Hacker News Reader

Hacker News

Comment by infogulch | Hacker News Reader

Full thread

infogulch

It's Bash Month on Hacker News, apparently:

My thoughts on writing a Minecraft server from scratch in Bash | https://news.ycombinator.com/item?id=30347501

Bash Pitfalls | https://news.ycombinator.com/item?id=30345487

Show HN: Toolbox – A framework for modular Bash scripts | https://news.ycombinator.com/item?id=30067708

Use Bash Strict Mode | https://news.ycombinator.com/item?id=30217719

A Wordle clone in 50 lines of Bash | https://news.ycombinator.com/item?id=30174112

View on HN

I've been sensing a stigma against shell scripting increasingly over the past decade. It's nice to see bash being used in creative and elaborate ways like this.

The stigma is well deserved and I will enthusiastically spread it.

Bash should be considered an esoteric programming language. Shell Programming Considered Harmful.

It's hard to program, and easy to make mistakes in. It should be left as a tool of last resort.

(that said, I understand the appeal of doing weird things with it, like the Minecraft server, for the same reason other esoteric programming languages have appeal)

What is so hard about it? It feels like python but with an even more english language like syntax, and the man files can be like tomes of knowledge. 'head file.txt' is a heck of a lot easier to wrap your own head around than reading something into an object and calling head on it. Python programmers today would probably import pandas for that.

Not the person you replied to, but - yes, `head file.txt` is perfectly reasonable, as are a few other use-cases like `grep`/`sed`-ing. But when you go beyond "processing streams of lines of text" and start having to introduce conditionals, maps, or functions, the syntax is (for me, at least) opaque and easy to make mistakes in.

Isn't that also a case against learning new programming languages?

I think bash has a lot of gotchas due to its original purpose, and it's definitely more error-prone to write bash scripts than to e.g. write Python scripts, so that should be a consideration.

Ideally we should all try to more programming languages that are harder for us to shoot ourselves in the foot with, rather than easier ;-)

I think considering the purpose of the language, it does rather well. It's really easy to spawn and pipe processes, far more so than in almost any alternatives. Like it's so easy that it's easy to forget that this is like a 5-10 line overhead dance to launch and read from a process in most languages.

The problem usually come when you attempt to use it for nontrivial programming. But that just not what it's for, any more than the difficulty in building an OS kernel is a legitimate complaint against PHP.

I think this is what the post you were replying to is saying.

Stephen Borne famously disliked C, and much preferred Algol.

This is important to know in the context of what you are saying.

The Algol structures in the Borne shell are the most incongruous parts of UNIX.

https://research.swtch.com/shmacro

I will just leave this here.

"I am a strong believer that Bourne-derived languages are extremely bad, on the same order of badness as Perl, for programming, and consider programming sh for any purpose other than as a super-portable, lowest-common-denominator platform for build or bootstrap scripts and the like, as an extremely misguided endeavor. As such you won’t see me spending many words on extensions particular to ksh, Bash, or whatever other shells may be popular."

http://www.etalabs.net/sh_tricks.html

I think it's easy to sort of gloss over how good shell scripts are at the sort of things they actually are good at.

Consider this:

  for f in *png; do convert $f $(echo $f | sed 's/png$/webp/'); done

This is like at least a screen of code in in most languages, you get that sinking feeling where you need to fork a process and deal with its lifecycle and deal with some standard library to list files and don't forget to close the file handle for the directory and god forgive if you run a program with actual output then you need to read from a pipe. Like, geesh, do I have to write a whole program to convert a few files?

In a shell, it's just something you type up because you needed to convert a few files. Don't get me wrong, shell scripts are extremely bad at most things involving non-trivial branches, long-lived state, really most things other languages are good at, but man are they ever good at some things most programming languages are bad at.

    lee@sp3tmp1:ls *.png
    'a b.png'   c.png
    lee@sp3tmp1:for f in *png; do convert $f $(echo $f | sed 's/png$/web'); done
    convert-im6.q16: unable to open image `a': No such file or directory @ error/blob.c/OpenBlob/2924.
    convert-im6.q16: no decode delegate for this image format `' @ error/constitute.c/ReadImage/575.
    convert-im6.q16: unable to open image `b.png': No such file or directory @ error/blob.c/OpenBlob/2924.
    convert-im6.q16: unable to open image `b.png': No such file or directory @ error/blob.c/OpenBlob/2924.
    convert-im6.q16: no decode delegate for this image format `' @ error/constitute.c/ReadImage/575.
    convert-im6.q16: no images defined `b.webp' @ error/convert.c/ConvertImageCommand/3229.

[etc.]

I think you could be making a more complicated point.

Better:

    for f in *.png; do convert "$f" "${f/%.png/.webp}" ; done

Quoting the arguments lets it correctly handle file names with funny characters, including spaces. Using the `${parameter/pattern/string}` parameter substitution means the shell does all the work itself without invoking an external command (other than convert, of course). The '%' in the parameter substitution requires the pattern to match at the end, so "foo.png.png" becomes "foo.png.webp" and not "foo.webp.png" -- an unlikely corner case, but you might as well do it right. ("foo.png.blah" doesn't match "*.png" in the first place.)

