[cleanup] Remove warning about index access out of bounds.

Closes #35 (closed).

@simon.praetorius: This is the fix I had for the warnings, however, I'm not sure if this is the best way to fix this.

Thanks. I will have a more detailed look. My initial solution broke something in my testsuite - so it was wrong.

The same condition would also be useful in the line above (254), since also there the corresponding array element is accessed. Assuming the conditions in the assert() in the first line of that function hold, in the whole else-block it holds codim < dim. And thus dim - codim - 1 >= 0. We could thus also transform the whole else-block into an else if (codim < dim) block. But then we would have a missing return path for the potentially impossible case codim > dim.

So, I was thinking. What if we transform the assert() checks in the top into proper exceptions that are always checked. Is this method anyhow time-critical? The introduced if-clause inside an inner loop is probably not performance-enhancing.

An alternative simple fix would be to change most else-clauses into else-if-clauses and add a final else { std::abort(); return 0; } or something. Maybe with an __unreachable() or similar attribute.

changed this line in version 2 of the diff

I have uploaded the alternative simple fix into this branch.

Great, thanks. Merge at will.

added 1 commit

• cd91d194 - Handle all cases in referenceEmbeddings explicitly

Compare with previous version

added 6 commits

• cd91d194...a51164b6 - 4 commits from branch master
• d231994d - [cleanup] Remove warning about index access out of bounds.
• 3996749d - Handle all cases in referenceEmbeddings explicitly

Compare with previous version

mentioned in merge request !238 (closed)

approved this merge request

This MR fixes the warnings mentioned in #35 (closed), but not all warnings shown by gcc12 with -Wall. There are more places in other functions. It would be interesting whether there is an underlying pattern / source of the warning that we could fix instead of the symptoms only.

changed the description

merged

mentioned in commit d1d63a4d

There are still a few warnings left. The question is: What happens in the case that dim == 0? Then codim must be 0 as well, due to 0 <= codim <= dim. This needs to be added as a check, i.e.,

if( 0 < codim && codim <= dim ) // => dim > 0

The second check should always be true, thus does not change the logic.

Maybe we should add a macro (into dune-common) DUNE_ASSERT_OR_ASSUME that is a simple assert in the debug case and an [[assume: ]] attribute (or something similar)` in the release case.

mentioned in merge request !239 (merged)