The build graph is incorrect when build an optimized segment sequence.

[OTime-star]

In the encoding process, we will construct a graph, and then use the single-source shortest path algorithm to find the optimal segment sequence, but the constructed graph seems to be problematic.

Keep other parameters as default, and then encode "ABCDEF123", you will get the following graph(Byte mode is omitted).

The cost from node ("ABCDEF[A]") to node ("123[A]") is 16, which is wrong, it should be 17.

[soldair]

If you would like to take a shot at making this work more correctly I'm happy to take a look at a pr

[OTime-star]

I created the PR, it is here #273

[Greenheart]

I added this fix to my fork and it works well. The output remains the same even with this patch applied. All tests continue to pass 👍

For the specific test case, it wouldn't change the behavior, because the cost from ABCDEF [A] to 123 [N] (24) is still less than the cost to 123 [A] (16, which should be 17).

This is why there is no difference in behavior for the Segments.fromString() method. However, this invalid graph might cause issues in some cases, so it would be good to fix this.

However, it would be good to add a unit test to cover this regression. I have migrated to another testing framework in my fork and used Vitest in-source testing to verify the updated behavior of this internal function, buildGraph()