On Saturday, 19 January 2019 11:17:19 UTC, dcs3spp wrote:
> On Saturday, 19 January 2019 07:33:50 UTC, dieter wrote:
> > dcs3spp via Python-list <[email protected]> writes:
> > > On Friday, 18 January 2019 07:39:00 UTC, dieter wrote:
> > > ...
> > > My situation is similar to the following....
> > >
> > > Assume the following two privately developed projects that I have
> > > written, each with their own setup.py:
> > > 1. parent exists in folder $HOME/project/
> > > 2. child exists in folder $HOME/a_different_project/
> > > *child could be also be used by other projects in the future....
> > >
> > > parent has a dependency requirement on child and so child is listed as an
> > > install dependency in parent setup.py
> > >
> > > if I try python setup.py develop from parent project it fails to find
> > > child dependency on build.
> >
> > Do you tell me to have a cyclic dependency structure?
> >
> > If not, first install "child" in your virtualenv (by whatever
> > means, maybe "python setup.py develop"), then "parent".
>
> Thanks for your suggestion.No, no cyclic dependency structure. Child does not
> depend on parent. Child could be used by separate projects in the future.
> Child will not be cyclic, it will not import projects that import it.
>
> Installing child first is what I tried yesterday,
> https://github.com/dcs3spp/setuptools_dependency_example/blob/master/README.md
> , when trying to understand the suggested approach of using python setup.py
> develop to pull dependencies.
>
> The advantage of the devpi approach, adopted earlier, is that this additional
> complexity is hidden from developers. They do not need to be concerned about
> the order that they install dependencies, e.g. install child first and then
> install parent. Developers just run python setup.py develop and it
> automatically fetches the child dependency.
>
> However, the devpi approach requires a private repository to be setup and I
> am not sure whether it is possible to give access to it for usage in CI cloud
> technologies, such as gitlab.com. If I stuck with devpi then I would
> presumably have to follow the route of a bare metal environment for CI server
> (e.g. Jenkins). Hence, my thoughts on paying subscription to a private pypi
> cloud repository…..
>
>
> My question is, can setuptools be configured to pull in child from a separate
> git repository when running python setup.py develop from parent folder? I
> have since found and tried this approach at
> https://stackoverflow.com/a/53706140/8325270
> It appears that later versions of setuptools can install via a PEP508 url. I
> currently trying to investigate this approach…..
After trying PEP508 url approach my conclusions are as follows.
A PEP508 url for a git repository can be used in *install_requires* of
*setup.py*. An example is listed below.
```
requires = [
'parent',
'kinto-http@git+https://github.com/Kinto/kinto-http.py',
]
...
install_requires=requires
```
The package can then be installed with pip, using ```pip install -e . or pip
install .```
However, installation with setuptools is then broken, i.e. ```python setup.py
develop``` and ```python setup.py install``` fails. setuptools looks for
packages in pypi indexes. To install using setuptools a devpi index would have
to be installed and configured or packages would have to installed from a paid
for pypi repository in the cloud. Alternatively, developers could manually
install each private package dependency individually, prior to running
```python setup.py develop``` for the source package. Unless, there are other
alternative(s) such as zc.buildout with mr developer plugin etc.....
If I want to have a Python private project, referencing other private
project(s), available under source control and CI via gitlab.com, it seems that
I can use the pip approach with PEP508 or use a requirements.txt file
containing the git projects referenced as PEP508 urls, i.e. ```pip install -r
requirements.txt```.
Confusion, stems from the fact that pip and setuptools dependencies are then
not synchronised, i.e. setuptools will break if PEP508 urls are listed for
install_requires. Presumably the approach is to use either pip or setuptools
but not both?
--
https://mail.python.org/mailman/listinfo/python-list
