Ok, I went for another tool in the case of the Python version: https://pypi.org/project/spylls/. Do we want to use the same tool for both? E.g https://pypi.org/project/wn/ instead.

@theseion What's the advantage of wordnet over the spell/ispell/aspell family for us?

@fzipi are you booked on using python for the PHP stuff or can we switch to shell and use wordnet there too? I do not see anything out of the ordinary in the python scripts. If we keep python, I think the gen.sh should be transferred to python as well.

I don't mind using other scripting possibilities, shell included (if someone wants to port it).

WordNet is a corpus with additional linguistics information. The other programs use dictionaries to correct spelling. These dictionaries also contain words that aren't really English words (because users would probably still want correction on them) like "xterm" and "comm". In addition, they weren't built for finding English words so their interfaces are really not built for solving our problem.

I spent some time with aspell and hunspell and compared results to WordNet and WordNet produces the least false positives. To give you an idea:

• WordNet does not think that "7z" is a word
• WordNet does not know "xterm" or "comm"
• WordNet is case insensitive
• WordNet detects words from variants automatically. For example, "printing" is correctly detected as a form of "print". aspell does not detect that (without additional arguments)

WordNet isn't perfect of course but it easily beat the other choices. Added benefit: WordNet is very easy to use in Python (e.g. with nltk), and probably other languages.

Primitive example of Python with nltk:

# don't store data in home
export NLTK_DATA=/tmp/nltk_corpi

import nltk
nltk.download('wordnet')
from nltk.corpus import wordnet

words = set(wordnet.words())
'xterm' in words

WordNet wn seems to be less universal than aspell which was originally talked about using. So potentially a similar problem with WordNet to the original problem with using spell: it's easily available for some OSs but not others, or packages may not be up to date, or it relies on 3rd party repos, etc.

There does appear to be a Debian package, though, which maybe should be a good rule for what tooling we do or don't include and rely on.

That is a point. But the features @theseion described make it very worthwhile in my eyes. Thus:

Importance of FP > ease of installation of developer tool

Fair point @RedXanadu. I did check for a Debian package. And since it's in Homebrew, it's available on macOS and Linux. And there's a Windows binary on the WordNet page too.

Look over this and it looks good. The spell.sh is not overly beautiful, but improving that was not the goal of this PR. So all OK and I'll merge now.

(@fzipi : I know you've been assigned during the meeting, but I had a few spare cycles.)