Tablegenerator sometimes does not generated a column with expected verdict

[marian-lingsch]

Currently whenever all expected verdicts are the same, the CSV produced by table-generator does not contain a columnd for the expected verdicts. This makes working with the csv's very difficult in automated scripts, since this edge-case needs to always be accounted for.

[PhilippWendler]

The idea behind that was that it is useless to write out redundant data in columns that are always the same. It has become clear that this creates more problems than expected, but of course we cannot just change it because that would also break existing scripts and use cases. We are planning on fixing this together with other issues of the CSV export by a new clearly defined and consistent export: #900.

[PhilippWendler]

We can fix this part of the problematic CSV format by adding a command-line parameter that makes table-generator always use all columns of the task id. This could be implemented quickly and would solve this particular problem without waiting for a full solution for #900.

[dbeyer]

I agree that it is much better to split issue #900 into pieces. Let's fix the problem of missing columns separately here.

I proposed two solutions for the problem (just dumping them here for having them archived):

Solution 1:

• A new parameter -csv-export-complete-task-id or similar is added and exports all components of the task id.
• The current behavior without parameter is not changed.
• Pro: Probably quick and easy to implement.

Solution 2:

• Extend the table-definition format to not only allow to specify which column tags inside a run in the XML are exported but also which attributes of the run are exported as table in the CSV export.
• The current behavior is not changed if the new feature is not used in table definitions.
• Pro: There is no new parameter needed.
• Pro: It allows the user to specify which parts of the task id to export as column.

[dbeyer]

For Solution 2, new additional syntax is not needed if we could assume that attribute names of the run and column names in the run are disjoint. In this case we could simply specify an attribute name as a column for the result and it would be exported.

[marian-lingsch]

Currently I would prefer Solution 2, since it also allows the naming of the column, which currently does not have a name, which makes parsing the CSV's more difficult than necessary 😅