Randomizing text between delimiters

Going non-greedy is a good thought, but doesn't quite do the trick. And you can add a loop:

perl -plE 'while(s!\{([^{}]*)\}!@x=split/;/,$1;$x[rand@x]!ge){}'

Notice that your sample input has unmatched braces, so this appears to output a spurious '}'

Nice challenge. What you need to do is to find a set of braces without interior braces, and pick a random item from in there. You need to do that globally. That will replace just the "level 1" braces. You need to loop over the string until no more matches are found.

use v5.18;use strict;use warnings;sub rand_sentence {    my $copy = shift;    1 while $copy =~ s{ \{ ([^{}]+) \} }                       { my @words = split /;/, $1; $words[rand @words] }xsge;    return $copy;}my $str = 'I have { {red;green;orange} fruit ; cup of {tea;coffee;juice} } and {nice;fresh} {sandwich;burger}.';say rand_sentence($str);say '';$str = <<'END';This {beautiful;perfect} {image;photography}, captured with the { {NASA;ESA}Hubble Telescope ; {NASA;ESA} Hubble Space Telescope }, is the{largest;sharpest} image ever taken of the Andromeda galaxy { {— otherwiseknown as M31;— known as M31}; [empty here] }. This is a cropped version of thefull image and has 1.5 billion pixels. { You would need more than {600;700;800}HD television screens to display the whole image. ; If you want to display thewhole image, you need to download more than {1;2} Tb.  traffic and use 800 HDdisplays }ENDsay rand_sentence($str);

sample output

I have  orange fruit  and fresh sandwich.This beautiful photography, captured with the  ESA Hubble Space Telescope , is thelargest image ever taken of the Andromeda galaxy  — otherwiseknown as M31. This is a cropped version of thefull image and has 1.5 billion pixels.  If you want to display thewhole image, you need to download more than 1 Tb.  traffic and use 800 HDdisplays

TXR solution. There are many ways to approach this.

Let's assume we're reading the data from standard input. How about we read the data in records which are delimited not by the usual newline character, but rather the braced-choices pattern? We do this by creating a record adapter object over the standard input stream. The third argument to the record-adapter function is a Boolean which indicates that we want to keep the terminating delimiter (the part which matches the record-delimiting regex).

Thus if the data looks like this foo bar {bra;ces} xyzzy {a;b;c} d\n it turns into these records: foo bar {bra;ces}, xyzzy {a;b;c} and d\n.

We then process these records as if they were lines of text using the extraction language. They fall into two patterns: lines which end in the braces pattern, and lines which don't. The latter are just echoed. The former are treated as required by the random brace substitution.

We also initialize *random-state* so that the PRNG is seeded to produce a different pseudo-random sequence on each run. If make-random-state is given no arguments, it creates a random state object initialized from system parameters like the process ID and system time:

@(do (set *random-state* (make-random-state)))@(next @(record-adapter #/{[\w;]+}/ *stdin* t))@(repeat)@  (cases)@*text{@switch}@    (do (put-string `@text@(first (shuffle (split-str switch ";")))`))@  (or)@text@    (do (put-string text))@  (end)@(end)

Test run:

$ cat dataI have {red;green;orange} fruit and cup of {tea;coffee;juice}.$ txr rndchoose.txr < dataI have red fruit and cup of tea.$ txr rndchoose.txr < dataI have orange fruit and cup of tea.$ txr rndchoose.txr < dataI have green fruit and cup of coffee.