Re: [PATCH-for-5.0?] configure: Do not leave sphinx in-tree artifacts

["Philippe Mathieu-Daudé" <philmd@redhat.com>]

On 4/3/20 7:49 PM, Eric Blake wrote:
> On 4/3/20 12:37 PM, Eric Blake wrote:
>> On 4/3/20 11:57 AM, Peter Maydell wrote:
>>> On Fri, 3 Apr 2020 at 17:54, Philippe Mathieu-Daudé 
>>> <philmd@redhat.com> wrote:
>>>>
>>>> When ./configure checks the sphinx version is new enough, it leaves
>>>> the docs/sphinx/__pycache__/ directory. Avoid this by using the '-B'
>>>> option (don't write .py[co] files on import) via the
>>>> PYTHONDONTWRITEBYTECODE environment variable.
>>>>
>>>> Reported-by: Eric Blake <eblake@redhat.com>
>>>> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
>>>
>>> This only happens for an in-tree build, right? I think in
>>> that case you're kind of OK with having random stuff
>>> left in the source tree... It seems easy enough to suppress
>>> them though, so I guess we might as well.
>>
>> It happens in VPATH too - and what's more, in VPATH, it is still 
>> creating it under srcdir rather than builddir, which feels like 
>> unnecessary pollution.  I was trying to prove whether 'make distclean' 
>> got us back to a pristine state; this was one of the files that 
>> escaped 'make distclean', so our choice is to either add it to the 
>> clean rules, or to avoid creating it in the first place.  I like the 
>> approach of not creating it in the first place :)
>>
>> Reviewed-by: Eric Blake <eblake@redhat.com>
> 
> Hmm, I spoke early.  Your patch only addresses the pollution during 
> './configure'.  But running 'make' (even in a VPATH build) equally 
> creates the same pollution.  Which means we really ought to be cleaning 
> it up during 'make distclean' rather than just trying to make 
> './configure' clever.

Oh I only checked ./configure indeed, sorry. Thanks for testing!