Comment by opan | Hacker News Reader

Comment by opan | Hacker News Reader

Parent Full thread

>the argument used to be you used Windows instead of Linux because on Linux you might occasionally have to use the scary terminal to fix an issue

This always annoys me because you're really handicapping yourself by ignoring the CLI on any OS. Sure I use it more heavily on GNU/Linux than I did on Windows as a kid, but that's because it's so good. If I'm ever in front of a Windows machine now I still like to have a terminal handy (and it's even better/more-familiar on macOS, of course), and I've learned things like "type is like cat", "robocopy is like rsync", "tasklist is kinda like ps and taskkill like kill/pkill" which help me to do things better on Windows than when I used it fulltime. I'm glad Microsoft invested more in the CLI with Windows Terminal, OpenSSH in the default install, winget, PowerShell, etc. I think it's better for everyone. I fear the CLI hate is spreading anti-intellectualism. Some people seem annoyed when they even have to use their keyboard instead of their mouse for something.

GUIs are great for when you're new to a bit of software as you can see the various options and get a feel for the possibilities. CLIs are nearly always more flexible once you've read the man page, but is a steeper learning path.

Automation/scripting is when CLIs really come into their own as otherwise you end up becoming a GUI click monkey. The best is when there's both a GUI and CLI (as long as they work the same way).

I wonder how good CLIs could have been if a fraction of the resources that have gone into GUIs had gone into making CLIs more user friendly. A sequence of words is a pretty natural way of conveying what you want done.

The problem with CLI isn't typing, its discoverability. Keep in mind conveying what you want done requires knowing what can be done first.

I often get my way through unknown CLI commands by just typing TAB and selecting the option that sounds like what I want. Works, most of the time, I don't think that is really an issue. For most CLI programs you also have a reference/cheat sheet, examples and an interactive hypertext systems of tutorials available.

Many GUI designers these days don't like discoverability and don't think it's important. They prefer to hide everything behind "gestures" that you just have to know somehow.

That is exactly the sort of thing I think could have been improved. We have very little at the moment - autocomplete and some help.

Along with the "-h,--help" options, an option to show common usage examples would be useful. I find it much easier to learn from an example as you can modify and build upon it. Otherwise, it's a case of skimming through the man page and then switching back and forth between the man page and the command that I'm writing. (This is most common with tools that I don't use all the time, yet have expert features such as "ffmpeg")

> an option to show common usage examples would be useful.

That's what the man page is supposed to be (and most do contain example) and why GNU wanted to split it into info (tutorials and exhausting documentation) and man (reference and examples).

Yep, the examples on man pages are helpful, but I'd like a slimmed down option to just show the examples and a one-line description for the examples.

I usually end up just doing a quick web search for the command if it's not one that I'm familiar with (i.e. where I have read the man page).

An extreme example of my issue would be trying to find BASH examples of how to process a list of files - the man page on BASH is fairly lengthy.

Meanwhile, Greg's wiki provides this example of processing mp3 files:

  while IFS= LC_ALL=C read -r -d '' file; do
    some command "$file"
  done < <(find . -type f -name '*.mp3' -print0)

Why do you serialize and then deserialize the list of files? Even if you do disable the splitting on reparsing the list, why don't you just do this:

    find . -type f -name '*.mp3' -exec some command {} ';'

That sounds just unnecessary and bug prune. Unless you target some odd platform where find doesn't support '-exec', but it is even in POSIX. I think due to your use of process substitution, your code has a higher chance of being unportable, so why do you want to complicate your code?

Stuff like that is why I personally prefer the man pages to random websites.

Honestly for something as complex as a shell (which describes both a language, an editor, and an implementation) the man page is surprisingly short. (6418 lines for me) I have just found the section on process substitution in <1min, without even using the search, just by reading. I looked it up, because I didn't knew the name of that syntax, so I needed to actually look for what I wanted and I do not use the man page of bash often.

