tsort: GNU test cases fail even though our output is still valid

For the test cases named tree-1 and tree-2 in the GNU test file tests/misc/tsort.pl, our implementation of tsort produces an output different from GNU tsort but still valid. (In general, there may be more than one valid topological ordering for a given directed acyclic graph.) The test cases will fail due to this difference.

One option is to use our build-gnu.sh script to modify the GNU test cases as follows:

diff --git a/tests/misc/tsort.pl b/tests/misc/tsort.pl
index f1ca28a08..1fb256139 100755
--- a/tests/misc/tsort.pl
+++ b/tests/misc/tsort.pl
@@ -41,9 +41,9 @@ my @Tests =
     {OUT => "a\nb\nc\nd\ne\nf\ng\n"}],
 
    ['tree-1', {IN => "a b b c c d d e e f f g\nc x x y y z\n"},
-    {OUT => "a\nb\nc\nx\nd\ny\ne\nz\nf\ng\n"}],
+    {OUT => "a\nb\nc\nd\nx\ne\ny\nf\nz\ng\n"}],
    ['tree-2', {IN => "a b b c c d d e e f f g\nc x x y y z\nf r r s s t\n"},
-    {OUT => "a\nb\nc\nx\nd\ny\ne\nz\nf\nr\ng\ns\nt\n"}],
+    {OUT => "a\nb\nc\nd\nx\ne\ny\nf\nz\ng\nr\ns\nt\n"}],
 
    # Before coreutils-5.0.1, given an odd number of input tokens,
    # tsort would accept that and treat the input as if an additional

Maybe there are other solutions as well.

Start with the tree-1 and tree-2 cases in tests/misc/tsort.pl and inspect how build-gnu.sh regenerates or updates GNU test cases. Run the relevant tsort test to confirm the current outputs differ while remaining valid topological orderings. Done means the cases accept the implementation's valid output without masking other failures.