REGEX PCRE Recursive expression for nested text matching

Here's a pattern that matches your example:

^\h*START\h*\n(?:\h*+(?!(?:START|STOP)\h*$)[^\n]*\n|(?R)\n)*\h*STOP\h*$

using the /mg flags (live at https://regex101.com/r/iK9tK5/1).

The idea behind it:

^                                  # beginning of line
\h* START \h* \n                   # "START" optionally surrounded by horizontal whitespace
                                   #   on a line of its own
(?:                                # between START/STOP, every line is either "normal"
                                   #   or a recursive START/STOP block
    \h*+                           # a normal line starts with optional horizontal whitespace
    (?!                            #   ... not followed by ...
        (?: START | STOP ) \h* $   #   "START" or "STOP" on their own
    )
    [^\n]* \n                      # any characters, then a newline
|
    (?R) \n                        # otherwise it's a recursive START/STOP block
)*                                 # we can have as many items as we want between START/STOP
\h* STOP \h*                       # "STOP" optionally surrounded by horizontal whitespace
$                                  # end of line

I've made \h*+ possessive in order to avoid accidentally matching " STOP" by 0 iterations of \h*, not followed by "STOP" (they're followed by " STOP" (with a space)). The + forces \h to match as many times as it possibly can, so it has to consume the space.

Alternatively you could pull \h* into the look-ahead: (?!\h*(?:START|STOP)\h*$)
That would also work, but then the look-ahead would skip over any spaces to see whether they're followed by START/STOP, only to have [^\n]* outside go over those same spaces again. With \h*+ at the start, we match those spaces once, with no backtracking. I guess it's a micro-optimization.