yuris/spacetimedb-defold
1
0
Fork 0
Code Issues 1 Pull requests Projects Releases 9 Packages Wiki Activity Actions

archive: fix the broken HTML5 install path and ship a consumer-only game.project #53

Merged
yuris merged 3 commits from feat/archive-fixes into main 2026-08-20 08:04:07 +00:00
Conversation 1 Commits 3 Files changed 9 +106 -44
yuris commented 2026-08-20 07:50:44 +00:00
Owner
Copy link

Release-gate work for #24: the consumer-archive checklist item, plus the version and install-URL bumps it depends on. Ten findings from an archive audit, all documentation or configuration — no stdb/ logic touched.

The one that mattered: the HTML5 install path was broken

README.md:95 was the only consumer-facing instruction for the only HTML5 path, and it did not work:

unzip -q .internal/lib/*websocket*.zip -d /tmp/ws

Defold does not name its dependency-cache entries after the dependency. The one entry in this repo's own cache is 14fb91dae55b356692919e5cdfec9d2f27beae6c-ImNjM2I4Y2FmZTY3OGY2ZjMzYmEwZjEx…i.zip — sha1 plus base64-of-URL-hash, with websocket appearing nowhere. Run verbatim:

unzip: cannot find or open .internal/lib/*websocket*.zip, … No zipfiles found.
exit=9

The repo's own harness already knew: dev/tools/html5-conformance.sh:44-52 opens every cached zip and identifies the extension by looking inside for websocket/ext.manifest, with a comment saying why filename matching is unreliable. So README.md:91's claim that the snippet was "exactly what dev/tools/html5-conformance.sh does" was false precisely at the step that breaks. That claim is deleted.

The install section now downloads 4.2.3 from upstream directly, sidestepping the Defold cache. Verified by extracting the instructions programmatically from the shipped README and running them as a script: exit 0, pristine md5 833ad14e… confirmed pre-patch, and CreateConnectionEmscripten(conn) relocated into LuaConnect at line 514 behind #if defined(__EMSCRIPTEN__). No .rej, no .orig.

This also fixes the quieter half: [library] include_dirs = stdb means Defold mounts only stdb/, so patches/ shipped inside the zip but never appeared in a consumer's project tree — the old path/to/patches/… was unresolvable without manually locating that same hash-named zip. The patch now has a direct raw URL.

The patch trailer contradicted the README

patches/extension-websocket-html5-subprotocol.patch is the only content of patches/ and its own only documentation. Its trailer said "apply to a fork … and point game.project at the fork" — the opposite of what the README says, and exactly what #15 frames as the thing to avoid. The archive was shipping the superseded instruction as the more discoverable one.

Trailer rewritten to the vendoring flow. The patch body is untouched: the audit found it genuinely self-describing and verified both hunks apply to pristine 4.2.3 with zero fuzz and zero offset. Re-verified after editing — patch -p1 --dry-run clean, hunks at 424 and 508, git apply --numstat 8/3.

game.project: split into a library manifest and a harness overlay

It declared [bootstrap] main_collection = /dev/example/main/main.collectionc and [input] game_binding = /dev/example/input/game.input_bindingc — both under export-ignored /dev, so both targets were absent from the archive. Plus three sections of harness-only display and script settings.

The tension: this same file is the harness's own project file, so the conformance runs need those keys. Resolved with Defold's own mechanism rather than an invented one — bob --settings, which layers settings files left to right. dev/example/harness.project carries the harness keys and both harness scripts pass the flag.

Verified the layering actually merges before relying on it: a settings file carrying a bogus dependencies#0 made bob resolve try to fetch it, in all four path spellings.

Tradeoff, stated plainly. A consumer who extracts the archive and opens it as a Defold project still cannot build it — bob resolve OK, bob build exit 1 with the '/logic/main.collection' can't be found. That is Defold's default main_collection, not a dangling reference. Judged acceptable: the honest state of this artifact is "a library with no application in it", and Defold's generic default says that, whereas the old value actively sent readers hunting for a dev/example/ directory deliberately stripped from their download. Making the build succeed would mean shipping a main collection to consumers purely to satisfy a project-open they have no reason to perform.

harness.project carries no comments because bob's settings parser rejects # both before the first section and inside one — it logs Could not parse at SEVERE and then silently ignores the file, which would have dropped the settings without failing the build. All three cases verified; the rationale lives in dev/tools/conformance.sh next to the flag instead.

