Bash quotes unescaped on command substitution

Question

Can someone explain me why this is working (listing the contents of a directory that has a whitespace in its name):

ret="$(ls "my dir")"

Shouldn't it rather be interpreted as:

ret="$(ls "
my dir
")"

As for example with:

ls "my" "dir"

and

ls "my dir"

Answer 1

The best guess so far seems to be that you are asking:

The command

    ret="$(ls "my dir")"

is a pretty terrible example of a command that contains four double-quote (") marks.  One might naïvely expect that the second quote would match the first one, and the fourth would match the third, like this:

    ret="$(ls "my dir")"
        ↑-----↑      ↑-↑

as, for example, happens with

    eat "afternoon snack" dinner "midnight snack"

Instead, it “works”, as follows:

    ret="$(ls "my dir")"
        ↑     ↑------↑ ↑
        ↑--------------↑

Why are the nested quotes parsed that way?

This paragraph of bash(1) may explain it:

Command Substitution

                    ⋮
…  When using the $(command) form, all characters between the parentheses make up the command ….

I guess this means what you have observed, namely, the text between the $( and the ) is treated as if it were a separate command, with its own parsing/pairing of quotes, independent of syntactic elements outside the parentheses.  I concede that, if you aren’t looking for that meaning, it’s very hard to interpret.

---

But your question, as written, doesn’t make sense.  The whole thing hinges on the very special meaning(s) of $ to the shell.  If we changed that to some other currency symbol,

    ret="£(ls "my dir")"

then it would be treated as two words:

        ret="£(ls "my
_{(followed by)}
                                    dir")"

because, if a non-blank, non-special character appears immediately before an opening quote or immediately after a closing quote, it is included in the same string as the characters enclosed in the quotes.  For example,

    cat Wal"*"mart

looks for a file called Wal*mart, not a file called Wal, one called *, and one called mart — Wal"*"mart is equivalent to Wal\*mart and "Wal*mart".  Similarly, the following example shows four words:

    one" "two  buckle" my "shoe  /don\'t/be/a/" "/cadet  sqrt"(2)"
    ↑-------↑  ↑--------------↑  ↑--------------------↑  ↑-------↑

I don’t know where you got the three-line breakdown that you showed in your question.