also looking at this
fix(ci): pantry cannot resolve `latest`, so ask for ranges instead
#1099
1 file
+19
-3
| @@ -1,5 +1,21 @@ | ||
| 1 | # Ranges, never `latest`. | |
| 2 | # | |
| 3 | # pantry's installer SDK resolves `latest` from a hardcoded per-domain switch | |
| 4 | # (nodejs.org, ziglang.org, ...) that ends in a bare throw. A domain the SDK | |
| 5 | # supports but has no case for cannot resolve `latest` at all — it fails with | |
| 6 | # `Cannot resolve latest version for <domain>`. A range takes a different path | |
| 7 | # that reads the package's published version list, so it always resolves. | |
| 8 | # | |
| 9 | # That is not hypothetical: `sqlite: latest` was fine only because the SDK did | |
| 10 | # not support sqlite.org and skipped it with a warning. Adding it upstream | |
| 11 | # promoted it from skipped to installed and reddened main. `bun` never broke | |
| 12 | # here precisely because it was already a range. | |
| 13 | # | |
| 14 | # mysql and postgres are still skipped as unsupported, so their spec is inert | |
| 15 | # today — they are ranges so that being added upstream is a no-op, not an | |
| 16 | # outage. | |
| 1 | 17 | dependencies: |
| 2 | 18 | bun: ^1.3.14 |
| 3 | mysql: latest | |
| 4 | postgres: latest | |
| 5 | sqlite: latest | |
| 19 | mysql: ^8.0.45 | |
| 20 | postgres: ^18.0.0 | |
| 21 | sqlite: ^3.52.0 | |