@IvanChepurnyi
Thank you for yet another valuable contribution.
In order to accept it, all code changes should be supplied with the corresponding unit/integration tests.
Also, to simplify further code review, the conventional bug report would be greatly appreciated:

1. "Steps to reproduce"
2. "Expected result"
3. "Actual result".

Added integration test. It fails on 3rd data-set with non-fixed core code and on passes if my fix is applied.

The expected result is correctly sorted totals by before and after properties. The actual results and steps to reproduce already specified in original bug report.

@IvanChepurnyi: Can you also add this patch to https://jira.magento.com/browse/MCACE-129 ?

@IvanChepurnyi
Thank you for adding lacking tests.

The contribution has been accepted with the following deviations:

• minimized usage of Reflections for testing
• implemented additional tests for ambiguous situations during sorting
  • replaced testing logic to compare against expected results in order to be able to validate ambiguous cases

Again, thank you for all the effort and valuable contribution.

Changes will become available with one of the nearest merges from the internal repository. Closing the ticket.

@IvanChepurnyi Is this a toplogical sort algorithm? I am wondering what happens if the specified orders are contradictory (total foo after bar, total bar after foo). Shouldn't that be detected and an exception thrown?

It is not a good idea to throw an exception, where it was not thrown before, as far as I know the fixes from Magento 2.0 go to current version as well, and if at some Magento version conflict were resolved by some of the properties but later on an exception were thrown, then existent shops who updates via Magento connect are in danger of crashing down...

But if there a contradictions the sort order might be completely undefined ... so I think an exception is much better than an undefinied state that could cause to further problems such as wrongly calculated totals (you might sell your stuff without tax then an so on)

For this case after has always higher priority. Since grand_total gets after tax, tax goes after shipping, discount and so on, you will never get a situation of wrongly calculated taxes. It only affects custom totals that manually setting up positions for themselves.

How is the result defined if A after B , B after A was defined?

If A is processed after B then A will be placed after B, if B is processed after A, then B will be after A.

So one of the conditions is not met. For those cases also the testcase would fail (if you provide it with such "invalid input").

I still have the opinion that this can cause business critical problems that are very hard to track down and should be detected well. Everything else makes the system unreliable.

If someone updates Magento and has incompatible modules that cause such a problem, an exception would be noticed very soon (I hope nobody does an Magento update on a live system without testing).
But if the error is such a hidden one, it might go unnoticed and cause several business damage.

Sorry, Alexander, but I don't see any reason to argue with you about the exception, the patch was already submitted to core team, covered with integration tests and accepted by them into the next release. If you do not agree with my contribution, you are always free to create another pull request with your exception throw statement.

For my opinion it should not be thrown, to not create fatal error on the frontend, maybe it better just to log debug message somewhere. At least, in another core functionality with before/after sorting, there is no exception thrown if two blocks have cycling reference in before and after attributes (layouts).

As for store owners who update live version via Magento connect, it is true, and I've helped such people enormous amount of time during all the 5 years of my Magento development experience. Of course it is not recommended to update Magento version without testing on staging environment, but a lot of people just do so.

No offense. My intention is also not to argue. Its just about constructively improving your pull request.

This total sorting problem caused a lot of people lots of headaches and I am thankful that you fixed it.

But if it does not reliably fulfil the conditions giving in the XML configuration and silently ignores some of them I am just afraid that a similar issue in Magento2 will waste a lot of peoples time who have to debug this.

I also understand your point, that fatal errors for endusers are not nice. I think there is room for discussion what is better: A order that does not work, or an order with possibly wronlgy caluclates totals (if a total B that was specified to run after A is not run after A, the required total fields in the quote might not be available and the total might return 0).
Both is not good - so I think just logging an exception to exception.log might be a good compromise.

Actually I did not yet have the time to reconstruct the algorithm used in your pull request to identify the place where we could log an exception. That is one reason I did not enhance your pull request myself.

(My pull request #65 does not provide the feature to generate a "nearly correct" order - it just fails completly so I could only throw an Exception with a fatal error to the end user).

Alexander

We discussed the question of "to throw or not to throw" with product management and agreed on the following:

• During checkout or anywhere on the frontend, the contradictory or ambiguous totals must not trigger errors and must not prevent customer from buying. The justification for this argument is that it is better to sell with slightly wrong price, rather than not to sell at all.
• When totals are calculated, add validation of their declaration. The validation errors must be logged to system/error log
• Implement an integrity test that would invoke validation of declared totals by the currently deployed Magento instance (it might include 3rd-party extensions with own totals). If ambiguous or contradictory declarations are declared -- fail the test. In such a way the developer or system administrator, when introducing an extension with new totals, will know if they break the system or not.

All the mentioned above has been implemented. Representation of totals is implemented as a graph (new library Magento_Data_Graph, but used only in validation for now). In this graph, we search for loops in various ways. If a loop is found, it means that it is logically impossible to transform it to a simple graph (or sort graph nodes).

The topological sorting algorithm should perfectly fit into the new model Magento_Data_Graph. We'll recommend it to author in the pull request #65. Changes will be rolled out later with next code updates.

Alright - thanks for elaborating this.

"The justification for this argument is that it is better to sell with slightly wrong price, rather than not to sell at all." - makes sense to me. But I think we can not be sure, that such problems cause only slightly wrong prices... - depending on the totals to be affected.

Integrity Testing and logging to system/error.log sounds good to me and might be a good trade off. So we could make it up to the users if they conduct such tests and check the error log.

@magento-team Is there any chance that this patch will get backported to Magento 1.9?

@Naitsirch in case you are still interested two years later, there is a patch for Magento 1 on StackOverflow

Thanks for the hint @amenk