Version and URL bumps

  • game.project version 0.1.0 → 1.0.0
  • Both install URLs → v1.0.0. Shape left alone — Forgejo serves /archive/<tag>.zip, and the token-as-query-param variant matches Forgejo's behaviour since basic auth 404s. grep for 0.1.0 across the repo afterwards: zero matches.

The raw/tag/<tag>/<path> shape was verified against the tag that exists (raw/tag/v0.1.0/README.md → 200); v1.0.0 404s until the tag is cut, which is correct for a release-gate branch since the README ships inside the v1.0.0 archive.

Smaller findings

  • .gitignore shipped to consumers — no export-ignore rule, while every other repo-only dotfile had one. It ignores /dev/test/.deps/, /dev/server/node_modules/, /result-*, /.direnv/ — paths absent from the archive. Rule added; its duplicate /.internal/ entry fixed.
  • The .gitattributes size comment claimed "69 KB before, 23 KB after". Real numbers now, in both places it appeared (the audit only named one): 54 entries / ~220 KB before, 11 entries / 87,646 B after; payload 232,696 B, README.md 76,502 B of it.
  • The README cited dev/ paths a consumer never receives. The ## Dev section now leads with the fact that dev/ is stripped from the release archive, so none of those commands exist in a downloaded copy.

Verified against the live forge, not a local simulation

The decisive check, because git archive locally is not byte-equivalent to the published artifact — the forge adds a spacetimedb-defold/ prefix:

GET /archive/feat/archive-fixes.zip -> 200, 87,743 bytes, payload 232,696 B
  spacetimedb-defold/README.md
  spacetimedb-defold/game.project
  spacetimedb-defold/patches/extension-websocket-html5-subprotocol.patch
  spacetimedb-defold/stdb/{cache,connection,init,protocol,schema,tables}.lua

.gitignore absent — confirmed fixed in production. Shipped game.project reads version = 1.0.0 with include_dirs = stdb and nothing else. Shipped patch trailer confirmed to be the vendoring text.

Suite

6 passed, 0 failed; stylua clean. Native conformance PASS both runs through PHASE5, identity persisted across a process restart, schema fixture matched. HTML5 conformance PASS — run rather than assumed, because this PR changes that script's bob invocation: handshake subprotocol=v1.json.spacetimedb → 101, phases 1–5, reconnect after restart, identity survived a reload.

