Terminal load errors discard the HTTP status the engine already computed

[edde746]

AetherEngine.swift:2343, :2470, :2610:

state = .error("Failed to load: \(error.localizedDescription)")

HLSIngestError is CustomStringConvertible but not LocalizedError, so localizedDescription falls through to Foundation's generic bridge:

description         : playlistUnreachable(500)
localizedDescription: The operation couldn't be completed. (HLSIngestError error 0.)

The status is raised correctly (HLSVODIngestReader.swift:499-501) and rendered correctly (HLSIngestError.swift:17-19), then dropped at the boundary. A server refusing a transcode is currently indistinguishable from a corrupt file.

Ask: state = .error("Failed to load: \(error)"), or conform the error enums to LocalizedError with errorDescription returning description. The second fixes the class rather than three call sites; no preference beyond that.

Scope: load path only. A mid-session failure on the nativeRemoteHLS bypass arrives through AVPlayerItem (NativeAVPlayerHost.handleItemFailed:610-660) and is not addressed here.

[superuser404notfound]

Fixed on main in f8ffa6d, unreleased.

Went with the second option. HLSIngestError and every other error type the engine can throw now conforms to LocalizedError with errorDescription returning description, so playlistUnreachable(500) survives the boundary intact.

Keeping localizedDescription at the call sites rather than switching to \(error) was deliberate: URLError and other Foundation errors reach those same catch blocks, and there localizedDescription is the better rendering. Conforming the enums makes both cases right at once.

The class fix reaches further than the three load-path sites: AetherEngine+Loading.swift:1459 and :1484 publish "Reload failed: ...", :1613 publishes "Audio track switch failed: ...", and SoftwarePlaybackHost / AudioPlaybackHost route "Playback error: ..." through onError. Same defect at all of them.

The audit turned up one case that is arguably worse than the reported one. DemuxerError carried no description at all, and it is the most common failure at that boundary. openFailed(code:) rendered as "(DemuxerError error 0.)" with the AVERROR code gone, which is the whole diagnosis. It now goes through av_strerror:

Demuxer: open failed (Invalid data found when processing input (-1094995529))

Also covered: AVIOReaderError, HLSLocalServerError, ProducerError, MuxerError, AudioBridgeError, DiscError, SubtitleDecoderError, FrameDecodeError, AudioDecoderError, both audio-tap enums, both diagnostics probes, and the two SMB error structs.

ErrorLocalizedDescriptionTests pins the invariant per type, with the reported case verbatim, so an error type added later without conformance fails there instead of in someone's error banner.

The scope note holds as written. The nativeRemoteHLS mid-session failure arrives as AVPlayerItem.error, an NSError from AVFoundation whose localizedDescription is already correct, so that path was never the same defect.

Thanks for tracing this to the boundary instead of the symptom. The raise-site and render-site pointers made it a confirmation rather than an investigation. Leaving this open until it can be retested against a release.

[superuser404notfound]

Released in 6.5.0.

A minor rather than a patch, since LocalizedError on HLSIngestError and the other public error types is added public API, and because the text a host renders from a caught error changes with it.

.package(url: "https://github.com/superuser404notfound/AetherEngine", from: "6.5.0")

[edde746]

Friendly bump now that 6.5.6 is out: this one-liner ("\(error)" or a LocalizedError conformance on HLSIngestError) is the last piece blocking our HTTP-status-aware load-failure UI on the host side — the mid-session case was fixed by #291, so load-time is all that remains of the original report.

[superuser404notfound]

This shipped in 6.5.0 on August 2, so 6.5.6 already carries it (f8ffa6de, present in every tag since, current release is 6.6.2).

Went with the second option from the report: HLSIngestError conforms to LocalizedError with errorDescription returning description, so the load boundary now renders

Failed to load: playlistUnreachable(500)

Same change for every other error type the engine can throw, and DemuxerError got a description at the same time, since it had none at all and openFailed(code:) was dropping the AVERROR code, which is the whole diagnosis. ErrorLocalizedDescriptionTests pins the reported case verbatim plus a sweep asserting that no engine error renders Foundation's bridge text, so a type added later without the conformance fails there instead of in an error banner.

If 6.5.6 still shows The operation couldn't be completed. (HLSIngestError error 0.), that is a new finding and worth chasing. Two things would pin it down: the AetherEngine version your Package.resolved actually resolves to (SwiftPM ignores a dependency's own resolved file and only honours the root's, so a transitive pin can hold an older engine than expected), and the exact string that reaches the UI.

Mid-session on the nativeRemoteHLS bypass stays as scoped in the original report: it arrives as AVPlayerItem.error, an NSError whose localizedDescription was already correct, so it was never the same defect.

Leaving this open for the retest.