[clang] [clang-repl] add %help, documentation, and tests for %commands (PR #150348)

[David Spickett via cfe-commits]

================
@@ -364,15 +366,34 @@ int main(int argc, const char **argv) {
       }
 
       Input += L;
+      // If we add more % commands, there should be better architecture than
+      // this.
       if (Input == R"(%quit)") {
         break;
       }
       if (Input == R"(%undo)") {
         if (auto Err = Interp->Undo())
           llvm::logAllUnhandledErrors(std::move(Err), llvm::errs(), "error: ");
+      } else if (Input == R"(%help)") {
+        OS << "%help\t\tlist clang-repl %commands\n"
----------------
DavidSpickett wrote:

I didn't think of that. That's a good question to be asking. I checked it:
```
/// This returns a reference to a raw_fd_ostream for standard output. Use it
/// like: outs() << "foo" << "bar";
LLVM_ABI raw_fd_ostream &outs();
```
And 1: the example uses outs() directly, so that's a big hint that it's fine. 2: It does create a raw_fd_ostream each time, but it always points to stdout. So it's not going to be doing much work constructing the wrapper.

There is one instance of someone doing `= llvm::outs();` outside a loop and reusing it, but there are plenty more where they don't.

But like you say, it's only going to be used for `%help` so if you're doing that, you're at an interactive prompt, and the performance there is not a problem or if it is, there are bigger issues than a few wrapper objects.

Again though, good thing to be thinking about. C++ especially can be very sneaky.

https://github.com/llvm/llvm-project/pull/150348