Release-gate work for #24: the consumer-archive checklist item, plus the version and install-URL bumps it depends on. Ten findings from an archive audit, **all documentation or configuration — no `stdb/` logic touched.** ## The one that mattered: the HTML5 install path was broken `README.md:95` was the only consumer-facing instruction for the only HTML5 path, and it did not work: ``` unzip -q .internal/lib/*websocket*.zip -d /tmp/ws ``` Defold does not name its dependency-cache entries after the dependency. The one entry in this repo's own cache is `14fb91dae55b356692919e5cdfec9d2f27beae6c-ImNjM2I4Y2FmZTY3OGY2ZjMzYmEwZjEx…i.zip` — sha1 plus base64-of-URL-hash, with `websocket` appearing nowhere. Run verbatim: ``` unzip: cannot find or open .internal/lib/*websocket*.zip, … No zipfiles found. exit=9 ``` The repo's own harness already knew: `dev/tools/html5-conformance.sh:44-52` opens every cached zip and identifies the extension by looking *inside* for `websocket/ext.manifest`, with a comment saying why filename matching is unreliable. So `README.md:91`'s claim that the snippet was "exactly what `dev/tools/html5-conformance.sh` does" was **false precisely at the step that breaks**. That claim is deleted. The install section now downloads 4.2.3 from upstream directly, sidestepping the Defold cache. Verified by extracting the instructions programmatically from the shipped README and running them as a script: exit 0, pristine `md5 833ad14e…` confirmed pre-patch, and `CreateConnectionEmscripten(conn)` relocated into `LuaConnect` at line 514 behind `#if defined(__EMSCRIPTEN__)`. No `.rej`, no `.orig`. This also fixes the quieter half: `[library] include_dirs = stdb` means Defold mounts only `stdb/`, so `patches/` shipped inside the zip but never appeared in a consumer's project tree — the old `path/to/patches/…` was unresolvable without manually locating that same hash-named zip. The patch now has a direct raw URL. ## The patch trailer contradicted the README `patches/extension-websocket-html5-subprotocol.patch` is the only content of `patches/` and its own only documentation. Its trailer said *"apply to a fork … and point `game.project` at the fork"* — the opposite of what the README says, and exactly what #15 frames as the thing to avoid. **The archive was shipping the superseded instruction as the more discoverable one.** Trailer rewritten to the vendoring flow. The patch **body** is untouched: the audit found it genuinely self-describing and verified both hunks apply to pristine 4.2.3 with zero fuzz and zero offset. Re-verified after editing — `patch -p1 --dry-run` clean, hunks at 424 and 508, `git apply --numstat` 8/3. ## `game.project`: split into a library manifest and a harness overlay It declared `[bootstrap] main_collection = /dev/example/main/main.collectionc` and `[input] game_binding = /dev/example/input/game.input_bindingc` — both under export-ignored `/dev`, so **both targets were absent from the archive**. Plus three sections of harness-only display and script settings. The tension: this same file is the harness's own project file, so the conformance runs need those keys. Resolved with Defold's own mechanism rather than an invented one — `bob --settings`, which layers settings files left to right. `dev/example/harness.project` carries the harness keys and both harness scripts pass the flag. Verified the layering actually merges before relying on it: a settings file carrying a bogus `dependencies#0` made `bob resolve` try to fetch it, in all four path spellings. **Tradeoff, stated plainly.** A consumer who extracts the archive and opens it as a Defold project still cannot build it — `bob resolve` OK, `bob build` exit 1 with `the '/logic/main.collection' can't be found`. That is Defold's *default* `main_collection`, not a dangling reference. Judged acceptable: the honest state of this artifact is "a library with no application in it", and Defold's generic default says that, whereas the old value actively sent readers hunting for a `dev/example/` directory deliberately stripped from their download. Making the build succeed would mean shipping a main collection to consumers purely to satisfy a project-open they have no reason to perform. `harness.project` carries no comments because bob's settings parser rejects `#` both before the first section and inside one — it logs `Could not parse` at SEVERE and then **silently ignores the file**, which would have dropped the settings without failing the build. All three cases verified; the rationale lives in `dev/tools/conformance.sh` next to the flag instead. ## Version and URL bumps - `game.project` `version` `0.1.0` → `1.0.0` - Both install URLs → `v1.0.0`. Shape left alone — Forgejo serves `/archive/<tag>.zip`, and the token-as-query-param variant matches Forgejo's behaviour since basic auth 404s. `grep` for `0.1.0` across the repo afterwards: **zero matches.** The `raw/tag/<tag>/<path>` shape was verified against the tag that exists (`raw/tag/v0.1.0/README.md` → 200); `v1.0.0` 404s until the tag is cut, which is correct for a release-gate branch since the README ships *inside* the v1.0.0 archive. ## Smaller findings - `.gitignore` shipped to consumers — no `export-ignore` rule, while every other repo-only dotfile had one. It ignores `/dev/test/.deps/`, `/dev/server/node_modules/`, `/result-*`, `/.direnv/` — paths absent from the archive. Rule added; its duplicate `/.internal/` entry fixed. - The `.gitattributes` size comment claimed "69 KB before, 23 KB after". Real numbers now, in **both** places it appeared (the audit only named one): 54 entries / ~220 KB before, 11 entries / 87,646 B after; payload 232,696 B, `README.md` 76,502 B of it. - The README cited `dev/` paths a consumer never receives. The `## Dev` section now leads with the fact that `dev/` is stripped from the release archive, so none of those commands exist in a downloaded copy. ## Verified against the live forge, not a local simulation The decisive check, because `git archive` locally is *not* byte-equivalent to the published artifact — the forge adds a `spacetimedb-defold/` prefix: ``` GET /archive/feat/archive-fixes.zip -> 200, 87,743 bytes, payload 232,696 B spacetimedb-defold/README.md spacetimedb-defold/game.project spacetimedb-defold/patches/extension-websocket-html5-subprotocol.patch spacetimedb-defold/stdb/{cache,connection,init,protocol,schema,tables}.lua ``` `.gitignore` **absent** — confirmed fixed in production. Shipped `game.project` reads `version = 1.0.0` with `include_dirs = stdb` and nothing else. Shipped patch trailer confirmed to be the vendoring text. ## Suite `6 passed, 0 failed`; stylua clean. Native conformance PASS both runs through PHASE5, identity persisted across a process restart, schema fixture matched. **HTML5 conformance PASS** — run rather than assumed, because this PR changes that script's `bob` invocation: handshake `subprotocol=v1.json.spacetimedb` → `101`, phases 1–5, reconnect after restart, identity survived a reload.
archive: fix the broken HTML5 install path and ship a consumer-only game.project
All checks were successful
ci / ci (pull_request) Successful in 20s
Details
e14773a585 
Ten findings from a verified audit of the v1.0.0 candidate archive. Every one
is documentation or configuration; no stdb/ logic changed.

