echo 'I wrote a Bash book' >> /dev/world

Certainly there's a lot of stuff in common. I read every single article on Greg's Wiki a couple years before starting this book, and it's informed a lot of my opinions and understanding of the fundamentals of Bash. I do disagree with what seems to be the prevailing wisdom on Greg's Wiki on some points though. For example, I strongly support using errexit rather than trying to implement all the error handling yourself.

Imagine they'd used '>' instead. I don't think that I would appreciate a hard reset of that particular block device.

You're right, I should probably have used `patch` instead.

they shall symlink /dev/world to /dev/null

I agree, Greg's Wiki has a lot of amazing content. Certainly there will be a bit of overlap, but the focus of the book is a bit different:

- It is more relevant for scripting than the interactive command line. For example, it includes

- I've taken a lot of care to write non-trivial example code, which is useful to show when and where constructs are useful, and which other constructs they are usually used with. For example, the argument parsing part shows not just how to use `getopt`, but also how it's meant to be used with `eval`, `unset`, `while`, `case`, and `shift`; the standard `--` option/argument separator; how to handle unimplemented options (anything mentioned by `getopt` but not matched by any `case` statements); and how to deal with the remaining arguments.

- The non-trivial code has been linted using `shellcheck` and a lot of it is tested with `shunit2` (533 lines of tests in total, which are included as a download).

Judging by the index it is more like a "shell 101" plus some bash. Also a paid one.

It's cool, people need such stuff but the title is misleading and the article is basically an ad for a paid service.

It was originally a book, however the publisher reformatted it to a book just before release, possibly because of the length.

As for the "theoretically" part, it is enough to prove that at least one such a system exists that will yield another result than expected. There could be various reasons, from human/software errors to hardware errors like memory bit flips. Since we know these do exist, it is not even theoretically possible to write a piece of software which will behave the same no matter how and where it is run, although it is certainly possible to write it in such a way that it will behave the same in a vast majority of cases.

I'd say that containers are actually something that makes Unix shell a lot more valuable and predictable, and one reason that I'm working on https://www.oilshell.org/ .

A container is basically a "virtualenv" for shell, and it's honestly better than that because:

- it's more general (it isolates every language, not just Python), and

- not even bigger (e.g. an Alpine Linux container is not that much bigger than the Python install in every virtualenv!)

Traditionally you do indeed have the problem that copying a shell script from one machine to another basically guarantees you nothing about how it will work. These days we deploy containers, which solve the problem, and that's why shell is used so much in the cloud (e.g. anything with Github Actions has a boatload of shell)

Even more on that: https://www.oilshell.org/blog/2021/07/cloud-review.html

I remember reading (sorry, can't remember the source) that it's been proven mathematically that there's no way to write code which will run the same on any Turing-equivalent processor. Obviously that's not saying much since Bash is pretty far abstracted from the CPU, it's more of a caveat that I've hardly ever seen a discussion about Bash which doesn't involve at least one "well, actually" comment about how the suggested code doesn't do what the autor purports on some incredibly niche platform or with some highly unusual configuration.

You can set IFS='' globally, which disables splitting. (And you have to make sure it stays that way, since there are some idioms which modify IFS)

Although there is the "empty elision" case related to splitting: if x='', then $x is different than "$x", even if IFS=''.

You could also use zsh which doesn't do word splitting, but does "empty elision". (I'd say zsh is a better interactive shell than a scripting language, but I've seen it used on occasion.)

Or you can use OSH, which will run your bash script as is. And then you can opt into shopt -s simple_word_eval or the group shopt -s oil:basic, and EVERYTHING will be "unmolested", quotes or not:

http://www.oilshell.org/blog/2021/04/simple-word-eval.html

Just be careful to quote filename variables. Don't write this:

    cp $FILENAME $TARGET

Write this:

    cp "$FILENAME" "$TARGET"

It's a bit annoying, but it's not that bad, and certainly not unmaintainable.

the true (and unpopular) solution is to disallow spaces in filenames

I strongly recommend the book Unix Power Tools for all such questions. Quotes become easier to grok once you think of different ways to remove meaning of special characters. Basically, 1) backslash removes special meaning of the next character, 2) surrounding by double quote removes special meaning of all characters except backslash, dollar, and backtick, and 3) surrounding by single quotes removes all special meanings.

Coffee is for closers^W quoters. ABQ.

Always. Be. Quoting.

Also, you should be checking your Bash scripts with shellcheck, which will tell you when to do this with ~99% accuracy.

(For the uninitiated: https://www.youtube.com/watch?v=GrhSLf0I-HM)

I'm afraid quotes are the way to go. The alternatives are basically escaping (which looks even worse) or using a different language. The book also mentions a few more tricks you probably want to know about for dealing with the really tricky stuff, like filenames starting with a hyphen or ending with a newline.