diff options
| author | Michael Forney <mforney@mforney.org> | 2019-11-17 14:48:05 -0800 |
|---|---|---|
| committer | Michael Forney <mforney@mforney.org> | 2019-11-17 15:24:25 -0800 |
| commit | f4720910b9b72bd1fe2aaa82cf6a38e226d5e149 (patch) | |
| tree | 773f1c9ea2fa5375243a4cb065c8345f0730a5cf /scripts | |
| parent | 564557663b9c36612df603e36e4c7a679e82c51b (diff) | |
Update ninja requirement to 1.8.0 and don't bother with atomic outputs
Since ninja 1.8.0[0] and samurai 0.2[1], output mtime is always
recorded after a job succeeds, and outputs are considered dirty if
the recorded mtime is older than any input. This means that even
if a command partially writes its output then fails (for example,
if run with stdout redirected to a file), it will still get re-run
during the next build.
This means that outputs no longer have to be written atomically
(i.e. renamed into place) in order to ensure a correct build, and
we can drop the idiom `command >$out.tmp && mv $out.tmp $out`.
[0] https://github.com/ninja-build/ninja/commit/04d886b11041bb59d01df794cce7a1e8cad2250d
[1] https://github.com/michaelforney/samurai/commit/799bc5a4719b90a1026c129b55b92c0bccf97eeb
Diffstat (limited to 'scripts')
| -rw-r--r-- | scripts/commit.sh | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/scripts/commit.sh b/scripts/commit.sh index d2d9a844..b2798b3b 100644 --- a/scripts/commit.sh +++ b/scripts/commit.sh @@ -21,4 +21,4 @@ if [ -z "${commit+set}" ] ; then commit=$(git -C "$repo" commit-tree -m "oasis $(git rev-parse --short=10 HEAD)" "$@" "$tag") git -C "$repo" update-ref "refs/heads/$branch" "$commit" fi -echo "$commit" >"$out.tmp" && mv "$out.tmp" "$out" +echo "$commit" >"$out" |