Preparing for a point release

Richard Laager rlaager at
Fri Dec 8 02:22:48 UTC 2017

On 12/07/2017 03:06 PM, Fred Wright via devel wrote:
> On Wed, 6 Dec 2017, Ian Bruene via devel wrote:
>> For installs the only remaining problem is that for unknown reasons it
>> sometimes doesn't follow the PREFIX when installing the python libs.
> There's nothing "unknown" about it.

Actually, there is something unknown. Gary's case in #414 seems to be
the default of prefix=/usr/local, which should be in sys.path, and yet
he is getting the python bits installed to /usr. Surely that is not the
intended behavior, right?

> The problem is that the location based
> on PREFIX isn't necessarily in the default sys.path, and thus may not
> actually work for imports
> the philosophy is
> to prioritize using a location that actually works over using a location
> that honors PREFIX.
I think prefix should be honored, full stop.

The default case is prefix=/usr/local, which (correct me if I'm wrong)
works without hacks.

The distro-packaging case is to install to /usr, which (correct me if
I'm wrong) works without hacks. For this case, keep in mind that the
final install location is /usr, but prefix is a temporary directory.

The only other real world case I can think of (for software generally,
not just ntpsec), is an install to something under $HOME. These days,
~/.local is a standardized location, that Python supports.

Are there other scenarios that you personally care about, or know of
someone that does? To be clear, I'm not interested in debating
hypothetical scenarios that *could* exist.

Can we drop fix_python_config and just print a warning (at configure
time), if prefix is something other than /usr, /usr/local, or ~/.local?


More information about the devel mailing list