This change would break the intention of r.stream.extract. The intention is to follow flow accumulation as much as possible to trace streams, e.g. use flow accumulation from r.terraflow or some external tool or some modified flow accumulation from r.watreshed (see example in the manual), but take advantage of the A* search of r.watershed which does not require DEM modifications to successfully extract stream networks.
A better solution would be to solve ties of equal flow accumulation by following elevation. I will propose a separate PR with this change.

Thinking about it, this change does not break the intention of r.stream.extract because it would still prefer to follow increasing flow accumulation. Therefore I suggest to apply both this PR and my PR #6554

Thinking about it, this change does not break the intention of r.stream.extract because it would still prefer to follow increasing flow accumulation. Therefore I suggest to apply both this PR and my PR #6554

I see your PR fixes the issue, so I wonder if this PR is needed and why was the code block there to start with?

I see your PR fixes the issue, so I wonder if this PR is needed and why was the code block there to start with?

As the code comment says

/* set main drainage direction to A* path if possible */

To elaborate, the intention of r.stream.extract is to use external flow accumulation for stream extraction, but if in doubt follow the A* search direction, i.e. if (fabs(wat_nbr[np_side]) >= max_acc).

Restoring this deleted code block with your PR in addition to my PR should make r.stream.extract robust for special cases as reported in #6541

Since this is a bugfix, backport to G84?