A preview system image installed makes the newest runtime its codename, not the highest API level
维护者通常 1 天内回复
评估
这个 Issue 还没有评估数据。
描述
What is wrong
A preview system image installed makes the newest runtime its codename, not the highest API level
Where this came from
Confirmed in the bug bounty sweep tracked on #648 (finding 4). The reproduction was reproduced independently by the orchestrating session, the proposed fix confirmed green against the surrounding suite, and the finding re-confirmed against current main.
Spec
Area: src/drivers/android/index.ts (runtime ranking: lease default, defaultRuntime)
Severity: wrong device booted, silently, on a real host configuration — this
reporting machine itself has android-Baklava in
~/Library/Android/sdk/system-images/ beside android-36, android-36.1 and
android-37.0. simlock lease --platform android with no --os boots an
API 35 preview instead of 37.0, and simlock catalog /
getCatalog().defaultRuntime tell every caller and the console that Baklava
is the newest installed runtime — so an agent that trusts the catalog to pick a
valid --os, the stated purpose of that command (docs/CLI.md:1594), is
pointed at the preview. Same for --image-tag without --os.
Kind: defect — four doc lines state what "newest" means; the ranking
disagrees with all four.
Problem
#installedImages records each image's level as the android-<x> folder
suffix (index.ts:1602 separately reads AndroidVersion.ApiLevel into
sourceApiLevel precisely because the folder name is not always the level).
For the preview folder, the recorded level is the string "Baklava".
installedApiLevels (index.ts:2260) sorts that list with compareApiLevels
(index.ts:2280): Number(left) - Number(right), falling back to
localeCompare when either side is not a number. "Baklava" vs "37.0" is
NaN, so the text branch wins and any letter-starting name sorts after
every number. resolveSpec then takes levels.at(-1) (:808) and
listCatalog takes runtimes.at(-1) (:1253, and :1673 for the
--image-tag path) — both hand out "Baklava" as the newest.
The lines this breaks:
docs/CLI.md:217 `--os` defaults to the newest runtime already installed
docs/CLI.md:311 without `--os` the lease gets the newest API level that has one
docs/CLI.md:1594 which installed runtime is the default (the newest)
docs/CLIENT.md:352 `defaultRuntime` is the newest installed runtime
Range requests are unaffected: src/contract/os-range.ts parses a non-numeric
version to NaN and refuses it, so the gateway never routes a range to the
preview. This is the no---os default path and the catalog's own words about
itself.
Root cause
src/drivers/android/index.ts:2280 — compareApiLevels ranks API levels by
Number(name) and falls back to left.localeCompare(right), so a name the SDK
stores as a codename always outranks a number. It knows nothing about
sourceApiLevel, which the same file reads 680 lines earlier precisely because
the folder name is not always the level. The ranking, not the identity, is what
is wrong: "Baklava" is a legitimate name to list and to lease explicitly — it
is only wrong as the newest.
image-fit.ts:85 (for (const candidate of [image.sourceApiLevel, image.apiLevel])) already applies the correct rule elsewhere in the same
module: this file's own convention, missed at one call site.
Evidence
Test file: src/drivers/android/scratch-newest-image.test.ts
Test titles: gives a lease with no OS the highest API level installed, not the
preview folder whose name sorts last
names the highest API level installed as the catalog's default runtime
Command: pnpm exec vitest run --project unit \
src/drivers/android/scratch-newest-image.test.ts
AssertionError: expected 'Baklava' to be '37.0' // Object.is equality
Expected: "37.0"
Received: "Baklava"
❯ src/drivers/android/scratch-newest-image.test.ts:156:28 (resolveSpec)
❯ src/drivers/android/scratch-newest-image.test.ts:168:36 (defaultRuntime)
The fixture's source.properties is the verbatim content of the real
~/Library/Android/sdk/system-images/android-Baklava/google_apis/arm64-v8a/source.properties
on this machine (AndroidVersion.ApiLevel=35, AndroidVersion.CodeName=Baklava
— orchestrator confirmed the file exists with those values beside 36/36.1/37.0).
Honesty checks (the finder's, then the orchestrator's):
- Both failures are preceded by
expect(catalog.runtimes).toContain("Baklava")
and the same formodelRuntimes["Pixel 9"]— so the failure can only be about
ranking: the preview really is a listed candidate, not an omission. - Preview folder removed from the fixture: both tests pass — nothing else causes
it. - Fixture reversed (
AndroidVersion.ApiLevel=38): both tests pass against the
broken code while still asserting the preview is newest — so the tests do
not merely prefer numbers over letters; they fail exactly because the recorded
level is ignored. - Orchestrator: fix applied → new tests green and
index.test.tsgreen
(274 passed | 1 skipped), including its existingandroid-Baklavatest at
:4877, which pins that the preview still pairs with Pixel 9 and is still
namedBaklava; then the full drivers module: 21 files, 1229 tests green;
then the entire unit lane: 269 files, 7029 passed | 4 skipped. Source
reverted, both assertions re-confirmed red.
