用 PCRE 提取“表达式语法”

Question

We're working on a project where we support extrapolation expressions from strings. Under the hood we're using Symfony's Expression Language to provide context parsing, but we're the ones extracting expressions from strings.

I would like to preface this by saying, I am no expert at regular expressions. My working knowledge is limited, and so the following regex will appear clunky and inelegant:

/\${(.*?)}(?=[\s\w\-_\/\\:;,.?!()|"\]&]|$)/

The theory is thus:

1. An expression starts with ${ . This is the starting anchor.
2. Match anything in there.
3. The expression ends with a closing } that is followed by either a line end $ , or one of the items from the character list.

Consider an expression that looks like this:

His name is " ${name} ", and he's a " ${thing} ".

The regex will successfully match the expressions name and thing , and will replace those with values from a value object.

However, if we take into account that users can also parse actual expressions and values, given this:

${{"name": "Pack Rat", "mana_cost": "{1}{B}", "cmc": 2}}

Meaning, evaluate that expression to a JSON object, the regex fails because it stops at the }" sequence in the part {1}{B} , and matching only {"name": "Pack Rat", "mana_cost": "{1}{B . Removing " as a possible stopping point in the lookahead character list fixes the JSON, but then it fails to extract the two expressions from the regular sentence.

Would it be possible to avoid premature stopping of this expression parser? Or is this something that is beyond the scope of a single regular expression?

Answer 1

You could use

\$(\{(?:[^{}]+|(?1))+\})

And use this to further analyze. See a demo on regex101.com .

---

In detail, this reads:

\$                       # "$" literally
(                        # opening bracket -> capture group 1
    \{                   # "{" literally
        (?:[^{}]+|(?1))+ # not { nor } or repeat the first group -> recursion
    \}                   # "}" 
)                        # end of group 1