Commit Graph

2 Commits

Author SHA1 Message Date
goodolclint-claude[bot] 0026ef2b57 refactor: typed models for the remaining dictionary-returning services (#157) (#229)
NodeService.GetNodeConfig/GetNodeDns, ClusterConfigService.GetClusterConfig,
BackupService.GetNotBackedUp and VmService.GetGuestExecStatus returned raw
Dictionary/List<Dictionary> instead of a Pve* model, per issue #157. Each now
has a typed model under Models/{Nodes,Cluster,Backup,Vms}/ with [JsonProperty]
for documented fields and a [JsonExtensionData]-backed AdditionalProperties
catch-all, following the PveVmConfig pattern. The five consuming cmdlets and
their [OutputType] attributes are updated to match.

GetClusterConfig also fixes a latent bug: GET /cluster/config returns a JSON
array (a directory index), but the old code did `data is JObject obj ? ... :
empty dict`, which silently always returned an empty dictionary since data was
a JArray. PveClusterConfigEntry decodes the array correctly and exposes a
typed Name property (the array items' schema documents no named fields, but
the endpoint's "links" metadata gives the child-URL template as "{name}").

The guest-exec poll loop in InvokePveVmGuestExecCmdlet keeps its exact
Stopwatch + Thread.Sleep(1000) structure (ADR 0001 accepted exception); only
the type it reads from changed. A TolerantBooleanConverter was added so
PveGuestExecStatus.Exited keeps accepting PVE's boolean/integer/string forms,
matching what ApiValueHelper.IsExited already tolerated for the old
dictionary path.

Reviewed with codex-rescue, correctness-reviewer and api-compat-reviewer
before commit; both real findings above (the name/subdir key and the Exited
string-form regression) came from that pass and are mutation-tested.

Co-authored-by: goodolclint-claude[bot] <323206664+goodolclint-claude[bot]@users.noreply.github.com>
2026-09-03 19:16:29 +00:00
goodolclint-claude[bot] 109dca657a refactor: make the offline Pester suite able to fail (#153) (#205)
Skip-IfMissing skipped a test whenever the cmdlet under test was absent
from the build, including the test asserting it exists. Every cmdlet
compiles into one assembly, so there is no partial-build case for it to
serve; all it did was hide a missing cmdlet. Delete the 37 copies and
every call site, and drop the residual conditional skips in the same
family (CmdExists probes, an attribute-presence skip in SdnCmdlets, a
lifecycle helper skipping on a cmdlet-name collision fixed long ago).

894 It blocks only reflected [Cmdlet] and [Parameter] attributes back at
the compiler: existence, CommandType -eq 'Cmdlet', Parameters.ContainsKey,
IsMandatory reflection, and per-file CmdletsToExport asserts for whichever
names an author remembered. Nothing reflected over the built assembly, so
a new cmdlet missing from the manifest shipped invisible. One data-driven
file replaces them: it diffs CmdletsToExport against the assembly's cmdlet
types in both directions and asserts the conventions reflection can see.

Behavioural tests are untouched: no-session errors, binding rejections,
ShouldProcess and -WhatIf, ConfirmImpact, ValidateSet and ValidateRange
values, parameter types, positions and pipeline binding.

The generated help covered 169 of 194 cmdlets. Regenerated with the repo's
own generate-help.ps1: 25 new markdown stubs, 13 existing docs picking up
parameters added in earlier waves, and a rebuilt MAML.

Co-authored-by: goodolclint-claude[bot] <323206664+goodolclint-claude[bot]@users.noreply.github.com>
2026-09-03 00:54:02 +00:00