r.watershed: fix streams and basins

[metzm]

The PR #2648 introduced a bug in r.watershed such that streams and basins are no longer correctly calculated. This PR fixes calculation of streams and basins in r.watershed.

[nilason]

Nice catch!

Underlines the advantages with always-use-braces-policy!
(Could probably have been prevented by ClangFormat InsertBraces: true https://clang.llvm.org/docs/ClangFormatStyleOptions.html#insertbraces).

[metzm]

This could also have been prevented by stricter tests, adjusted in 0ec0004

[neteler]

This could also have been prevented by stricter tests, adjusted in 0ec0004

Thanks.

FYI: I tried the new test my unfixed copy and it fails as expected:

======================================================================
FAIL: test_thresholdsize (__main__.TestWatershed.test_thresholdsize)
Test the expected range of basin output values
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/mneteler/software/grass_main/raster/r.watershed/testsuite/r_watershed_test.py", line 167, in test_thresholdsize
    self.assertRasterFitsUnivar(
  File "/home/mneteler/software/grass83/dist.x86_64-pc-linux-gnu/etc/python/grass/gunittest/case.py", line 301, in assertRasterFitsUnivar
    self.assertModuleKeyValue(
  File "/home/mneteler/software/grass83/dist.x86_64-pc-linux-gnu/etc/python/grass/gunittest/case.py", line 283, in assertModuleKeyValue
    self.fail(self._formatMessage(msg, stdMsg))
AssertionError: Basin values must be in the range [2, 12] 
r.univar map=test_basin percentile=90.0 nprocs=1 separator== -g difference:
mismatch values (key, reference, actual): [('max', 12, 4)]
command: r.univar map=test_basin percentile=90.0 nprocs=1 separator== -g {'map': 'test_basin', 'separator': '=', 'flags': 'g'}

(BTW: the issue was reported on the Italian GRASS GIS mailing list)

[neteler]

@wenzeslaus can we change the Centos7 CI runner to not required? Otherwise we'll not be able to merge this rather important and urgent bug fix...

[wenzeslaus]

Having a failing main and all PRs is bad, too. I suggest reverting the change which caused that and/or removing the CentOS 7 check given how outdated it became and its low usefulness.

[neteler]

Having a failing main and all PRs is bad, too. I suggest reverting the change which caused that and/or

This is not an option as the old runner will go EOL on Sept 11, 2023.

removing the CentOS 7 check given how outdated it became and its low usefulness.

Done in #3141

[neteler]

PR #3141 (drop Centos7) has been merged, this PR needs to be rebased to become merge'able.