RFC: get panics out of std::ffi::CString

[mzabaluev]

Bring CString in line with other library functions on the failure mode conventions.
There should be no panic lurking.

Rendered

[mzabaluev]

Cc @alexcrichton @aturon

[kennytm]

Alternative: Add a checked_from_slice() method, just like we have x/y which may panic and x.checked_div(y) which does not.

[mzabaluev]

@kennytm I have added a paragraph on why simply adding functions is worse than changing the "shortest name" API.

In case of division, a zero divisor arguably is a programmer error. A restriction on NULs in a string, on the other hand, is arbitrary for the value domain in general (a Unicode text with NUL characters is valid). There are also performance concerns regarding the implementation of division.

[mzabaluev]

That said, a migration path is probably needed whereby the Result-returning functions are added under temporary names and the current functions deprecated. Suggestions?

[aturon]

Thanks so much for throwing this together! I'm increasingly in favor of this approach for C strings.

cc @wycats @nikomatsakis @brson @huonw

@pcwalton I know CString is somewhat common in Servo, do you all have a preference here?

[huonw]

Seems fine to me.

[pcwalton]

@aturon +1 from me.

[aturon]

After discussion here and on discuss, we've decided to move forward with this long-requested change. Thanks again, @mzabaluev! We'll try to land this ahead of alpha2 if possible.

Tracking issue