A build cache serves a dependency rebuild after a branch switch or a `git clean` instead of
recompiling it, which is most of what the in-loop verify spends. It stays a convenience and
never a requirement: with nothing on `PATH` nothing is exported and a clone needs only stable
Rust and a C compiler, exactly as before.
`check.sh`, `gates.sh` and `smoke.sh` source it, so the loops pick it up without being asked
and a machine that would rather not have it sets `BENILLA_NO_SCCACHE=1`. An existing
`RUSTC_WRAPPER` wins. Only rustc is wrapped — the vendored Lua build wants its own
`CC="sccache cc"`, so it is not cached by this.
Logging in to an account or realm with no characters lands on the empty character list. Until now the login cycle created a male Human Warrior on the server, named `WOW_CHAR` or "One", on every login that found an empty roster, a typed one included and again after Change Realm.
- **The reference.** The client builds `CMSG_CHAR_CREATE` in one place, `0x5aac50`, reached only from the glue binding `CreateCharacter` (`0x4719e0` → `0x4711d0` → `0x46b290` → `0x5aac50`), which stock `CharacterCreate.lua` calls only from `CharacterCreate_Okay`. With an empty roster, stock `CharacterSelect.lua`'s `UpdateCharacterList` disables Enter World and Delete and shows Create New Character, which benilla's select screen already does.
- **The change.** The create-if-empty step is gone from the per-realm cycle, and `NetConfig`, which existed only for it, goes with it. The create screen's own path is untouched.
- **Scripted runs.** `WOW_CHAR` must now be on the account: a driverless run whose `WOW_CHAR` is missing fails at select, as it already did on a non-empty roster, and a player's run shows the list. `docs/CONTRIBUTING.md` says so, and names `WOW_PROBE_CHARCREATE` as the explicit way to make one.
Checked: `an_empty_account_parks_at_select_without_creating_a_character` drives the real login cycle against a fake realmd (the server half of SRP6) and a fake world server answering an empty roster. It fails with the old step restored (the client sends `CMSG_CHAR_CREATE` on its own) and passes without it. Live against a local vmangos on a fresh account with no characters: before, the server gained a level-1 "One"; after, the client parks on the empty list and the account still holds no character. With a `WOW_CHAR` not on the account, a driverless run exits at select, and nothing is created. Gates and the smoke run green on macOS; no platform seam touched.
Fixes#343🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A maintainer reviews it against the reference and usually finishes it on the contributor's branch
(a rebase, a fix, a test, a comment) rather than sending it back, so "Allow edits by maintainers"
stays ticked; it lands as one squash-merged commit with the contributor as its author, or, when
what lands would be mostly ours, as our own commit crediting them as co-author.
Docs only; no gate reads it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The gates of scripts/gates.sh that need neither the install nor a display (fmt, clippy, the
workspace tests without an install, the doc-link and render-pass lints, the player build and its
tests), on ubuntu-latest, so a red pull request is seen before anyone builds it. A branch of this
repository skips the job: it runs the full chain before it lands. The job reads the repository
only, keeps no credentials in the checkout, pins its two actions by commit, drops debuginfo for
the runner's disk and caches dependencies. gates.sh and CONTRIBUTING say what runs where.
actionlint passes; the system packages are the ones CONTRIBUTING names for Linux. The workflow's
first run is a manual one on main once it lands.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The probe skill was a Claude-only file, and its promises held on one machine: its shield needs an
account named probe and digits, which CONTRIBUTING never asks for, and its long-run advice was
macOS's. The switches every checkout has for an unattended run are plain text in CONTRIBUTING
now, read by any tool; METHOD points there, AGENTS.md drops the skills folder, and .gitignore
leaves .claude/ to each person.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A committed .claude/settings.json bound everyone who opens the repo in Claude Code: it switched
their auto memory off, and its one hook guarded the install against Claude's Edit and Write
tools only, never a write through the shell or by another agent. The rule it backed is enforced
for every tool by the code and by scripts/smoke.sh's install watch, which METHOD now cites. jq is
no longer a prerequisite, and .gitignore shares only the skills under .claude/.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The workflow that closed every pull request on arrival is gone, the README's closed-doors
notice becomes a short Contributing section, and CONTRIBUTING.md says how to report a bug.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
465 tests loaded the stock interface off the install without declaring
it, and 25 addon tests assumed the install beside the corpus; each now
opens with `wow_data_or_skip!`. The gates run the workspace tests a
second time with `WOW_DATA=`, so a test that reads the install
undeclared fails at land instead of on a contributor's clone. The skip
log writes each line in one call so parallel tests do not interleave.
Verified: the workspace tests with `WOW_DATA=` in a checkout that has
the addon corpus pass 6972, fail 0, and log 1555 skips; on a fresh
clone of main before this change they failed 465.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
What a clone lacked: the gates went green with a thousand skipped tests and said nothing, every
login script assumed a worktree-pool slot and its probe account, the client defaulted to the
maintainer's account, a docker path and a home directory were in the code, the temp files broke
on GNU coreutils, and two generators could only run against a sibling repo.
Now the test gates print how many data-gated tests skipped and why; a scripted run logs in as
the account its checkout declares in `.probe-identity` or as the three variables name, never a
default, and the client refuses a scripted login on any other account; the event and message
tables regenerate from vendored inputs, byte for byte; the docs state the real prerequisites.
Out of the tree, into the maintainer's own tooling: the worktree pool, its hooks, the perf leg
runner, the cross-platform compile and the Vulkan container leg — a way of working, and setups
not every machine has. One hook stays: no writes into the install.
Two tests that could never run are gone with their subject: the quest-timer geometry diff
compared a copy of the window that no longer exists (and joined a chain path onto a folder, so
it always skipped), and the FrameXML flag gate now reads the stock files off the player's own
patch chain instead of an extracted folder only one machine had.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Until now this repo was an export of a private tree, pushed as squashed snapshots. From here on
it is the working tree itself, and the private repo is archived.
The root holds AGENTS.md and the Rust and GitHub files only. Everything an agent needs to orient
is in docs/: METHOD.md (the rules and how the project is built, 123 lines, project method only),
MAP.md (generated from the code at every land) and CONTRIBUTING.md (what gets in and how a
change is judged). One skill, `probe`, holds the procedure for running the client unattended
against the local server.
Brought over because the project needs it: reference/ (the 1.12 name catalogues 34 source
files read; the export failed its own tests without them), the gate chain and the pool
(scripts/), the hooks that enforce the install rule and the worktree rule, the full .gitignore,
the `cargo play` alias, and the original wording of eleven source files the export had reworded.
Left out: the decision records (archived with the private repo), the export tooling, the
operator's Discord and triage loop, the lab-laptop scripts and the instruments nothing
references, which live outside the repo now. What read the records is gone from the tooling:
no corpus lint, no records index, no `reserve` verb, no number-collision check.
Three hooks are gone: fmt-at-commit and the file-size nudge were redundant with the gates, and
the assets/ui guard is now a test (`assets_ui_does_not_grow`) that fails for everyone, not only
for a Claude Code session. The remaining pool hooks stand down on a machine without a pool.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
chairmanpow