The thing I wished I had learned earlier is "quick and dirty assertions". If you write lots of functions in Bash, you quickly end up getting tripped up by cases where an argument is omitted and the function does something totally batshit given the missing (empty string) argument. Now, the canonical way to handle this is to put validators on your input, (and make sure those validators don't crash with cryptic errors if someone calling your function is using "set -u") like so:
function() myfunc {
local foo="${1:-}"
if [ -z "$foo" ]; then
echo "Invalid first parameter!" >&2
return 127
fi
...
}
...but man, that's time consuming when you have lots of parameters.
function myfunc() {
local foo="${1:?First parameter must be provided}"
...
}
Much quicker, especially when throwing things together in a hurry. It has a gotcha, though: ":?" assertion doesn't cause a function to return early, it shuts down the whole interpreter after outputting the error. So it's more like a true assert() statement than an input validator. If you'd only ever call your function in a subshell, this won't matter (because the subshell will exit early with a nonzero code, big deal), but otherwise it can be a nasty surprise to users when an argument-validation issue inside a function shuts the program down. Then again, the "return 127" in the first example would also shut the program down if someone was using "set -e".
...and while we're on the subject of "set -e", I think that the ["unofficial Bash strict mode"](http://redsymbol.net/articles/unofficial-bash-strict-mode/) (putting "set -euo pipefail" and "IFS=$'\n\t'" at the top of your scripts) has been a bigger bug-prevention/rapid development aide to me than anything else. To be clear, I think it's a means of detecting some kinds of bugs. I've read Wooledge and others' objections to those patterns, especially "set -e", and agree with the point that this does not make your programs objectively safer and shouldn't be counted on as a crutch. Then again, neither does a linter, but it still helps you detect and avoid some kinds of bugs, so why not use it?
as a precondition. I must admit that it's longer to write, but I can write a bunch of preconditions for my function then write the logic with an appeased mind
Instead, the quick and dirty way is to just "assert" via [parameter expansion](https://www.gnu.org/software/bash/manual/html_node/Shell-Par...):
Much quicker, especially when throwing things together in a hurry. It has a gotcha, though: ":?" assertion doesn't cause a function to return early, it shuts down the whole interpreter after outputting the error. So it's more like a true assert() statement than an input validator. If you'd only ever call your function in a subshell, this won't matter (because the subshell will exit early with a nonzero code, big deal), but otherwise it can be a nasty surprise to users when an argument-validation issue inside a function shuts the program down. Then again, the "return 127" in the first example would also shut the program down if someone was using "set -e"....and while we're on the subject of "set -e", I think that the ["unofficial Bash strict mode"](http://redsymbol.net/articles/unofficial-bash-strict-mode/) (putting "set -euo pipefail" and "IFS=$'\n\t'" at the top of your scripts) has been a bigger bug-prevention/rapid development aide to me than anything else. To be clear, I think it's a means of detecting some kinds of bugs. I've read Wooledge and others' objections to those patterns, especially "set -e", and agree with the point that this does not make your programs objectively safer and shouldn't be counted on as a crutch. Then again, neither does a linter, but it still helps you detect and avoid some kinds of bugs, so why not use it?