Repository navigation
Description
The next tag has to be at least v0.9.0, and nothing records that
v0.8.5 is the highest tag. main is 66 commits ahead of it, and at least one of those commits is breaking for a Ruby-facing API:
Augeas.create used to accept three positional arguments, matching Augeas.open:
Augeas.create("/root", "/lens", 5) # worked before #952#952 gave it the gem's real signature — Augeas::Facade::create(opts={}), an options hash — so that call is now a TypeError. The change is right (the old signature was simply not ruby-augeas's), but a program written against it breaks. Below v1 the convention in force is breaking → MINOR, so the next tag is v0.9.0, not v0.8.6.
Why this is an issue rather than a note in a commit
ghrelease takes the tag as an argument:
ghrelease [flags] [owner/repo] <pr-number> <tag>
It validates that the tag is above the repository's highest release (--allow-lower to override), and nothing else about it. Nothing computes or checks MAJOR/MINOR/PATCH against what actually changed, so whoever cuts the next release picks the number by hand, 66 commits after the last one, from a release-notes list generated by GitHub.
The failure mode is quiet: v0.8.6 would publish, be marked Latest, and break Augeas.create for anyone who pinned a patch range — with no red build anywhere, because semver is not something CI measures.
⛔ What I am NOT claiming
I have not read all 66 commits. #952 is one breaking change I know of because I wrote it — not necessarily the only one in that window. Anyone cutting the release should treat v0.9.0 as a floor established by this one, not as a verdict that the window holds exactly one.
A second thing worth checking at the same time, same window: #951 removed Resolv::DNS#close's blanket NotImplementedError and gave it MRI's arity-0 nil, and #952 added Augeas#close plus a SystemCallError from every method of a closed handle. Both are additions rather than removals, so neither forces a MAJOR, but they are API surface changes a release note should name.
A tool-shaped alternative
If the fleet wants this enforced rather than remembered, the shape that fits ghrelease is a refusal rather than a computation: let a commit or PR body carry a marker (⛔ BREAKING: / Breaking-Change:), have ghrelease scan the commits between the previous tag and the one being cut, and refuse a PATCH bump when any of them carries it. That keeps the judgement with the author who made the change — who is the only one who can tell a fix from a break — and puts the refusal before the publish, where a gate belongs. Filed as an idea, not a design: it needs go-fleettools' agreement about the marker before it is worth writing.
Refs #952, #951 — both already merged; they are named as the API changes this issue is about, and nothing here asks for either to be reopened.
🤖 Generated with Claude Code
Prior art: priorart "the next tag must be a MINOR because a breaking change landed since v0.8.5" at 2026-11-16 04:21 CET reported prior art; checked: The 899 hits are generic keyword matches (next/must/because/change) across the open tracker -- #950, #860, #912, #866, #887, #940, #941, #909 and the rest are parity defects and CI issues. None is about release versioning, semver bumps or ghrelease's tag argument. Nothing covers this.
Activity