extends with increments can result in missing values

[djay]

The following test fails

def increment_buildout_with_multiple_extended_without_base_equals():
    r"""

    >>> write('buildout.cfg', '''
    ... [buildout]
    ... extends = base1.cfg base2.cfg
    ... parts += foo
    ... [foo]
    ... recipe = zc.buildout:debug
    ... [base1]
    ... recipe = zc.buildout:debug
    ... [base2]
    ... recipe = zc.buildout:debug
    ... ''')
    >>> write('base1.cfg', '''
    ... [buildout]
    ... extends = base3.cfg
    ... parts += base1
    ... ''')
    >>> write('base2.cfg', '''
    ... [buildout]
    ... extends = base3.cfg
    ... parts += base2
    ... ''')
    >>> write('base3.cfg', '''
    ... [buildout]
    ... ''')

    >>> print_(system(buildout), end='')
    Installing base2.
      recipe='zc.buildout:debug'
    Installing foo.
      recipe='zc.buildout:debug'
    Installing base1.
      recipe='zc.buildout:debug'
    """

Currently fails as follows

Expected:
    Installing base2.
      recipe='zc.buildout:debug'
    Installing foo.
      recipe='zc.buildout:debug'
    Installing base1.
      recipe='zc.buildout:debug'
Got:
    Installing base2.
      recipe='zc.buildout:debug'
    Installing foo.
      recipe='zc.buildout:debug'

If you remove [buildout] from base3.cfg or remove the extends=base3.cfg the test will pass.

Might be related to #1060236

[djay]

The reason this happens is that in _update_section during the first merge

_update_section("", "parts+= base1") -> "parts= base1"
_update_section("", "parts+=base2") -> "parts=base2"

Then finally

_update_section("parts=base1", "parts=base2") -> "parts=base2".

I suspect the way to fix this would be that unless you are merging += with an = then the result should still be +=. Then once everything is merged, switch any leftover += to a =.

[djay]

I meant to add this is related to #22 but I think the solution is different

[distributist]

I've created a pull request that (I hope) fixes this issue: #226