A4 (HIGH) was a proven break, not a style fix. README:95 was the only
consumer-facing instruction for the only HTML5 path, and it did not work:

    unzip -q .internal/lib/*websocket*.zip -d /tmp/ws

Defold names dependency-cache entries by hash, not after the dependency. Run
verbatim against a real populated .internal/lib it exits 9 with "No zipfiles
found" -- the substring "websocket" appears nowhere in the filename. This
repo's own harness already knew: html5-conformance.sh iterates every *.zip and
identifies the extension by looking inside it for websocket/ext.manifest, with
a comment explaining why filename matching is unreliable. So the README's claim
that the snippet was "exactly what dev/tools/html5-conformance.sh does" was
false precisely at the step that broke. Replaced with a download of the 4.2.3
tag from upstream, which sidesteps the cache entirely.

A5 falls out of the same fix. [library] include_dirs = stdb means Defold mounts
only stdb/, so patches/ ships inside the zip but never appears in a consumer's
project tree, which made the old `path/to/patches/...` unresolvable without
locating that same hash-named cache zip. The patch now has a direct raw URL on
the tag, so patches/ being invisible to Defold stops mattering.

A6: the patch's own trailer told readers to fork extension-websocket and point
game.project at the fork -- the opposite of what the README says, and the exact
thing #15 exists to avoid. patches/ is that file's only documentation, so the
archive was shipping the superseded instruction as the more discoverable one.
Trailer prose only: both hunks are untouched and still apply to pristine 4.2.3
with zero fuzz and zero offset.

A2: game.project declared [bootstrap] main_collection and [input] game_binding
under /dev, which .gitattributes strips, so both targets were absent from every
archive a consumer downloaded. It also shipped the harness's [script],
[display] and [graphics] settings. Moved all of them to
dev/example/harness.project and layered them back with bob --settings, which is
Defold's own mechanism for exactly this. game.project is now only what a
consumer needs: [project] title/version and [library] include_dirs. Both
conformance tiers still pass.

A7 and A8: version 0.1.0 -> 1.0.0, and both install URLs from
/archive/v0.1.0.zip to v1.0.0. The URL shape was already correct -- Forgejo
serves /archive/<tag>.zip and refuses basic auth -- only the tag was stale.

A1: .gitignore had no export-ignore rule, so it landed in every consumer
archive despite being unambiguously dev-only: it ignores /dev/test/.deps/,
/dev/server/node_modules/, /result-* and /.direnv/, none of which exist in the
archive. An omitted rule, not a deliberate exception.

A3: the .gitattributes size comment claimed "69 KB before, 23 KB after", which
was true when the rules were added and is now stale by 2.6x. Measured on this
tree: 54 entries / 224,894 B of zip before, 11 entries / 87,583 B after.

