Release 1.0

[maxammann]

It seems this crate can be thought of as "done" as it implements a spec. Are there any blockers that would need to be done before this can reach 1.0?

I think log and geo_types are both blockers because they are not 1.0 yet.

[urschrei]

The main question is whether we're happy with perf, which is related to the way we handle errors: #197, #176, #180

[michaelkirk]

I think log and geo_types are both blockers because they are not 1.0 yet.

Can you articulate why depending on a non-1.0 dependency blocks us from releasing a 1.0? I'm not aware of why it would.

[michaelkirk]

A bigger thing that I might be worried about before 1.0 is making the data types opaque. Currently they are all just aliases for Vec.

So that we can do things like: https://github.com/georust/geojson/pull/222/files#diff-b1a35a68f14e696205874893c07fd24fdb88882b47c23cc0e0c80a30c7d53759L427

[michaelkirk]

I think log and geo_types are both blockers because they are not 1.0 yet.

Can you articulate why depending on a non-1.0 dependency blocks us from releasing a 1.0? I'm not aware of why it would.

Firstly, semantic version lawyering is not my strong suit, but in a moment of synchronicity, I may have answered my own question, when it was asked differently in another context:

pka/http-range-client#8 (comment)

Since log isn't used in our public API anywhere, I think it's not relevant, so long as we trust the maintainers of log to not make backwards incompatible changes without bumping the 0.Y.0.

Technically anything goes when you're 0.y.z,

Major version zero (0.y.z) is for initial development. Anything MAY change at any time. The public API SHOULD NOT be considered stable.

But I'm comfortable betting on the log maintainers to bump 0.Y.0 when they make backwards incompatible changes, as they have for the last many years.

geo-types is a little tricker.

Since we're implementing traits on geo-types, it's part of our public API, so semver breakage in geo-types means breakage in geojson (right?). Although geo-types isn't yet 1.0, like log, it's relatively stable and we are careful about properly bumping geo-types 0.Y.0 whenever we make a change that is not backwards compatible.

So if we want to release geojson 1.0 before geo-types 1.0 (which we could), geojson would get an extra major version bump if/when geo-types finally ever gets to 1.0.

[maxammann]

Firstly, semantic version lawyering is not my strong suit, but in a moment of synchronicity, I may have answered my own question, when it was asked differently in another context:

pka/http-range-client#8 (comment)

Since log isn't used in our public API anywhere, I think it's not relevant, so long as we trust the maintainers of log to not make backwards incompatible changes without bumping the 0.Y.0.

Technically anything goes when you're 0.y.z,

I thought I would find some better list than this but I think the rule is that all your dependencies need to be 1.0, even if they are not public facing. I could be very wrong on this though. Imaging that log adds a rust-version tag to their Cargo.toml That would break compiling geosjon with the older Rust version. That would be a breaking (Cargo) change imo.

Releasing a 1.0 for log seems to have stalled: rust-lang/log#375

So if we want to release geojson 1.0 before geo-types 1.0 (which we could), geojson would get > an extra major version bump if/when geo-types finally ever gets to 1.0.

That could make sense yeah. I'm a bit mixed on the situation of depending on non stable crates in a stable crate. I'll try to find more information of the communities opinion on that.

[maxammann]

Found the reference! https://rust-lang.github.io/api-guidelines/necessities.html#c-stable

btw, the reason is that I want to create a crate based on this and though: Hey! this could be come stable pretty quickly!

After all, this migth be more a geo-types discussion, first and foremost :)

[michaelkirk]

Found the reference! https://rust-lang.github.io/api-guidelines/necessities.html#c-stable

My read of this is we don't need to worry about log, since it's not a public dependency.

Imaging that log adds a rust-version tag to their Cargo.toml That would break compiling geosjon with the older Rust version. That would be a breaking (Cargo) change imo.

WRT to a dependency bumping MSRV specifically: It's true — that could break someone's build. Much ink has been spilled on whether MSRV bumps should result in bumping your library's major version. I see both sides, but agree with the majority - namely that MSRV should not constitute a breaking release: rust-lang/api-guidelines#231

So I think geo-types is the only relevant dependency in terms of 1.0.

[michaelkirk]

So I think geo-types is the only relevant dependency in terms of 1.0.

Having thought about this more, geo-types 1.0 is relevant, but not necessarily blocking us from releasing geojson 1.0.

Though some people think "anything goes when you're pre-1.0.0", cargo's interpretation of semver is specific:

• with x.y.z, bumping x constitutes a breaking release.
• with 0.y.z, bumping y constitutes a breaking release.

As one of the maintainers of geo-types, I can assure you any breaking changes published in the last 5+ years have been, and will continue to be, in line with cargo semver convention (also, in practice geo-types rarely breaks semver - last in 2021).

We're pre-1.0 in a lot of georust crates, and I value jumping through some hoops to maintain backwards compatibility when reasonable, even though they're "not stable".

But, I'm not married to backwards compatibility, and I would not expect us to ossify at v1 forever. That's old pre-semver thinking. I'm happy to continue making breaking releases (v2.y.z, v3.y.z...) when breaking the API significantly improves the user experience. Our current pace (roughly once or twice a year, but slowing down), seems reasonable to me.

My motivation for "going 1.0" is that I'm excited to distinguish between "this release adds new features, but doesn't break any APIs" vs. "this is a patch/bugfix release".

For example, the proposed 0.24.2 release adds a lot of new features and changes a lot of code. Commensurately, I'm less confident that we're not breaking something.

I wish it could be a minor release, rather than a patch release, but we can't technically have minor releases in this pre-1.0.0 purgatory.

So my fellow semver PhD's, any reason to not just publish geojson-1.0.0 immediately (with no further changes)? I'm not an expert on this (nor have I ever met someone who claims to be). I would love both confirmation that this seems reasonable or a factual refutation for why we shouldn't.

[frewsxcv]

For all the reasons you stated above, I approve of a 1.0.0 release. Seems like a great belated 10th birthday present for the crate.