Regression: Quadlet: conflict names is nondeterministically failing

[edsantiago]

Has something changed in the processing of $QUADLET_UNIT_DIRS?

not ok 215 |252| quadlet conflict names in 2083ms
# #/vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv
# #|     FAIL: quadlet should show Notify=yes
# #| expected: =~ Notify=yes
# #|   actual: .... Notify=no

First seen today.

Reproducer:

$ while :;do hack/bats --rootless 252:conflict || break;done
...
fails on my laptop within 3-4 iterations. [EDIT: Usually. Right now, of course, on my final test, it took about 20 iterations]

[sys] |252| quadlet conflict names

• debian-13 : sys podman debian-13 rootless host sqlite
  • PR #23275
  • 09-23 13:16
• fedora-40 : sys podman fedora-40 rootless host sqlite
  • PR #24041
  • 09-23 10:28

x	x	x	x	x	x
sys(2)	podman(2)	debian-13(1)	rootless(2)	host(2)	sqlite(2)
		fedora-40(1)

For extra credit, a patch similar to this one might be appreciated by future maintainers looking at the test failure.

diff --git a/test/system/252-quadlet.bats b/test/system/252-quadlet.bats
index 3ccf97d269..ada1198106 100644
--- a/test/system/252-quadlet.bats
+++ b/test/system/252-quadlet.bats
@@ -230,13 +230,13 @@ EOF

     cat > $dir1/$quadlet_file <<EOF
 [Container]
-Image=$IMAGE
+Image=quay.io/libpod/this-is-the-one:wewant
 Notify=yes
 EOF

     cat > $dir2/$quadlet_file <<EOF
 [Container]
-Image=$IMAGE
+Image=quay.io/libpod/bad-bad-bad:nonono
 Notify=no
 EOF
     QUADLET_UNIT_DIRS="$dir1:$dir2" run \

[edsantiago]

OBTW it's not an alphanumeric-sorting issue: I tried dir1=aaa, dir2=bbb and vice-versa, and it still flakes unpredictably.

[vrothberg]

@ygalblum, ideas? I didn't find the time to follow all recent Quadlet improvements.

[giuseppe]

commit 133ea31ffb9e6cd9ccbbc01053463f04c11f7f18 introduced the regression. Moving from []string to map[string]struct{} broke the assumption that the entries are ordered

[ygalblum]

@giuseppe is correct, thanks. I think we need to keep the order assumption because it provides predictability as to which file is processed and which is skipped. So, I'll need to rework on that change.

[giuseppe]

the previous code was not deterministic for sub dirs. I think we need to sort the dir entries too

[ygalblum]

Are you sure? Because, the previous code used WalkDir which states that The files are walked in lexical order, which makes the output deterministic and each new directory was appended to the end of the list.

[giuseppe]

sorry you are right. WalkDir is fine