Do not flag identical negative numbers as different. #4490
Conversation
Yeah, I think it makes sense to disable that behaviour. Tracing the history, this mode seems to have been present since commit 51c7a97 (originally OPM/opm-output@a2897a2, arguably since OPM/opm-output@4aa250e, OPM/opm-output#55) and never really justified anywhere. I believe the original goal was to prevent including reference solutions with negative saturations &c, but since we've since added such solutions it has not met its goal. Altering the comparison to accept negative values as long as they don't change (too much) is probably beneficial in the short term. Long term I do think we should strive towards never generating such negative values, but that's a much bigger program. |
|
jenkins build this please |
1 similar comment
|
jenkins build this please |
|
@bska ok to merge this? |
This applies to the comparison program that checks regression failures. With the current approach, keywords in the "keywordDisallowNegatives" set will be treated such that even if the reference and test results are identical, if negative it will still be flagged as a failing test (if the negative values are larger than the absolute tolerance). This is unproductive when applied to the regression tests, and just gives unnecessary failures. If the intention is to disallow any negative values at all in certain fields, the code must be changed to do just that (and not allow negative values with small absolute value), I would support a move in that direction, although it would also require modification to Flow, which currently can indeed produce negative values for SGAS and SWAT.
The change proposed here is a simple disabling of this (in my opinion) strange behavior, but we could discuss more fundamental changes as outlined above.