dep-tree check list of violations non-deterministic?

[jhovell]

dep-tree check seems reliable in determining there are no dependency rule violation in my project (when there are none), and in a sense reliable at determining there are dependency errors in my project (when indeed there are errors).

The issue is that the list of dependency errors is different every time I run. My guess is the program is performing some best-effort list of errors or halting it when it has some but not exhaustively returning all errors which makes cleaning up violations a bit of a game of whack-a-mole & it's difficult to analyze problem scope from the outset. I have never seen more than maybe ~15 violations when I suspect there are many more. The number of files scanned stays the same.

I had been on version v0.19.9 and just upgraded to v0.23.4 (latest available in brew) and this issue exists in both versions... python project being analyzed.

Is this a known issue?

[gabotechs]

The issue is that the list of dependency errors is different every time I run. My guess is the program is performing some best-effort list of errors or halting it when it has some but not exhaustively returning all errors

If there are cycles in the dependencies, it's possible that the reported errors are not deterministic, as I don't think the underlying graph computation library used for detecting cycles is deterministic in its reports.

When you say it's not deterministic you mean that:

• without changing any code, running dep-tree twice reports different sets of violations?
• or after changing some code dep-tree reports new violations that weren't there on the first run?

[jhovell]

When you say it's not deterministic you mean that:

• without changing any code, running dep-tree twice reports different sets of violations?
• or after changing some code dep-tree reports new violations that weren't there on the first run?

The former. I have run the dep-tree check command 5 times in a loop without changing any code, and the output differs notably on each run. The number and specific violations reported fluctuate between runs, with some dependencies reported inconsistently or only in some runs. Output from my run:

• Run 1: 11 violations reported
• Run 2: 11 violations reported (but different ones!)
• Run 3: 13 violations reported
• Run 4: 8 violations reported
• Run 5: 14 violations reported

In these runs:

1. Different violations each run - Some runs show module x -> module y`, others don't
2. Different counts - Ranging from 8 to 14 violations
3. Missing/appearing violations - module z appears in Run 5 but nowhere else
4. File inclusion varies - Sometimes module violations appear, sometimes not

I had been trying to use this tool as a gate for CI but having different results on each run makes it confusing for developers and not reliable to track actual compliance improvements.

Thanks for any help here!