Does `OFFSET` really correspond to the second measure's `fumenOffset`?

[vivaria]

Right now, my converter assumes that OFFSET directly corresponds to the second measure's fumenOffset. See:

• https://github.com/vivaria/tja2fumen/issues/4#issuecomment-1575248722

While this holds for the majority of cases, sometimes an ESE song will be off by anywhere between 0-25ms. Normally, I just correct the TJA file so that it matches the official fumen's OFFSET value.

But, what if I should be leaving the TJA offset values as-is? ((There may be some interaction with #BPMCHANGE commands that cause OFFSET to not line up with the second measure's fumenOffset.))

So, I will need to see how these "wrong OFFSET" songs play on TJAPlayers, and compare that to how they play on TDMX, to make sure that my fumenOffset calculations are correct.

[vivaria]

Does OFFSET really correspond to the second measure's fumenOffset?

ANS: No. Very much no.

sometimes an ESE song will be off by anywhere between 0-25ms.

While it was difficult to pinpoint before due to the small desync times, I've found a song that is now way off: SUPERNOVA.

• The OFFSET of this song is -0.129
• The existing tja2bin.exe converts it flawlessly, with Ura fumenOffset values -1003, -720, 1544 for the first three measures
• tja2fumen.exe instead generates Ura fumenOffset values of -154, 129.0, 2393 (note the assumption here that OFFSET == second measure)

The measure durations (i.e. the time between each fumenOffset) are identical. It's just the offset as a whole that's way off. So, my assumptions around OFFSET are definitely off.

Here is the relevant part of the Ura chart:

#BARLINEOFF
#MEASURE 1/4
580000,

#BARLINEON
#MEASURE 4/4
40040040,
,

My hunch is that it has to do with the #MEASURE commands, as butou5.tja was off a little too before I corrected the .tja file, and its first measures include #MEASURE commands as well.

Maybe OFFSET corresponds to the fumenOffset value of the first full (4/4) measure. But, since the second measure of the chart occurs only after 1/4 of a measure has passed, then the offset values are way off..

[vivaria]

Maybe OFFSET corresponds to the fumenOffset value of the first full (4/4) measure. But, since the second measure of the chart occurs only after 1/4 of a measure has passed, then the offset values are way off..

This is exactly it.

If I subtract by an extra 3/4 of a measure (i.e. 3/4 * 1132), then I get the "correct" fumenOffset values (-1003, -720, 1544).

So, in other words:

# Formula
fumenOffset[2] = OFFSET - (measureDuration * (1 - time_sig[1]))

# In practice:
fumenOffset[2] = 129 - (1132 * (1 - 1/4))
fumenOffset[2] = 129 - 849
fumenOffset[2] = -720

And, because most of the time time_sig[1] is 4/4, then there's no adjustment to the OFFSET value.