Catch edge cases were delimiters are wrongfully caught

[Jan200101]

related to #308

implements more robust logic that splits on the first delim it finds that isn't followed by the same delim

[abitrolly]

This part of code badly needs some test table with expected inputs and outputs. I am having a hard time trying to juggle them in my head.

I spent some time juggling with re.split but then it turned out to be too complicated with no solution.

In [4]: delim = re.escape('+')                                                                                                                                            

In [5]: re.split(delim, '/g++ whoa')                                                                                                                                      
Out[5]: ['/g', '', ' whoa']

In [6]: re.split("%s+" % delim, '/g++ whoa')                                                                                                                              
Out[6]: ['/g', ' whoa']

In [7]: re.split("%s+" % delim, '/g++')                                                                                                                                   
Out[7]: ['/g', '']
...

In [21]: re.split("%s(?:[^%s])" % (delim, delim), '/ge+ll33+++')                                                                                                          
Out[21]: ['/ge', 'l33+++']
...

[Jan200101]

I don't have a complete list of would be inputs, but here are a few test cases that should correctly split query and section:

rm+-rf, rm -rf, g++

[abitrolly]

@Jan200101 how about moving that function to the root scope without modification in a separate commit?

This way we will be able to test previous behavior in automated fashion with bisect.

[Jan200101]

sure thing

[abitrolly]

Well, I actually thought about moving only one function, as it is the only thing which behavior needs to be tested right now. Otherwise the diff becomes quite unreadable. And the move commit needs to be code modification commit for the old behavior to be testable.

[Jan200101]

I had taken the initiative to move out all functions but reversed that

the first commit now moves the nested function out of the scope
and the second commit implements the logic

[abitrolly]

So the way to go from here. I need to setup a development environment to fix tests in master before going further. Then I need to run tests to intercept all queries passed to _add_section_name() and record queries to file. Then record input and output as a test. Then apply these changes, and run the tests.

[abitrolly]

@Jan200101 I've merged one of your commits and added tests on top to the master. If you rebase this and add a couple of edge cases to the list of tested strings, this will be ready to go in.

[Jan200101]

I've added some tests that have + in their name that should be unchanged.

I had added another test case locally but kept it out because it fails the test.
I'll leave it up to you if the test is valid or not

split = """
g++ -O2
g++/-O2
"""

[Jan200101]

I've reworked the logic slightly to deal with the test case above as well as to ease the exception handling a little bit.

[Jan200101]

CI failure seems unrelated to PR.

[abitrolly]

CI errors are strange. Need to investigate them before going on.

[abitrolly]

Fixed tests in #317. Should become green after rebase.

[abitrolly]

While going through the logic I found that it could still strip ++ in lines, so that g++g++ becomes g+/g++, so in #318 I replaced it with a regexp that only matches a single + in the middle. Thanks for the PR pointing out to the solution.