A9: the README cited dev/ paths a consumer never receives. Both the
html5-conformance.sh citation and the Dev section now say so explicitly.

Also dropped a redundant .gitignore block: /dist/headless/ was already covered
by /dist/, and /.internal/ was listed twice.

Verified: lua suite 6/6, stylua clean, native conformance PASS, html5
conformance PASS (101 Switching Protocols with the subprotocol negotiated), the
README command block runs clean when extracted verbatim from the file, and the
edited patch still applies to pristine 4.2.3 with no .rej and no .orig.

Refs #24, #15
archive: make the install snippet fail loudly and rerun cleanly
Some checks failed
ci / ci (pull_request) Has been cancelled
Details
48a2cfcca7 
Review of #53 found three robustness gaps, all verified by running the
shipped instructions rather than reading them.

`curl -sL` on a 404 pipes the 11-byte body `Not found.` into patch, which
answers `Only garbage was found in the patch input` and never names the
cause — which is exactly what a consumer hits until the v1.0.0 tag is cut.
`-f` turns that into curl naming the 404. Same class on the zip download,
where an HTML error body would otherwise be written into /tmp/ws.zip.

`unzip -q` into a hardcoded /tmp/wsx is not idempotent: a second run stops
at a replace prompt and exits 1 under a non-tty. `-o` fixes it.

Stripping game.project to the library manifest also broke the Defold editor
as a harness entry point: the two scripts and CI pass `bob --settings`, but
the editor has no equivalent, so Build/Run no longer boots dev/example. That
was undocumented; flake.nix's shell banner now says so.
flake: anchor the JSON vendoring to the repo root, not $PWD
All checks were successful
ci / ci (pull_request) Successful in 21s
Details
3379c51584 
vendorJson used relative paths, so entering the dev shell from any
subdirectory minted a second dev/test/.deps tree there. Found as an
untracked dev/server/dev/test/.deps/json.lua in the main checkout — and
.gitignore only covers the one at the root, so the stray would have been
committed by any `git add -A`.

Verified: `cd dev/server && nix develop` now creates nothing locally and
the root symlink still resolves into the store.
yuris commented 2026-08-20 08:04:06 +00:00
Author
Owner
Copy link

Review — automated reviewer, posted for the record

Verdict: READY TO TAG. All five review items verified by running them; no blocking defects. Five findings, all non-blocking; B2/B3/B4 have since been actioned in 48a2cfc and a root cause the review surfaced in 3379c51.

Install instructions — run, they work

The shipped README from the live forge zip contains exactly one sh fenced block in 1369 lines, so extraction is unambiguous. Run in a clean empty dir with no repo access: exit 0, patch applied, no .rej/.orig, pristine md5 833ad14e… → 8eb2ab3e…, hunks at 424 and 508. A second run using the branch's own shipped patch fetched by URL produced a byte-identical result. The old exit-9 failure is gone and nothing in the flow touches Defold's cache.

raw/tag/<tag>/<path> confirmed a real Forgejo route via the tag that exists (raw/tag/v0.1.0/README.md → 200). v1.0.0 404s until cut, as expected.

game.project split

Live-forge copy is 175 B / 7 lines: [project] title + version = 1.0.0 + dependencies#0, and [library] include_dirs = stdb. Every /dev path and all three harness-only sections gone.

Both conformance tiers PASS on a fresh clone with an isolated port — native through PHASE5 with identity persisted across a process restart, HTML5 handshake → 101 with PHASE1-5 on both loads and identity surviving a reload.

B1 — premise right, mechanism wrong. The PR body said bob "silently ignores" a settings file containing comments. Measured: bob exits 1 and prints Could not parse: <file> twice, aborting before it resolves libraries. All four comment placements (#/;, before and inside a section) fail the same way. So the decision stands — no comment syntax parses, the rationale must live elsewhere — but the real behaviour is safer than described.

Layering was proven to genuinely merge by using a bogus dependencies#0 as the observable: no --settings → exit 0; with it → Failed to fetch library … HTTP 404.

