Iterating matches, and a manual that has fallen behind

Matching one string is the simple case. Most real uses walk an input, and CTRE has three ways to do that which are easy to confuse.

Start with the documentation problem

The readthedocs page teaches ctre::range and describes range support as preliminary. Compile that today and the compiler disagrees:

warning: 'ctre::range<...>' is deprecated: use search_all

The same documentation carries a literal TODO more detailed regex information where the syntax reference should be. Neither of those is a criticism of the library, which is under active development, but it does set the working rule for this series: read the header, not the manual. Everything below was checked against a compiler rather than a page.

search_all

Every match of the pattern, skipping whatever lies between:

for (auto m : ctre::search_all<"([0-9]+)">("123,456,768"sv)) {
    m.get<1>();   // 123, then 456, then 768
}

split

The pieces between matches of a separator, which is the inverse view of the same walk:

for (auto piece : ctre::split<",">("alpha,beta,gamma"sv)) {
    // alpha, beta, gamma
}

tokenize, and the distinction that matters

tokenize consumes the input from the front, match after match, and stops at the first position where the pattern does not match. That sounds like a smaller search_all and behaves like a different tool. The same pattern over the same input:

Input ab!!cd	Result
search_all<"[a-z]+">	[ab] [cd]
tokenize<"[a-z]+">	[ab]

search_all treats unmatched input as something to skip past. tokenize treats it as the end of the road.

For scanning a log line for every timestamp, skipping is correct and search_all is the tool. For lexing a language, unmatched input is a syntax error, and silently skipping it turns a broken program into a subtly different valid one. That is why the lexer in the next episode is built on tokenize: stopping is the feature.

What these give you as ranges

All three are usable in a range-for and compose with the standard range machinery, so the ordinary vocabulary applies:

auto numbers = ctre::search_all<"([0-9]+)">(input)
             | std::views::transform([](auto m) { return m.get<1>().to_view(); });

Which is the quiet payoff of the library being ordinary C++ rather than a separate engine: the matches are just a range, and everything that works on ranges works on them.

---

Sources: CTRE · CTRE documentation