin/nix --extra-experimental-features 'nix-command flakes' eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 5.838 ± 0.023 | 5.799 | 5.867 | 1.01 ± 0.01 |
| `result/bin/nix --extra-experimental-features 'nix-command flakes' eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 5.788 ± 0.044 | 5.715 | 5.876 | 1.00 |

| Command | Mean [s] | Min [s] | Max [s] | Relative |
|:---|---:|---:|---:|---:|
| `result-asserts/bin/nix --extra-experimental-features 'nix-command flakes' search --no-eval-cache github:nixos/nixpkgs/e1fa12d4f6c6fe19ccb59cac54b5b3f25e160870 hello` | 15.993 ± 0.081 | 15.829 | 16.096 | 1.01 ± 0.01 |
| `result/bin/nix --extra-experimental-features 'nix-command flakes' search --no-eval-cache github:nixos/nixpkgs/e1fa12d4f6c6fe19ccb59cac54b5b3f25e160870 hello` | 15.897 ± 0.075 | 15.807 | 16.047 | 1.00 |

Fixes: lix-project/lix#4

Change-Id: Id3a6f38274ba94d5d10b09edd19dfd96bc3e7d5f
This commit is contained in:
jade 2024-04-07 17:19:37 -07:00
parent 06f17a5c78
commit 0d37b99a15
3 changed files with 31 additions and 3 deletions

View file

@ -0,0 +1,23 @@
---
synopsis: Lix turns more internal bugs into crashes
cls: [797, 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][apple-crashreport] for more details.
[apple-crashreport]: https://developer.apple.com/documentation/xcode/acquiring-crash-reports-and-diagnostic-logs#Locate-crash-reports-and-memory-logs-on-the-device

View file

@ -4,7 +4,7 @@
# This has an overhead of 0-4% on gcc and unmeasurably little on clang, in
# Nix evaluation benchmarks.
DEFAULT_SANITIZE_FLAGS = -fsanitize=signed-integer-overflow -fsanitize-undefined-trap-on-error
GLOBAL_CXXFLAGS += -Wno-deprecated-declarations -Werror=switch $(DEFAULT_SANITIZE_FLAGS)
GLOBAL_CXXFLAGS += -Wno-deprecated-declarations -Werror=switch $(DEFAULT_SANITIZE_FLAGS) -D_GLIBCXX_ASSERTIONS=1
GLOBAL_LDFLAGS += $(DEFAULT_SANITIZE_FLAGS)
# Allow switch-enum to be overridden for files that do not support it, usually because of dependency headers.
ERROR_SWITCH_ENUM = -Werror=switch-enum

View file

@ -344,6 +344,12 @@ add_project_arguments(
'-Wimplicit-fallthrough',
'-Werror=switch',
'-Werror=switch-enum',
# Enable assertions in libstdc++ by default. Harmless on libc++. Benchmarked
# at ~1% overhead in `nix search`.
#
# FIXME: remove when we get meson 1.4.0 which will default this to on for us:
# https://mesonbuild.com/Release-notes-for-1-4-0.html#ndebug-setting-now-controls-c-stdlib-assertions
'-D_GLIBCXX_ASSERTIONS=1',
language : 'cpp',
)
@ -351,8 +357,7 @@ if cxx.get_id() in ['gcc', 'clang']
# 2024-03-24: jade benchmarked the default sanitize reporting in clang and got
# a regression of about 10% on hackage-packages.nix with clang. So we are trapping instead.
#
# This has an overhead of 0-4% on gcc and unmeasurably little on clang, in
# Nix evaluation benchmarks.
# This has an unmeasurably low overhead in Nix evaluation benchmarks.
#
# N.B. Meson generates a completely nonsense warning here:
# https://github.com/mesonbuild/meson/issues/9822