bug#33959: 26.1.90; python.el font-lock buffer wreaks havoc when company is enabled

[Noam Postavsky]

Carlos Pita <carlosjosepita@gmail.com> writes:

> Hi Noam,
>
>> And my " *Python-font-lock*" looks like this:
>>
>>
>>     1 + len('123') + 99 + len('aa')
>>
>>
>>
>>     1 + len(
>>     1 + len('123'
>>     1 + len('123') + 99 + len('aa')
>>
>> The behaviour seems rather inconsistent.  I think to solve this properly
>> we need some deterministic tests which reproduce the problem.
>
>
> The state of your " *Python-font-lock*" buffer is exactly what I mean.
> The duplications you see are caused by spurious newlines added when
> empty output is passed to the filter. There are two bad things there:
> i) empty strings are transformed into new lines and 2) empty strings
> are an undesirable side effect of company being triggered in a hidden
> buffer (this is analogous to the problem with org mode hidden buffer
> and yasnippet that I recently reported). Try disabling company mode in
> python font lock buffer; even though, this is not necessary if (i)
> above is fixed, which IMO should be done independently of the nasty
> company interaction.
>
> Besides, remember that patch 2 is fixing two different issues. This is
> somewhat undesirable but there are a lot of overlapping between them.
> Nevertheless, I provided patch 1 in case you are unwilling to apply
> one of them.

The above state happens after I've applied your patch #2.  So I can't
tell if it fixes anything.  That's why I think we need some
deterministic tests to show what is fixed and what is broken.