hidraw: Implement hid_error()

[z3ntu]

Here is my first try at implementing error messages for the Linux/hidraw implementation.

• I used a thread-local variable for non-device error messages.
• Non-device errors can be obtained by passing NULL into hid_error as I suggested in libusb: allow hidapi to be configured without using iconv #123.
• I belive to have covered all system calls that could fail; please check this.

A few extra questions:

• I am not sure why in utf8_to_wchar_t the input is called utf8. As far as I know, the encoding used in mbstowcs depends on the current locale, if if I haven't set my locale to use UTF8, then the input is not UTF8.
• The register_error equivalent in windows/hid.c doesn't use its argument const char *op. Is that intended?
• In the Windows implementation, do functions like HidD_GetPreparsedData actually set the error code for GetLastError()?

Originally signal11/hidapi#125

[Youw]

When a client code gets an "error string" as char* (or, in our case wchar_t*), it likely assumes, that that is a valid c-string, and for instance try to:

printf("Last HIDAPI call status: %S", hide_error());

and that would immediately lead to crash, since hide_error() sometime returns NULL.

I'd suggest always return a valid c-string (non-NULL) instead.
If there is no error, we might return "" or "Success", or smth like that.

[Youw]

my suggestion stands

@z3ntu if you accept - please update PR accordingly, otherwise give your note why you wouldn't do it within this PR; maybe we'll have a separate task for it

[z3ntu]

Yes, will update it. I've just been travelling a lot in the past couple of days and spending time on other projects inbetween.

Do you think it's better that if the register_global_error function receives a NULL argument, that it sets the last_global_error_str variable to Success or that the register_global_error function should be called with parameter Success? I'm personally for the first option.

[Youw]

• first looks OK
• register_global_success is an alternative
• hid_error could return "Success" if last_global_error_str is NULL

to be honest I'm more concerned of usage on the user side, and I'll support any reasonable good implementation, so it is up to you to decide
(and yes, I think passing "Success" in each place wouldn't look very nice)

[z3ntu]

The last sounds the best imo

[z3ntu]

Should be good for a review now

[Youw]

@jdk, @todbot, @Qbicz your comments?

[Youw]

Looks like documentation has to be updated(

[Youw]

Thanks to comment from @canselcik, we see, that the part about NULL isn't true.

[Youw]

And it is not thread-safe, when you're using same device from different threads.