Double safety check for Xorshift bit values.

[WebDrake]

Add a little bit of extra robustness/future-proofing to the design of Xorshift. It's a check that should have been in my previous pull request but that I overlooked.

[monarchdodra]

What i originally meant is that while double checks are more robust, if the second check fails, then it is a phobos internal error, and should read and be diagnosed as such: format("Phobos Error: %s bits not implemented", bits). Or format("Phobos Error: Failed to implement XOrShift for %s bits", bits).

As is written, the user wouldn't be able to tell if he made a wrong code call, or if the implementation made an error.

But this is mostly a rhetorical issue, since this is simple code. But as a rule of thumb, I think a lot more static asserts should simply read as Phobos Error.

[WebDrake]

I actually deliberately designed the messages to be different for each case, but they are subtle differences. We have 3 possible sources of error we need to catch:

• a user tries to instantiate an XorshiftEngine with an unsupported bit value -- Error I.
• a developer adds an extra bit value to the allowed list, but forgets to add instantiation for seeds_ - Error II.
• a developer adds an extra bit value and instantiation, but forgets to add an update rule in popFront() - Error III (this case).

So, I'll tweak all the error messages accordingly, and make use of the "Phobos Error" message you suggest.

[WebDrake]

Here you go. I didn't make use of format() or text() to create the error message string -- that can be corrected if you prefer.

[monarchdodra]

Here you go. I didn't make use of format() or text() to create the error message string -- that can be corrected if you prefer.

It's all CTFE, so it shouldn't really make any difference.

Looks good to me.

[WebDrake]

Tests passed ... ping? :-)

[repeatedly]

After current testing passed, I will merge.

[WebDrake]

Thanks! :-)