--help and -h are automatically supported by the command-line parser, unless overridden by the tool. The behaviour of the PrintHelpMessage being used for -h prior to this patch is subtly different to that provided by --help automatically (it omits certain elements of help text and options, such as --help-list), so overriding the default is not desirable, without good reason. This patch removes the explicit specification of -h and its behaviour, so that the default behaviour is used.
Thanks, I wasn't aware of that one (and hadn't yet bothered running every lit test for this). Just to be clear, are you recommending that I just delete the bits of that test relating to llvm-dwarfdump, since it will now follow the default process, or do you think there should be something else to replace it? I can replace it with llvm-opt-report, which seems to do something similar to llvm-dwarfdump.
Yes, it can be removed. If you want, you can add it to the ones above that do use the new default -h, but I don't think that's necessary. And yes, if there's another one that does something special, adding it, i.e., s/llvm-dwarfdump/llvm-opt-report/ makes sense.
Okay, I'm going to remove it and then commit, without replacing the removed test part. My initial evaluation was using an out-of-date llvm-opt-report, and I was unable to find any other suitable candidates. I reviewed the other tools, and the only tools I could find with explicit -h help handling are dsymutil and clang-offload-bundler. In both cases, I think the existing handling is a bug (in dsymutil at least, you actually have to provide a positional argument before it will process '-h'!).
That's because they're processing -h after cl::ParseCommandLineOptions returns. But as you say, InputFiles requires at least one argument, so cl::ParseCommandLineOptions never does return if this condition isn't met. The only way to fix that is to remove the special -h processing.
static list<std::string> InputFiles(Positional, OneOrMore, desc("<input files>"), cat(DsymCategory));