Do you know is the track length as written on the CD box likely to be the actual track length in milliseconds truncated to the nearest second or rounded to the nearest second
i.e.
would 3 minutes, thirty seconds and 600 milliseconds be shown as 3:30 (truncated) or 3:31 (rounded to nearest second, or rounded up always)
would 3 minutes, thirty seconds and 300 milliseconds be shown as 3:30 (truncated, or rounded to nearest second) or 3:31 (rounded up always)
and same question for Vinyl releases ?
I would assume it will rarely be that well thought out. Different publishers over the decades will have their own rounding up\down\close enough thoughts when they write the track times down. Hence why the actual times from the CD media are the only ones to really be trusted.
Actually if you are using Release Editor to add a release you cannot enter an accuracy to more than a second so if you have the accuracy to milliseconds what approach should you enter in MusicBrainz (truncated or rounded to nearest) ?
I guess I am fairly common example. I don’t manually edit track times. I generally use Picard to take times directly from the CD and upload the CD data directly.
The only time I would type numbers is if I didn’t have that actual CD. Then I would be either copying from the back of the Vinyl or CD cover and checking the times in my digital music player. All of these tend to just work to the nearest second.
If I was going to go into a full on pedant mode - I would use normal Maths rounding rules and “round to nearest”.
My brain starts churning into overtime on this one. 
Truncating the times seem illogical to me. As this is a music track it would take up space somewhere on tape\CD\whatever so would be more useful to be a rounded up figure so as to always make sure there was enough space on the medium.
Of more interest to me would be the fact there are gaps around tracks… so where are the gap times added?
Questions like that mean I try and just stick with what the digital media says is the length as I am mainly working with CDs and FLAC or MP3 files.
Yes, but if just displaying to the second then that depends on the software you are using, i.e does the software truncate or roundup the value before displaying
To get an answer to that you’ll need to ask MediaInfo, VLC, KODI, Media Player Classic and the rest… but that is a totally different question to your original one.
I think the answer is “Yes” 
And also the third option applies: None of the two. I have seen it often that track lengths written on the releases are more than a second of. The reason is probably that layout and marketing material is created before the final CD mastering happened.
The right answer is it depends. I have seen both. From the top of my head I don’t even know how Picard does it. What does your software do?
I was just pointing out to you that this approach can be inconsistent, i.e if you view your digital media in different tools you can get different results.
Good question, it gets the value from the underlying jaudiotagger library and this rounds to the nearest second. Having said that, at one point there was certain audio formats that jaudiotagger was truncating rather than rounding.
I suppose more to the point is what is the MusicBrainz style on this ?
If you don’t have the Disc (so the tracks are not being added based on discid) but you do have digital versions of the file so you know long they are (at least your versions) should you enter the track length truncated or rounded to nearest second ?
I’ve never given it any thought before but I know I’ve used track timings from my media player for digital releases rather than the timings shown on amazon which were different. FWIW, I use foobar2000 which rounds the values and that makes the most sense to me.
edit: in case anyone is wondering, the “Properties” dialog for a track shows the actual length to several decimal places compared the rounded length you’d see in the playlist view.
This question is not really important for CD as we eventually submit the Disc ID. Then MB stores milliseconds and displays proper rounded value, like foobar2000.
For what it’s worth, I have a macro to convert freedb.org entries to proper rounded values (otherwise freedb.org displays floored values, cut-off milliseconds).
For analogue releases (cassette, LP, etc.) I think it is clever to just copy what is printed. Because there are many ways to rip them, the offset will rarely be the same from one rip to another.
For DVD, I rip it with chapter split and foobar2000 gives me round values until we can actually submit milliseconds (which would also be useful for freedb.org entries).
For Blu-ray, as I cannot rip them, I compute track lengths from chapter indexes, which is only second accurate, not millisecond.
True, but it is important (well may be important is pushing it) until that time comes. Since MusicBrainz displays rounded values then I guess it best practise to enter rounded rather than truncated values as well.
You can use release editor seeding to enter sub-second accuracy track times. E.g., using the Bandcamp importer userscript, spotify2musicbrainz, the Picard “Add Cluster As Release” plugin, … the releases added will have millisecond precision track lengths.
You often have a couple of seconds between recordings which are absorbed into the track times, so when you use Picard to import times directly from the CD, you end up with the time for a track being slightly different to the official time for the recording which is printed on the CD.
This can cause an issue with compilations having different track times for the same recording(but that is probably another topic in itself).
Eh. It’s been working for forever. If there was a period where it wasn’t working, then that was a bug (that may or may not have been fixed). It’s been working (or intended to work) for at least 4 years.
In the “olden days” my cd player would play a track and would then give a 3 second countdown to the next track.
Mathematically, a CD length would be 30 seconds longer than the sum of all of the posted lengths. Plus the difference of the rounding of fractional seconds over multiple tracks.
And in the “modern days”, for digital tracks, most of my programs give the option of adding 2 seconds of silence.
That is because the CD you had playing had a three-second pregap between the tracks (index 0 of the next track was three seconds before the index 1). This is still very common today.
I find it unfortunate that this information is usually lost on the ripping process (this is the reason I always rip CDs to full album files with cue sheets that save this), for example, MusicBrainz disc IDs only take the index 1’s into account.
There are three different methods you could use to calculate the track duration that all make sense: prepend, remove and append pregap. Out of which, only “append” is possible without knowing where the index 0’s are while “remove” is a quite often used metric next to the track names on backs of releases. The remaining one, “prepend”, would also make sense, as the pregap is technically a part of the next track.
There can also be indices higher than 1, up to 100. These would also be similarly lost if not specifically trying to detect them. (I have never seen these)
I happened to analyse this disc’s track durations in detail: Edit #56021058 - Set track lengths
Very few CDs have them, because very few players support them, because very few CDs have them, because very few CDs have a use for them. While I have seen them perhaps a dozen times over the years, I know offhand of only one that has indexes > 1: The Beavis and Butt-head Experience