The tradeoff is the right call. A dependency's game.project is read by Defold only for include_dirs; no consumer path performs the project-open that fails. Making bob build succeed would mean shipping a main collection plus a third top-level directory that include_dirs = stdb would not even expose. And the failure a consumer now sees is Defold's own generic default, where the old value actively sent readers hunting for a dev/example/ directory deliberately stripped from their download.

Archive contents — zero accidental exclusions

Decisive check: diff of unzip -Z1 on /archive/main.zip against the branch's archive shows exactly one line, -spacetimedb-defold/.gitignore, and nothing in either direction besides it. 13 → 12 entries. No module, patch or README lost.

The .gitattributes numbers reproduce to the byte: 54 entries / 225,054 B before, 11 / 87,646 B after; payload 232,696 B, README 76,502 B (32.9%), stdb/ 153,302 B — the parts sum to the whole.

Version and URLs

grep -rn '0\.1\.0' over the whole checkout: zero matches, shipped or not. v1.0.0 at exactly three places, all README. The upstream 4.2.3 URL occurs 5× and none was disturbed; release.yml builds its dependency URL from \${TAG}, so nothing to bump there.

Patch trailer

Body byte-unchanged — diff -u yields a single hunk located entirely above the --- a/ line. The word fork now appears only as "no fork is needed". Still applies cleanly, proven by the install run that fetched this exact shipped file by URL.

Findings

  • B1 INFO — PR body's bob-parser claim. Wrong reason recorded as history; true behaviour is safer. Corrected in this comment; no repo change.
  • B2 LOW — flake.nix:180. bob build at the repo root without --settings now fails, so the editor advertised in the shell banner no longer boots dev/example. Actioned in 48a2cfc.
  • B3 LOW — README.md:105,108. curl -sL on a 404 pipes Not found. into patch → Only garbage was found in the patch input, exit 2, cause never named. Actioned: curl -fsSL.
  • B4 INFO — README.md:105-106. Hardcoded /tmp/wsx made a rerun stop at a replace prompt and exit 1 under a non-tty. Actioned: unzip -qo.
  • B5 INFO — patch trailer's own snippet assumes the file sits in the project root while the README pipes it from curl. Reader is holding the file; noted for consistency only.

Also fixed, found while actioning

vendorJson in flake.nix used relative paths, so entering the dev shell from any subdirectory minted a phantom dev/test/.deps tree there — an untracked dev/server/dev/test/.deps/json.lua was sitting in the main checkout, and .gitignore only covers the root one, so any git add -A would have committed it. Anchored to the repo root in 3379c51 and verified from dev/server.

