r4925 - why not use exact float for NTSC?

[taltamir]

But it is truncated. The float type has 24 bits precision, or about 7 decimal digits. Anything longer is truncated.

Ah, i see what you mean now. 59.94005994005994005994005994006 isn't truncated. But if we try to input it into "float" (the command) it will get truncated because a float is a 24bit floating decimal number which can only store so many significant digits.

What about using a double then? it should be good for 52 significand and the number above is only 31.

[avih]

I haven't looked at the code, but my guess is that increasing accuracy by 3 orders of magnitude (from 59.94) seemed like a good balance between small enough change for big enough gain. Double might have been better, and actual nominator/denominator would have been best, and yet, that's how gigahertz decided to change it

And it is good enough. Recall it was 60 before these changes, then eventually changed from 2 digits precision to 7 digits, making it 100K times more accurate than it originally was. The implications of the original inaccuracy was 1 second A/V drift every 15 minutes. 59.94 would have been 1s every 1500 minutes, and now it's 1 second every 1.5 million minutes of recording. That's good enough in my book, even if ultimately not 100(.000000000000000000000000)% accurate

Yes, Computer Science should be an accurate science, but in practice that's not always the case

[taltamir]

And it is good enough. Recall it was 60 before these changes, then eventually changed from 2 digits precision to 7 digits, making it 100K times more accurate than it originally was. The implications of the original inaccuracy was 1 second A/V drift every 15 minutes. 59.94 would have been 1s every 1500 minutes, and now it's 1 second every 1.5 million minutes of recording.

My understanding was that the current value is 59.94
What is the current value that has an inaccuracy of 1 second every 1.5 million minutes of recording?

[avih]

My understanding was that the current value is 59.94
What is the current value that has an inaccuracy of 1 second every 1.5 million minutes of recording?

Current value is (more of less) the best approximation at which a float can represent the result of (30/1.001)*2, which happens to be 59.940056

[taltamir]

Current value is (more of less) the best approximation at which a float can represent the result of (30/1.001)*2, which happens to be 59.940056

I see, thank you for clarifying.