Use iostream::fail() to check for I/O error

[jhenin]

Fixes #521

[giacomofiorin]

Looking again at this change and the related issue, it is really not self-apparent why one should use fail() instead of bad(). Our use case has always been "test if there are any errors (no matter of what kind) and then read the stream".

So the most intuitive check is probably just using the bool cast operator, and testing the stream object directly (I believe this was also @HanatoK's suggestion earlier on).

Approving because it's a bugfix, but my preference would be for simpler semantics here, i.e. either just use the bool operator throughout or (as a less elegant option) set the badbit ourselves inside colvarproxy::input_stream() before returning it to the caller.

[jhenin]

it is really not self-apparent why one should use fail() instead of bad(). Our use case has always been "test if there are any errors (no matter of what kind) and then read the stream".

The intended use of bad() is not very clear to me, that is, based on the name I would expect bad() to be the broadest way of catching errors. But I've accepted that it isn't :-)

I did consider using the bool cast, which is semantically equivalent to fail(), and thought that fail() was more explicit and slightly easier to read, but I'm not strongly opinionated on this. I'm ok with testing the stream objects themselves.

[giacomofiorin]

I did consider using the bool cast, which is semantically equivalent to fail(), and thought that fail() was more explicit and slightly easier to read, but I'm not strongly opinionated on this. I'm ok with testing the stream objects themselves.

Indeed fail() is easier to read: what's not easy at all is distinguishing it from bad(), nor is it accepting the fact that (! good()) != bad().

Ok, can you please change to bool and then merge?

[jhenin]

accepting the fact that (! good()) != bad().

Yup, that's f-ed up. Thank you C++.