QueryPerformanceFrequency failure mode not handled

[graydon]

In the runtime, in timer.cpp, timer::timer, failure of QueryPerformanceFrequency is not handled. Find out if it's possible for it to fail and, if so, get a workaround in place. Or failure. One or the other.

[lifthrasiir]

Quick research reveals the following:

• QueryPerformanceFrequency is supported since Windows 2000, so if we are ignoring Windows 9x we don't need a conditional compilation.
• QueryPerformanceFrequency always defaults to GetTickCount. So it wouldn't fail even in the absence of high resolution timer.
• QueryPerformanceFrequency rarely fails. The alternative implementation of Wine does not fail at all. The actual Windows API seems to fail with an misaligned pointer, but that can be made impossible by ensuring proper alignment (which Rust should do anyways).

[metajack]

Is this really a production readiness thing? Seems internal. Nominating for discussion.

[msullivan]

If we think that this can't actually fail, then we should add an assert.

[graydon]

just a bug, removing milestone/nomination.

[pnkfelix]

Visiting for triage, email from 2013 sep 30

@msullivan you mean to assert that the input address we feed to QueryPerformanceFrequency is quad-word aligned? Or do you mean we should assert that the value stored in ticks_per_s after it returns is non-zero? (Or both?)

[brson]

Same for QueryPerformanceCounter. Let's just assert that the functions succeed.