Arch Linux

Ah, crap, really sorry about that, seems I forgot to test it on a clean system. I wonder what the best way to solve this currently would be. Would prefer not having to move it back to the AUR, but it seems like that's sadly necessary for now, hmm.

Oh, I think part of the reason I failed to realize this is that python-html5lib-git actually builds from a more than 3 versions old commit, hmph.

Your work to get this package from AUR to community repo is really appreciated. I also think it should be available here, to make jupyter-nbconvert work. Downgrading the python-html5lib community package is probably not an option, since other packages depend on it.


A possible temporary solution is to bundle the older html5lib-0.9999999 (7 9's) directly in the python-bleach package, until html5lib-1.0 with stable API is released and bleach adapted it ( bleach issue: https://github.com/mozilla/bleach/issues/229 ).

For bundling some changes has to be done:

* install html5lib in site-packages/bleach
* add local html5 package folder to search path in bleach
* remove html5 dependency from bleach (to avoid pkg_resources errors)


I have done this in an updated PKGBUILD. Tested and working with Python-3.6. The bleach library can be imported and also "jupyter nbconvert" works (with an also tweaked python-entrypoints package with added egg-info file). What do you think about it?


# Maintainer: Johannes Löthberg <johannes @ kyriasis.com>

pkgbase=python-bleach
pkgname=(python-bleach python2-bleach)
pkgver=1.5.0
pkgrel=3

pkgdesc="An easy whitelist-based HTML-sanitizing tool"
url="http://pypi.python.org/pypi/bleach"
arch=('any')
license=('Apache')

makedepends=('python-setuptools'
'python2-setuptools')

source=("python-bleach-$pkgver.tar.gz::https://pypi.org/packages/source/b/bleach/bleach-$pkgver.tar.gz"
"https://github.com/html5lib/html5lib-python/archive/0.9999999.tar.gz")

md5sums=('b663300efdf421b3b727b19d7be9c7e7' '2ca78b1ec5852779bc121a97da6e8d4d')

prepare() {
cp -a bleach-$pkgver{,-python2}
}

build() {
cd "$srcdir"/bleach-$pkgver
# add local html5lib to search path
sed -i 's/import re/import re\nimport os, sys\nsys.path.insert(0, os.path.dirname(__file__))/' ./bleach/__init__.py
# remove html5 dependency
sed -i '/html5lib/d' ./setup.py
python setup.py build

cd "$srcdir"/bleach-$pkgver-python2
# add local html5lib to search path
sed -i 's/import re/import re\nimport os, sys\nsys.path.insert(0, os.path.dirname(__file__))/' ./bleach/__init__.py
# remove html5 dependency
sed -i '/html5lib/d' ./setup.py
python2 setup.py build
}

package_python-bleach() {
cd "$srcdir"/html5lib-python-0.9999999
python setup.py install --root="$pkgdir"/usr/lib/python3.6/site-packages --install-purelib=bleach --optimize=1

cd "$srcdir"/bleach-$pkgver
python setup.py install --root="$pkgdir" --optimize=1
}

package_python2-bleach() {
cd "$srcdir"/html5lib-python-0.9999999
python2 setup.py install --root="$pkgdir"/usr/lib/python2.7/site-packages --install-purelib=bleach --optimize=1

cd "$srcdir"/bleach-$pkgver-python2
python2 setup.py install --root="$pkgdir" --optimize=1
}

for a quick workaround, i’ve also created https://aur.archlinux.org/packages/python-html5lib-9x07/

Okay, so, I've now pushed python{,2}-html5lib-7-9s to [community]. Please tell me if there are any issues with it, but for now I'll close this issue since it seems to be working with that version.

Thank for the solution, but it's not working yet. The PKGBUILD for the python{,2}-html5lib-7-9s package is missing the lines:

provides=('python-html5lib')

and

provides=('python2-html5lib')

also this means we’re stuck on an old html5lib version. as said, it’s a workaround no solution.

the solution would be what sebastian said:

1. put a matching version of the module to a non-$PYTHONPATH location (e.g. “…/site-packages/bleach/vendor/”)
2. patch bleach to use it (add “…/vendor” to sys.path or so)
2. patch out html5lib from “…/site-packages/bleach-1.5.0-py3.6.egg-info/requires.txt” so pkg_resources doesn’t complain

i’d however write a real patch instead of using sed, then we aren’t incentivised to use shortcuts like adding “…/site-packages/bleach” to sys.path, as this adds all submodules to pythonpath instead of just html5lib. instead:

here = os.path.dirname(__file__)
sys.path.insert(0, os.path.join(here, 'vendor'))

That isn't really a solution either, it's just another workaround. Anyway, at first it didn't work for me, but just realized what I did wrong, and just pushed a version with the vendored copy can use the newer version.