Release trains 🚆?

[martinbonnin]

Hi everyone 👋

The September 2025 edition of the spec is now out 🎉

Before that, there was a 4 years hiatus on spec releases which means users have to either wait and/or rely on the draft.

At GraphQLConf, the idea was floatted to do release trains. This would encourage maintainers participation by giving them visibility into when their contributions can be released. This would also help users anticipate when changes are coming.

The tradeoff is vs administrative work and more noise.

Thoughts?

[benjie]

I'm unsure what you're proposing, could you be more explicit?

[martinbonnin]

I'd suggest we use a fixed schedule where we cut a release of the spec (this is in comparison to train schedules but for software, I have just added the link to the definition to the original post).

An example could be once every year before GraphQLConf. But it could be every 6 months or every 2 years if we decide to. Other projects using this are Java, Kotlin, Ubuntu, etc...

Everything that is ready to ship goes in. If a feature misses the train, it'll take the next one (meaning waiting one year but it's still better than 4, edit: or maybe it's not if we favor stability, this issue is for discussing that).

[benjie]

Oh right; we agreed to do ~yearly releases at GraphQLConf, the aim is to do a cut just before the conference each year rather than being on a fixed schedule (since the conference itself is not on a fixed schedule). Are you proposing always doing a cut in September, no matter what, instead?

[martinbonnin]

we agreed to do ~yearly releases at GraphQLConf

Nice, I wasn't sure where was the decision here so I guess this issue can serve as a decision record.

since the conference itself is not on a fixed schedule

Ah yes, forgot about that.

Are you proposing always doing a cut in September, no matter what, instead?

Excellent question, I'm not sure. There is some value in doing it before the conference so I think I'd go with that indeed. Even if there is no fixed schedule for GraphQLConf, I don't expect GraphQLConf schedule to become completely erratic in all cases.