apache / apache/cloudstack-go

Generator fabricates list-response JSON wrapper tags from parseSingular(), causing recurring silent-nil-slice bugs

未关闭
#155 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看
主要语言
Go
星标
39
派生
47
PR 合并指标
30 天内没有已合并 PR

描述

Splitting this out as the systemic root cause behind a recurring class of bugs (#135, #136 / #137, and the four just fixed in #154).

### Problem
For `ListResponse` structs, `generate.go` derives the JSON wrapper tag from `strings.ToLower(parseSingular(listName))`. But the key CloudStack actually returns is set server-side via `setObjectName(...)`, and it is frequently **not** what `parseSingular` produces:

- camelCase: `listHypervisorCapabilities` → server `hypervisorCapabilities`, generator `hypervisorcapability`
- `-xes` plural: `…Ipv6Prefixes` → `Prefixe` (also corrupts the element **type name**), server `guestnetworkipv6prefix`
- hand-chosen keys: `listLBHealthCheckPolicies` → server `healthcheckpolicies` (drops the `lb`, stays plural), generator `lbhealthcheckpolicy`

When the tag is wrong, `json.Unmarshal` populates `Count` but leaves the slice **nil** — silent data loss — and the generated `GetByID` helper then panics on `slice[0]`.

### Current mitigation doesn’t scale
`generateResponseType` already carries a hand-maintained override `switch` (~30 cases, incl. the metrics ones) that exists precisely to patch this. Each new case is added **reactively**, only after a user hits a nil slice. #154 adds four more; there are almost certainly others not yet reported.

Note: `listApis.json` (the generator input) does **not** expose the list wrapper object name — it carries each item’s `response` fields and `related`, but not the `setObjectName` value — so the correct key can’t simply be read from there.

### Proposed directions
1. **Harden `parseSingular`** for the regular English plurals it currently mishandles — `-xes/-ses/-ches/-shes` → drop `es` (it already does `-ies`/`-sses`). This fixes the `Prefixe` type-name corruption and several tags for free, shrinking the override set.
2. **Make the override set verifiable instead of reactive:** add a CI/test check that asserts each `ListResponse` tag against the keys in the `test/testdata/*.json` fixtures (and/or a captured real response), so drift is caught at build time rather than by users in production.
3. Optionally, a curated `command → objectName` override map (cleaner than inline `switch` cases) seeded from the verified keys.

I verified the #154 keys against the CloudStack server source (`setObjectName`) and have a 4.22 environment; happy to take on (1) and/or (2) if a maintainer signals a preferred direction.

贡献指南

这个仓库没有索引到贡献指南

调研方向

从 generate.go 开始,重点查看 parseSingular 和 generateResponseType,并检查现有的 override switch。将生成的 ListResponse tags 与 test/testdata/*.json 中已验证的 keys 以及服务器端的 setObjectName 值进行比较;listApis.json 不包含这些 wrapper 名称。完成的标准是:所提出的复数处理和/或验证方案能够防止静默的 nil slices,而不是仅依赖反应式 overrides。

由索引模型根据 Issue 内容生成。

评估

技术栈
go
领域
api, tooling
Issue 类型
缺陷
难度
4/5
预计耗时
3-5 天
活跃度
冷清
描述清晰度
基本清楚
新手友好度
45/100

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。