That depends entirely on 1) what grammar you're parsing, 2) what memory pressure you're in, 3) how fast you want to do it, and 4) how comprehensive you want the error handling. It's actually quite difficult to parse C++ on a low-memory embedded device without resorting to clever techniques.
Anyway, C++ is just a strawman that seems to have been contrived deliberately in order to wreck most of the rational arguments defending "programming languages are easy to parse".
Parsing as programming language is exactly as difficult as the hole we dig ourselves into. Programming language parsing isn't some information processing problem that exists in the world, or in another domain, that computer science is tasked with to solve; it's a problem that computer scientists and software people inflict on themselves.
By making languages hard to parse by machine, we are also making them hard to use by humans. Humans need to parse the language also, and not have a magic oracle for doing so. (Indentation only goes so far, and can tell a lie).
Still, there aren't actually difficult problems in parsing C++. It's more of a "death by a thousand blows". There is just the sheer size of it alone. The C++ parser source file in GCC is larger than some programming language implementations: their parsers, interpreters/compilers and all intrinsic routines combined. It's hard to comprehend the C++ grammar so that you can identify where the ambiguities lie, and form a strategy to deal with them, so as not to be surprised later.
This is a specific problem of C++ though, given the fact that its grammar is extremely complex.
Clang aims to support very general transformation of C++ with their parser and AST. I was shocked to learn that Clang's AST has 60K lines of header files! That's not including the code, and it's not templates! That's class definitions!!! (wc -l include/clang/AST/*.h)
> It's actually quite difficult to parse C++ on a low-memory embedded device without resorting to clever techniques.
Compiling C++ programs on low-memory embedded devices is not a mainstream use-case. Unless you can demonstrate that it is, your point is moot.