Proposed fix
src/drivers/android/index.ts:2260 — rank by the level the SDK recorded, keep
the folder suffix as the name:
function installedApiLevels(images: readonly SystemImage[]): string[] {
const rank = new Map(
images.map((image) => [
image.apiLevel,
/^\d/.test(image.apiLevel) ? image.apiLevel : (image.sourceApiLevel ?? image.apiLevel),
]),
);
return [...new Set(images.map((image) => image.apiLevel))].sort(
(left, right) =>
compareApiLevels(rank.get(left) ?? left, rank.get(right) ?? right) ||
left.localeCompare(right),
);
}
Verified above. No doc change needed — the four doc lines already describe this
behaviour.
Alternatives rejected
- Store
sourceApiLevelas the level itself — renames the runtime across the
catalog,--os,images[].runtimeandmodelRuntimes, and breaks the
existing pin atindex.test.ts:4885that the runtime is namedBaklava. - Pick the max with
contract/os-range.ts'scompareVersions—Baklavaparses
toNaNthere and would sort as the oldest, and it repeats the ranking in
three call sites instead of one. - A table of known codenames — invents a list the SDK owns and rots at the next
preview.
Risk
compareCatalogImages(index.ts:2272) sortsimages[].runtimewith the same
comparator, socatalog.imagesorder shifts (the preview moves next to API
35 — where it belongs). A reviewer should confirm no console or CLI snapshot
pinned the old order; the orchestrator's full-unit-lane run says none did.- Two previews side by side (
Baklava,Baklava-2, same recorded level) keep a
deterministic order via thelocaleComparetiebreak. resolveSpecwith an explicit--os Baklavastill resolves — the fix changes
only who is newest, never what is listable or leasable.
Written by an agent.
Written by an agent.
- 主要语言
- TypeScript
- 星标
- 20
- 派生
- 2
- 平均合并
- 7 小时 26 分钟
- 30 天内合并 PR
- 253
环境准备
这个项目没有提供开发容器、Dockerfile 或贡献指南,环境需要你自己搭建:先看它的 README,通用步骤见我们的新手贡献指南。
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
callstackincubator/simlock 的其他 Issue
-
Console lane deletes a failed daemon's home, so a start failure names no cause可能已有人在做 @V3RON 今天认领。 未关闭bug:triage
难度 2/5 1-3 小时 新手友好度 76/100
callstackincubator/simlock#673 · 已指派 1 人 ·
维护者通常 1 天内回复
-
bug:needs-info
难度 2/5 1-3 小时 新手友好度 76/100
callstackincubator/simlock#657 · 1 条评论 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 70/100
callstackincubator/simlock#639 ·
维护者通常 1 天内回复
-
bug:new
难度 4/5 3-5 天 新手友好度 45/100
callstackincubator/simlock#684 ·
维护者通常 1 天内回复
-
`worker.rejected` carries `count: 1` on an isolated refusal, which the catalogs say it must never do可能已有人在做 @V3RON 于 1 天前认领。 未关闭bug:ready
callstackincubator/simlock#677 · 已指派 1 人 ·
维护者通常 1 天内回复
查看 callstackincubator/simlock 的全部 Issue
相似的 Issue
-
Unsampled pageload span is profiled when profileLifecycle is 'trace' (browser UI profiling)可能已有人在做 @Jxxunnn 今天认领。 未关闭Browser Waiting for: Product Owner
难度 2/5 1-3 小时 新手友好度 82/100
getsentry/sentry-javascript#25497 · 1 条评论 ·
维护者通常 1 天内回复
-
e2e-failure ready-to-code
难度 2/5 1-3 小时 新手友好度 77/100
redhat-developer/rhdh-plugin-export-overlays#4438 · 1 条评论 ·
维护者通常 1 天内回复
-
resource: destroying a resource calls AbortController.abort() without a reason, creating an AbortError with a stack trace each time可能已有人在做 关联的 PR 仍在进行中或已合并。 未关闭area: core
难度 2/5 1-3 小时 新手友好度 75/100
angular/angular#71422 · 1 条评论 ·
维护者通常 1 天内回复
-
bug status: needs triage
难度 2/5 1-3 小时 新手友好度 72/100
facebook/docusaurus#12739 · 1 条评论 ·
维护者通常 1 天内回复
-
factory-active task-identify-harness-labels-done task-identify-issue-type-done
难度 1/5 1-3 小时 新手友好度 68/100
维护者通常 1 天内回复