For elaborate commands like this, I typically run it once with an "echo" inserted ("echo convert ..."), confirm that the resulting commands make sense, and then hit up-arrow to edit the command and delete the echo.

https://www.gnu.org/software/bash/manual/html_node/Shell-Par...

Quoting is not enough, as that won't handle filenames beginning with dash correctly.

Your script fails if the files referenced by $f contain white spaces or a newline.

You should script defensively, for when this is run on Windows or elsewhere [maybe Android] that will break this assumption.

Newlines can also be part of a filename. For extreme rigor, use find -print0 and xargs -0.

I have covered some of this here:

https://www.linuxjournal.com/content/parallel-shells-xargs-u...

(Somebody took me to the Hacker News front page for this, and nobody said that I made a mistake.)

I agree. It also fails if the filename starts with a dash.

> You should script defensively, for when this is run on Windows or elsewhere [maybe Android] that will break this assumption.

I think you are missing the point. I wouldn't run a shellscript on Windows or any other place than the terminal I entered it into. It's an expression of something I want to do at that particular time in that particular directory.

The entire point of my post is that if I wished to write something portable and robust, I wouldn't use bash in the first place. That's not what bash is good at. But quick one-offs, holy crap is it good at that.

I've run many Borne/POSIX shell scripts on Windows for things that I've needed to do.

The most straightforward POSIX shell on Windows is the Busybox port, which uses the Almquist shell (ash) with many bashisms reintroduced.

https://frippery.org/busybox/

> fails if the files referenced by $f contain white spaces or a newline.

This is said every time Bash scripting is mentioned. Why people can't get into the habit of quoting all variables in Bash is beyond me and the mention of white space makes sense; but how often, if ever, have you encountered a file name that contained a newline?

Most of the Unix tools are line oriented by default anyway and are frequently used to manipulate lists of file names and paths.

It's safe to omit quotes for POSIX shell variables that are initialized as numeric, and you enforce to remain numeric.

All else should be quoted, and newlines can happen, primarily from abuse. Do not trust user input!

Calling fopen() on a file with every ASCII character will work on POSIX, as long as it does not contain a null or forward slash (/). Coding defensively means that this does not crash your processing.

I'm not objecting to quoting; I do it even when the variable contains, or should contain, a number.

I was objecting to the knee jerk inclusion of warnings against file names including newlines. Has anyone encountered one in the wild that was not itself an error?

>> I think it's easy to sort of gloss over how good shell scripts are at the sort of things they actually are good at.

>> Consider this:

> for f in *png; do convert $f $(echo $f | sed 's/png$/webp/'); done

This is buggy, and it actually represents exactly the problem with Bash scripting.

By that logic shouldn’t we also stop using programming languages that aren’t threadsafe? Segfaulting could be prevented if C++ wasn’t so esoteric.

I can tell you’ve dealt with lots of Python programmers.

>What is so hard about it?

quick: without looking anything up, write something in bash that takes a string of comma-separated values and turns it into an array of those values.

then check to see if your answer is on this very long list of wrong implementations:

https://stackoverflow.com/a/45201229/6107981

Easy, I just avoid these sort of mind game situations when programming, and I come up with a simple 'wrong' approach that will probably work fine for my purposes anyhow. I don't care if my implementation breaks if I have a newline in my data or whatever, because I already know based on what I am working on that I won't have to worry about a newline in my data or any other weird quirks they bring up in stack overflow comment sections that I have yet to come across in my years of writing code that gets the job done. I don't care if my solution is fractions of a second slower, it's probably already going to be mindbendingly fast compared to the code running on hardware from 40 years ago when people had to be a lot more mindful about selecting the most performant solutions when writing in bash or anything else.

Okay, but how do you know that those weird quirks won't be present in your data? Or what those weird quirks even are? The details of the stuff in that Stackoverflow comment have already fallen out of my head; I couldn't tell you about them now. With Bash there are so many traps and footguns and (subjective) unknown unknowns, it would keep me up at night worrying about it. There's always another maddening fact to discover.

I would just use Python, where strings are sane and lists are sane: my_string.split(", "), boom, done, no jumpscares.

> Bash should be considered an esoteric programming language. Shell Programming Considered Harmful.

I disagree. Bash is fine. Nothing esoteric about it.

> It's hard to program, and easy to make mistakes in. It should be left as a tool of last resort.

It's easy to program, but it has a learning curve. Use shellcheck. It's OK to go first to bash if you need a shell script. Other languages make pretty crappy shell scripts.

I copy/paste from a previous post. "Exoteric" is a stretch (in particular, because there are no alternative glue languages, if we exclude all the shell variants), but cryptic, unintuitive and error-prone fits well.