It's to deal with problematic filenames (e.g. containing a newline) and enables setting of variables within the current process. Putting multiple commands into the "-exec" of find is possible, but looks horrible. To be fair, my example didn't use multiple commands, so your version is okay for simple processing.

Don't get me wrong, the BASH man page is great, it's just that I prefer to work with examples.

For reference, here's the Greg's Wiki explanation: https://mywiki.wooledge.org/BashPitfalls#for_f_in_.24.28ls_....

On my computer

    touch "$(echo -en 'a\nb')"
    find . -type f -name '*.mp3' -exec bash -c "echo '{}'; ls '{}'" ';'

works just fine, but maybe it doesn't work everywhere.

If you don't like how the multiple commands look like, you can always write it like this:

    find . -type f -name '*.mp3' -exec bash -c "
        echo '{}'
        ls '{}'
    " ';'

Yep, you can chain multiple commands with find's "-exec", but I'm not a fan of it myself. I suspect setting variables in the current process is trickier though.

(Very minor nitpick, it should be 'a\nb.mp3' to be included, but that does work fine)

Incidentally, ShellCheck isn't happy with that although I don't follow their reasoning:

  find . -type f -name '*.mp3' -exec bash -c "
                                           ^-- SC2156 (warning): Injecting filenames is fragile and insecure. Use parameters.

https://www.shellcheck.net/wiki/SC2156

> although I don't follow their reasoning

I think it is sound. Imagine what happens when the filename contains:

    ' && shutdown now && '.mp3

Of course that makes sense now.

Anyhow here's an example of how I would use the while loop and process substitution in a BASH script:

  declare -i file_count=0
  while IFS= LC_ALL=C read -r -d '' file; do
    file_count+=1
    printf "file: %s\n"  "${file}"
  done < <(find . -type f -name '*.mp3' -print0)
  printf "Processed %d files\n" "${file_count}"

I think that'd be tricky to do using just a find/-exec command.

I see, but now you are essentially operating on multiple files, at once, so the serialization makes some sense. Although for just this, I wouldn't write the operation in bash at all:

    find . -type f -name '*.mp3' | wc -l

Honestly I don't really view the shell / filesystem interface as a security boundary. I use the shell mainly for (automation of) interactive use, so any screwup due to e.g. quoting issues is my own fault, maybe even of using stupid filenames. Shell is a great language to connect streams of different programs into each other, not so much for doing any work. If I do that, I would reach for C.

The serialisation is just to work around the fact that filenames can have any character except for \0 which is why the "-print0" is used. It doesn't by itself allow for concurrent processing.

You're right about just using "wc -l", but I was just trying to demonstrate how you can set variables. A real use would be doing more than just counting files as your example would likely be quicker (assuming that calling an external programme is quicker than running a naive loop in BASH).

I am guilty of using BASH for stuff that most people would use a different language for - I just find that for system admin work that BASH is just at the right level of abstraction for me and is ubiquitous.

If CLIs functioned as LLM and you could talk freely with it, there would be no problem. In reality, CLIs are inconsistent and basically function as robotic interfaces, a lot of them not that far of programming.

> If CLIs functioned as LLM and you could talk freely with it, there would be no problem.

On face value, I find this suggestion hilarious. People are having sandboxing issues left, right and center with AI agents and MCPs, so clearly there would be enormous problems with giving an LLM full unscoped terminal access. Remember the guy who had his hard drive wiped?

Do you just mean that the ABIs are inconsistent and you want a more unified way to specify what you want (but the user still more-or-less spells out what command will be invoked)? I have some sympathy for that concern, yes.

If CLIs functioned as LLM and you could talk freely with it, it would kiss your ass while it hallucinates about the files on your hard drives, invent command line switches that either don't do anything or does something other than what you want, and every command that used to take just 5-10 characters would now require you to type paragraphs. No thank you.

Reply on news.ycombinator.com