Re: Do not create conditional installation directories

[Ralf Wildenhues]

* Ralf Corsepius wrote on Thu, Mar 19, 2009 at 03:01:45PM CET:
> Ralf Wildenhues wrote:
>> * Akim Demaille wrote on Wed, Mar 18, 2009 at 06:31:50PM CET:
>>   
>>>>>> "RW" == Ralf Wildenhues <address@hidden> writes:
>>>>>>           
>>>  > But the question remains: how do you think we should address the bug
>>>  > that bothers Akim, other than by this patch and a big warning in NEWS?
>>>
>>> My focus was more about an empty $(foodir), not an empty $(foo_DATA).
>>
>> Can you describe situations where an empty $(foodir) happens?
>>   
> The classic use case are packages which use a directory but not put any  
> files into it.

We are still talking past each other.  AFAIU, Akim is talking about
cases where the `make' macro `foodir' is initialized to be the empty
string.  You are talking about a use case where the macro `foodir' is
set properly but `foo_DATA' is empty.

Of course my last reply on this issue was confused, too: it would never
happen that things are installed in '/' by mistake, only in $(DESTDIR)
if that happens to be nonempty.  An empty destination directory variable
causes different things with 1.10.2 and 1.11a, with the latter either
causing an error from the `install' program, or, if exactly 2 files are
to be installed, overwriting a file in the source or build tree.  I
agree that both of these outcomes are a bad thing.

I'll fix this issue.

Cheers, and thanks for being so patient with me,
Ralf