Fix as.exe crash on Windows

[grendello]

Code previously used to convert command-line arguments from wide encoding to UTF-8
we used when as.exe was built with MinGW, is no longer necessary. It was causing
the app to crash on Windows.

[grendello]

Doesn't crash, but no longer finds the wide-char filenames... search continues.

[jonpryor]

This actually worries me; the "full expression" is this entire line, so once executable_path.string().c_str() has finished evaluation, it is potentially invalidated, meaning that epath may be garbage.

I think it would be safer -- and would also fix the -Wdangling-gsl warning -- if you instead did auto path = executable_path.string(), and then everywhere you had epath you instead used path.c_str(). Then we know the value returned from .c_str() won't be destructed until path is destructed, which is at end of scope, vs "end of the full expression".

[grendello]

That means code duplication and doesn't really change anything though... It's a cosmetic change with no real effect - the temporary is still removed and we still duplicate the C string we get from the temporary

[jonpryor]

Yes, but that's later. My recollection is that this full expression is basically:

auto __tmp = executable_path.string();
const char *const epath = __tmp.c_str();

which looks fine, except for the all important question: when is __tmp destructed? At various points in time this was up to the compiler, and some compilers would destruct __tmp at end of enclosing scope (i.e. "you're fine") and some would destruct __tmp at end of expression, i.e.:

const char *const epath;
{ auto __tmp = executable_path.string();
  epath = __tmp.c_str();
  /* __tmp destructed at end of full expression */ }
/* `epath` now pointing to garbage */

Perhaps the semantics have since been standardized, but at least in C++98 this was not well specified, and different compilers would do different things. The warning is trying to warn you about this.

[grendello]

The local object is strictly destroyed when it leaves its scope, in this case the function.

[jonpryor]

I'm confused wrt how this actually transforms to UTF-8 on Windows, when platform::string is std::wstring. This looks like it'll just truncate wchar_t to char.

[grendello]

platform::string is now uniformly std::string, I stopped using wstring but kept the abstraction "just in case"