simplecpp's output is correct for most code, and it passes nearly all the macro examples
from C11 §6.10.3.5 and the C23/C++20 __VA_OPT__ examples. But the code doesn't follow
the translation phases in C §5.1.1.2 and C++ [lex.phases], and that causes bugs.
Comparing with gcc -E shows these differences:

• Line joining (phase 2) happens after tokenizing, so ab\<newline>cd becomes two tokens.
• Tokenizing (phase 3) splits the input into single characters and joins them again in
  combineOperators(). Numbers like 0x1e+e and 1.2.3 are split, and <=>, ->*, .*
  and digraphs aren't supported.
• #line is handled while the file is read. It takes effect inside #if 0 and does not
  expand macros.
• #elif is evaluated even after an earlier group was taken.
• #if arithmetic has no unsigned math (-1 > 0u is false) and no short-circuit
  (0 && (1/0) is an error). true is 0 in C++.
• ## is checked against a hand-written list: valid <<=, -> and && are rejected,
  and invalid .. is accepted.
• # escapes every backslash (str(\n) gives "\\n") and drops the space a comment
  stands for.
• Macro expansion approximates the standard rescanning rules with an expanded-macro set
  and special cases, instead of hide sets.

Proposed steps, each one a separate PR:

1. Small fixes: #line in the directive loop, skip #elif after a taken group, a new
   #if evaluator, fix # escaping.
2. Do line joining in the character stream, and replace combineOperators() with a
   longest-match tokenizer.
3. Move cppcheck-specific changes (raw-string conversion, &= split, #/## in plain
   text) to options applied to the output.
4. Rewrite macro expansion using Prosser's hide-set algorithm.

The full report with test cases is attached (preprocessing-report.md).

You may want to split it into one main issue for the restructure and separate small issues for the step 1 bugs (#line, #elif, #if arithmetic). Each of those can be fixed on its own.

preprocessing-report.md