Need a better way to deal with wheels when using pip

[antevens]

We have some problems when dealing with wheels and pip.

Pip does not support wheels prior to version 1.4.0
Pip wheels require setuptools/distribute > 0.8
Python 2.6 and older does not support setuptools/distribute > 0.8
Pip >= 1.5.0 tries to use wheels by default, even if wheel package is not
installed, in this case the --no-use-wheel flag needs to be passed
Versions prior to 1.4 don't support the --no-use-wheel flag

So depending on the version of pip is installed and if wheel support is available we now have to:

< 1.4.0 no flag

= 1.4.0 and < 1.5 ideally pass the use wheel flag if wheel support is installed but no flag is acceptable
= 1.5.0 pass the --no-use-wheel flag if wheel support isn't available, otherwise pass no flag

I started working on this on January 2nd when 1.5 was released but I'm still running into issues for some pip versions/cases.

Right now I'm using the following command but there are some cases where it does not work as it should, we need a better way of dealing with this.

"${pip_env} wheel --help > /dev/null 2>&1 && { ${pip_env} wheel --version > /dev/null 2>&1 || wheel_support_flag='--no-use-wheel'; } ; ${pip_env} --log ${cwd}/pip.log install ${install_args} $wheel_support_flag ${proxy_flag} ${source}

[Ooblioob]

This is a great start, but a slight correction is needed. If an install_arg is passed that requires a second argument, such as -e, then when --no-use-wheel is added pip will pick this up as the argument.

For example, this is what gets generated: pip install -e --no-use-wheel /path/to/my/repo

It gives the following exception:

--no-use-wheel should either be a path to a local project or a VCS url beginning with svn+, git+, hg+, or bzr+
Storing debug log for failure in /home/vagrant/.pip/pip.log

This could be solved by switching the order of ${install_args} and $wheel_support_flag so that wheel_support_flag is first

[teajaymars]

+1 to @Ooblioob, I have just run into this problem and came here to report the issue. I'm having to hack my manifests to work around the problem.

[tremblaysimon]

I don't know if it is related, but I have that issue currently on a fresh system install, without using virtualenv, when I use pip:

Notice: /Stage[main]/Jenkins_ovh::Install/Python::Pip[fabric]/Exec[pip_install_fabric]/returns: no such option: --no-use-wheel Error: pip wheel --help > /dev/null 2>&1 && { pip wheel --version > /dev/null 2>&1 || wheel_support_flag='--no-use-wheel'; } ; pip --log //pip.log install $wheel_support_flag fabric returned 2 instead of one of [0] Error: /Stage[main]/Jenkins_ovh::Install/Python::Pip[fabric]/Exec[pip_install_fabric]/returns: change from notrun to 0 failed: pip wheel --help > /dev/null 2>&1 && { pip wheel --version > /dev/null 2>&1 || wheel_support_flag='--no-use-wheel'; } ; pip --log //pip.log install $wheel_support_flag fabric returned 2 instead of one of [0]