A different type of tarbomb.

While this tarbomb is contrived, the fundamental problem of digging a command’s syntax without leaving the command line is real. Going to a browser to get help risks introducing human task switching. The traditional answers are:

  1. Man pages:

    $ man tar
  2. Help options:

    $ tar --help

Neither are bad options, but it takes a longish time to scan through detailed manpages while the terser help text doesn’t include all the details you might need. So my habit is to type a few words into Google and click on the Stack Overflow link, which works as long as I can avoid taking a quick look at Twitter or something.

Apparently I’m not alone in being distracted by browser tabs as there are (at least) three command line interfaces for Stack Overflow. In order of GitHub stars:

howdoi (4,540 ★)

Like all of the commands listed here, howdoi feeds the provided search string to a search engine (Google in this case) in order to find the best Stack Overflow question. It then fetches the page and uses pyquery to find the first chunk of code in the top-voted answer. This normally works well, but it can be tricked:

$ howdoi untar multiple files
$ tar xf a.tar b.tar

In this case, the answer1 is explaining why this command won’t work as expected:

$ tar xf *.tar

Fortunately the -a option shows the entire answer. You can also use the -n option to fetch more answers, including this one by konsolebox:

for F in alcatelS*.tar; do
    tar -xvf "$F" 

Still, picking the first codeblock in the top returned answer is the least possible context switching and probably works well enough for most situations.

how2 (4,393 ★)

Written in Node.js, how2 uses the Stack Exchange API to grab complete answers. Unlike the other commands listed here, it defaults to searching Unix & Linux—Stack Exchange. However, you can get answers from Stack Overflow by specifying a tag with the -l option:

$ how2 -l unix untar multiple files

I find this slightly less useful than simply searching for keywords since I don’t always know (or care) where my answer exists. To compensate, there is an option to browse more questions and answers simply by pressing the space bar. It’s a great looking option, though it risks the sort of distraction this tool is designed to avoid.

If you like that sort of thing, the output uses rudimentary syntax highlighting to show the question title, code blocks and other bits of Markdown syntax. I find the code blocks a bit too light against the white background I prefer on terminal window. There are no options to adjust this behavior.

socli (338 ★)

The newest entry is also written in Python and uses Beautiful Soup to extract both the top answer and the question from Stack Overflow. Unlike the other options, it doesn’t use Google but Stack Overflow’s native search to find questions. Search is not one of our core competencies so we use Elasticsearch. It’s good, but not Google good.2

The upshot is the results are a bit off. For instance, this command:

$ socli -q untar multiple files

returns the accepted answer to How can I untar multiple tar files over ssh?. The answer works, but includes the needless complication of involving ssh. socli also includes the question itself, which does provide useful context in cases like this.

Isn’t this a novelty?

While writing this post I did try to use the three commands to solve programming problems. I think with a bit of self-training one of these commands would be mildly helpful. In particular, howdoi wins the most-potentially-useful prize since it uses Google for (shallow) natural-language searching and defaults to just showing code.

But I don’t think I will use any of these commands just yet. When I’ve forgotten to use one, I don’t seem to have that much difficultly getting the answers I need through the browser. Having used Stack Overflow extensively, I don’t find it particularly distracting. My habit has long been to switch between editor, command line and browser.

A deeper problem remains: answers to Stack Overflow questions just aren’t broad enough. For years, we’ve been trying to encourage people to edit questions to be more general, but people are too hung up on post ownership. That’s one of the reasons we’re building Stack Overflow Documentation to encourage collaborative editing. There’s clearly a need for quick access to the dusty corners of programming, so we are banking on Stack Overflow users stepping up to create long-tail examples the way they’ve created long-tail answers. With their help (and a little luck), programmers everywhere will be spending less time reading manuals and more time creating amazing software.

  1. Unfortunately, when I found them these commands didn’t to a good job of showing where answers came from. Technically that’s a violation of Stack Overflow’s Attribution Required policy and practically it means you need to do a redundant search if you want to get the answer’s URL to paste in your code. As of this writing, I’ve submitted pull requests to address this issue:

    • howdoi—Accepted on principle, but still working on the right comment-like prefix.
    • how2—No response.
    • socli—Accepted and improved upon. (Yay!)

  2. One way that native search could be better than Google is if it makes use of metadata unique to the site. On Stack Overflow, one good choice is votes.It’s an easy fix, but I still think a Google search would be better yet.