Arrays:

  - `${#myvar[@]}`: unreadable

  - `mapfile myvar < <(grep pattern file)`: "mapfile" is not exactly a clear term (there's the synonym
    "readarray", but "mapfile" is the reference; worse, you may find both used inconsistently); the
    command as I wrote it also has two problems

Strings:

  - `${myvar##.*}`: unreadable

  - `echo $(IFS=,; echo "${myvar[*]}")`: unreadable; most also don't know the difference between `[@]`
    and `[*]`; this doesn't also work for two-char join operations

  - `[[ $myvar =~ $(echo '\bpattern\b') ]]`: unreadable, and most don't know why command substitution
    is needed

Redirections:

  - `3>&2 2>&1 1>&3`: most don't know what this means

Yes, it can be cryptic and not beginner-friendly, but it's powerful and once you know it, you know it; it will be useful for years if not decades, as opposed to some tools or frameworks du jour that are sometimes as cryptic.

The main issue is that many people try (by necessity or not) to cobble scripts together without really learning the language and its idioms. It's not that hard to really learn it though, it's basically one man page then the mandatory [wooledge wiki][1].

[1]: http://mywiki.wooledge.org/BashGuide

I don't understand, none of that is unreadable or weird to me. Are you sure the issue is not a lack of familiarity with the language? The language can appear deceptively simple in the beginning.

And what a tool of last resort it is.

It may be much easier (and most of the time I would generally prefer) to use Python for scripting purposes, but Bash is a) almost always available and b) widely portable.

If you manage any servers where you never know if Python is available or if it’s version 2 or 3 (and you have no power to change this for one reason or other), you know the value of this. No, it’s not ideal, but we live in a non-ideal world. Long live Bash.

Bash has the highest rate of Stackoverflow searches per line of code produced. I swear I do about 10 searches per line of code produced.

And when you consider there really isn't a standard library, mostly gee-I-hope-this-UNIX-util-is-installed...

I think the big thing going for it is it's low footprint and mostly universal existence on all servers I work on.

For example, I'm working on a Hashicorp Vault server install and due to it's sensitive nature, I'm strongly considering using bash to set it up over something like python. I love the ability to test in a robust language, but the extra dependencies are hard to justify, at least on a security sensitive application.

> hard to program

Well, we "write" one-liner programs with it all day every day - since it's the shell.

I'm not saying bash doesn't have a bunch of warts, but - what other shell language would you suggest as an alternative to it?

Yeah it gets bashed a lot.

Yet we continue to use it unabashed.

The stigma is not against shell scripting, the stigma is against using shell scripting for things and in settings it wasn't intended for.

Don't get me wrong, I have nothing against fun and creative projects. If someone wants to script something complicated in bash for the lulz, the challenge, to learn something, or just to show off bash-skills, I'm all for it.

What I am against, is when I get asked to debug a 2000 LOC system used in production that was thrown together in bash, waste 2 days on it, only to end up doing what should have been done in the first place, and rewrite the damn thing in 200 lines of Python or 400 lines of Go.

It can, and should, be used to automate some things, string some commands together, make certain tasks easier. I use and write bash scripts in my daily work, and make my life easier with it.

But there comes a point, and that is sooner rather than later, when something should no longer be written in bash, but in a language that wasn't designed primarily as a command line interpreter. Can I make large complicated systems in bash? Sure. Can I carve the handle of a hammer so that it can be used as a screwdriver? I'm sure someone more skilled in woodworking than me could do it. But sometimes the question isn't if it can be done, but if it should.

There has always been stigma against shell programming, not shell scripting, and I think that more-or-less is still the case. It isn't so much anti-bash stigma as, using a less than ideal tool for the job being a smell that'll cost you to fix later.

Shell scripts to automate processes with other tools is exactly what they are designed for, and adding in other logic and loops is fine too, but if you are setting out to create what would be considered a full program rather than an automated task you should definitely consider alternatives (of which there are several suitable choices) that usually end up being easier to maintain.

It used to be that the main alternative was perl, but frankly that was/is no better (IMO much worse) than bash for encouraging the generation write-only (difficult to maintain) code. Python is often a good option, and I've seen node used, though these don't have the ubiquity that sh/bash does (or perl used to have) which is one advantage of bash: it is almost everywhere, even Windows these days (or stick with core posix sh and you'll find support a little wider again), though maybe without some of the supporting externals which is when tricks like these come in.

If you are writing some simple automation, or maybe slightly less simple, then bash is fine as a go-to, but if you are setting out to write something complex bash is a bad choice unless compatibility with many environments you can't dictate have another interpreter present is a non-negotiable requirement.

There is the grey area when a script, or collection of scripts, becomes a larger program/environment over time. Sometimes the effort to rewrite (and fully test) in another environment is not worth the small benefit of doing so, and is therefore a bad use of developer/admin time that could be used elsewhere, particularly when high performance is not an important feature. Though if you find yourself with a lot of bash script, do yourself (and your later replacement when you've moved on to newer projects) a massive favour and make sure it is well documented even if that is only in in-script comments.

Reply on news.ycombinator.com