## Review — automated reviewer, posted for the record **Verdict: READY TO TAG.** All five review items verified by running them; no blocking defects. Five findings, all non-blocking; B2/B3/B4 have since been actioned in `48a2cfc` and a root cause the review surfaced in `3379c51`. ### Install instructions — run, they work The shipped README from the **live forge zip** contains exactly one `sh` fenced block in 1369 lines, so extraction is unambiguous. Run in a clean empty dir with no repo access: exit 0, patch applied, no `.rej`/`.orig`, pristine `md5 833ad14e…` → `8eb2ab3e…`, hunks at 424 and 508. A second run using the branch's own shipped patch fetched by URL produced a **byte-identical** result. The old exit-9 failure is gone and nothing in the flow touches Defold's cache. `raw/tag/<tag>/<path>` confirmed a real Forgejo route via the tag that exists (`raw/tag/v0.1.0/README.md` → 200). `v1.0.0` 404s until cut, as expected. ### game.project split Live-forge copy is 175 B / 7 lines: `[project]` title + `version = 1.0.0` + `dependencies#0`, and `[library] include_dirs = stdb`. Every `/dev` path and all three harness-only sections gone. **Both conformance tiers PASS** on a fresh clone with an isolated port — native through PHASE5 with identity persisted across a process restart, HTML5 handshake → 101 with PHASE1-5 on both loads and identity surviving a reload. **B1 — premise right, mechanism wrong.** The PR body said bob "silently ignores" a settings file containing comments. Measured: bob **exits 1** and prints `Could not parse: <file>` twice, aborting before it resolves libraries. All four comment placements (`#`/`;`, before and inside a section) fail the same way. So the decision stands — no comment syntax parses, the rationale must live elsewhere — but the real behaviour is *safer* than described. Layering was proven to genuinely merge by using a bogus `dependencies#0` as the observable: no `--settings` → exit 0; with it → `Failed to fetch library … HTTP 404`. **The tradeoff is the right call.** A dependency's `game.project` is read by Defold only for `include_dirs`; no consumer path performs the project-open that fails. Making `bob build` succeed would mean shipping a main collection plus a third top-level directory that `include_dirs = stdb` would not even expose. And the failure a consumer now sees is Defold's own generic default, where the old value actively sent readers hunting for a `dev/example/` directory deliberately stripped from their download. ### Archive contents — zero accidental exclusions Decisive check: `diff` of `unzip -Z1` on `/archive/main.zip` against the branch's archive shows **exactly one line**, `-spacetimedb-defold/.gitignore`, and nothing in either direction besides it. 13 → 12 entries. No module, patch or README lost. The `.gitattributes` numbers reproduce to the byte: 54 entries / 225,054 B before, 11 / 87,646 B after; payload 232,696 B, README 76,502 B (32.9%), `stdb/` 153,302 B — the parts sum to the whole. ### Version and URLs `grep -rn '0\.1\.0'` over the whole checkout: **zero matches**, shipped or not. `v1.0.0` at exactly three places, all README. The upstream 4.2.3 URL occurs 5× and none was disturbed; `release.yml` builds its dependency URL from `\${TAG}`, so nothing to bump there. ### Patch trailer Body **byte-unchanged** — `diff -u` yields a single hunk located entirely above the `--- a/` line. The word `fork` now appears only as "no fork is needed". Still applies cleanly, proven by the install run that fetched this exact shipped file by URL. ### Findings - **B1** INFO — PR body's bob-parser claim. Wrong reason recorded as history; true behaviour is safer. *Corrected in this comment; no repo change.* - **B2** LOW — `flake.nix:180`. `bob build` at the repo root without `--settings` now fails, so the editor advertised in the shell banner no longer boots `dev/example`. **Actioned** in `48a2cfc`. - **B3** LOW — `README.md:105,108`. `curl -sL` on a 404 pipes `Not found.` into `patch` → `Only garbage was found in the patch input`, exit 2, cause never named. **Actioned**: `curl -fsSL`. - **B4** INFO — `README.md:105-106`. Hardcoded `/tmp/wsx` made a rerun stop at a replace prompt and exit 1 under a non-tty. **Actioned**: `unzip -qo`. - **B5** INFO — patch trailer's own snippet assumes the file sits in the project root while the README pipes it from curl. Reader is holding the file; noted for consistency only. ### Also fixed, found while actioning `vendorJson` in `flake.nix` used relative paths, so entering the dev shell from any subdirectory minted a phantom `dev/test/.deps` tree there — an untracked `dev/server/dev/test/.deps/json.lua` was sitting in the main checkout, and `.gitignore` only covers the root one, so any `git add -A` would have committed it. Anchored to the repo root in `3379c51` and verified from `dev/server`.
yuris merged commit 65282fb637 into main 2026-08-20 08:04:07 +00:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels   
area:defold
Defold engine integration

  
area:docs
Documentation and examples

  
area:infra
Flake, CI, secretspec

  
area:protocol
Wire protocol conformance

  
area:testing
Test harness and fixtures

  
kind:bug
Something broken

  
kind:chore
Maintenance

  
phase:breadth
UnsubscribeMulti, procedures, BSATN

  
phase:core
Core SDK layers (protocol/cache/connection)

  
phase:hardening
Reconnect, token persistence, errors

  
phase:slice
Vertical slice: first playable end-to-end

No labels
area:defold
area:docs
area:infra
area:protocol
area:testing
kind:bug
kind:chore
phase:breadth
phase:core
phase:hardening
phase:slice
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
yuris/spacetimedb-defold!53
No description provided.