NoFAQ – A tool that corrects your console commands | Hacker News Reader

NoFAQ – A tool that corrects your console commands

nofaq.cs.wisc.edu

21 pointsbased220 comments

I like that, in response to a name resolution failure in a ping command provided by the "See a random example" button, it suggested

    sudo rm -rf /*

Maybe a little curation might be of value here?

Indeed curation is required, as an example you forgot to include ..

    --no-preserve-root

;)

Using '/*' doesn't need --no-preserve-root. Only 'rm -rf /' needs it.

It seems like most people who are entering commands don't get what it's supposed to do, since most of the commands under "Help Fix Commands" are either malicious, don't have an associated error message, or both.

Also, it seems like it's being a little bit racist about what a "console command" is, because it won't let me suggest...

  (dir -rec).FullName

...as a fix for...

  PS C:\> dir /s/b

  dir : Cannot find path 'C:\s\b' because it does not exist.

  sudo chmod -R a-x /

would work as is.

That's definitely in the works. I'm working on a reporting/curating functionality for "fix-requested" commands as well as for suggested fixes. I'm hoping to push that to the live environment as soon as possible.

Not sure how well it works:

- Suggested `ls` for `sl` despite the fact that the latter [0] is a wonderful utility.

- `chomd` didn't yield any suggestion. Neither did `fskc`.

A similar utility that actually helps is thefuck [1].

[0] https://github.com/mtoyoda/sl

[1] https://github.com/nvbn/thefuck

Oddly enough, both of the points you bring up are handled perfectly by zsh's built-in correction. zsh will only suggest 'ls' for 'sl' if 'sl' isn't installed, and it'll catch 'fksc' and 'chomd'.

How well it works is: exactly to the extent that the crowd-sourced data is accurate. No more and no less.

It is not reasonable to suggest "ls" for "sl". If we regard this as typo correction, we have to consider that no letter in the command is in the correct position.

For typo correction to be plausible, it has to affect only some reasonable fraction of the command. The command has to be fairly long and the edit required to correct it has to be fairly small. E.g. showmnt -> showmount.

A lot of these suggestion things are nifty - another one is a machine learning code helper (spots errors, identifies errors, etc).

Not sure if this even has an api.. but i want it to. Despite that, my problem with them is that there's no way i'm going to send every failed command or all my code to some random 3rd party. It just isn't happening.

I'd love for this thing to be a local install, and then i can pipe any command failures i have to it locally, and feel semi-secure about them.

Really, i think i just need to bite the bullet and improve my understanding of the math behind ML (i'm terrible at math), and take a course or two on ML. I feel like there are so many opportunities for ML in your local life, that you would never ship off to a 3rd party. Hard to profit from, but seems great for open source projects.

(Note, i didn't mean to imply that this site used ML at all.. the other code suggester service does.. and the two, from a UX perspective, feel similar to me)

Thanks for your feedback! Currently, we don't have an external API, for curation and server load reasons. Also, we want to be as transparent as possible about what data is collected - the current website is a wholly "opt-in" sort of design.

Warren Teitelman implemented a rule-based command correction facility in BBN Lisp in the 1960s (https://dspace.mit.edu/handle/1721.1/6905) that later became known as DWIM in InterLisp (http://www.softwarepreservation.org/projects/LISP/interlisp-...). It was notorious for introducing weird errors: https://groups.google.com/d/msg/comp.lang.misc/hBQ1autWqlI/I.... There is also an anecdote about it deleting files in ESR's Hacker's Dictionary (http://www.catb.org/jargon/html/D/DWIM.html) that seems to be ESR's made up bullshit - the syntax is wrong, the anecdote is not in the old jargon file, and also the fact that Interlisp had undelete. A lot of people seemed to have liked the facility though.

I remember a nifty little python app called 'fuck', where you'd typo something, then type "fuck", and it'd take a guess at correcting it for you.

Was pretty cool!

It's even cooler now! https://github.com/nvbn/thefuck

Mr. Iakolev's tool helped inspire our work, actually! https://nofaq.cs.wisc.edu/about.html

    $ git pull origin master
    Your local changes to the following files would be overwritten by merge: user.abcd.suo

Suggestion: git rm --cached user.abcd.suo

Wooaa, no! (Now your local changes are gone.)

And maybe that's desired.

Or maybe it isn't.

The main problem with this, as I see it, is the lack of knowledge of the user's intention.

Without that, it's only good for total basics that built-in correction or 'thefuck' would handle.

With that, well.. that's StackOverflow.

Needs a way to downvote bad data. Someone put in:

   Command: ls -l .bla
   Error: [ no output ]
   Suggestion: ls -lah .bla

Is this real?

Which environments have a Unix-like ls which refuses, without -a, to list an explicitly named entry which exists, but starts with a dot?

Flagging of bad data is currently in the pipeline. We hope to push the feature soon.

This definitely needs some cleanup. I like the idea, but the suggested fix for "git remote" is "sudo rm -rf /". Thankfully it doesn't have --no-preserve-root, but still..

Reply on news.ycombinator.com