fix(version): heal ContentType for legacy octet-stream files on labelled version

[kptdobe]

Summary (revised after board pushback)

POST /versionsource returned a silent 500 for legacy imports whose source object was stored with ContentType: application/octet-stream (or no ContentType). The diagnostic from #284 confirmed 9/24h occurrences with that identical fingerprint.

Fix (was A, now B)

Original proposal (A) was to widen the gate: createVersion = shouldCreateVersion(contentType) || update.label != null. Board flagged that we should fix the underlying ContentType rather than let bad metadata through.

This PR ships fix B: in postObjectVersionWithLabel, infer the correct mime from daCtx.ext when the stored ContentType is missing or octet-stream, then pass that inferred type through update.type.

// src/storage/version/put.js
function inferVersionableType(contentType, ext) {
  if (shouldCreateVersion(contentType)) return contentType;
  if (ext === 'html') return 'text/html';
  if (ext === 'json') return 'application/json';
  return contentType;
}

Effects:

• Legacy .html files stored as octet-stream now get a labelled version. The version snapshot is correctly tagged text/html.
• The main object's PUT (the one that always runs in this code path) writes the inferred ContentType into S3 metadata, so the file is self-healed for all future requests. Auto-versioning on the next plain PUT now works because shouldCreateVersion(text/html) returns true.
• The shouldCreateVersion gate is unchanged. Binary files (jpg/pdf/mp4/etc.) still cannot be labelled-versioned, matching the project's existing "binaries do not version" semantics that JPEG/PNG/PDF/MP4/SVG/ZIP tests already assert.
• The diagnostic log from fix(version): log diagnostics on POST /versionsource silent 500 #284 stays in place and is extended with inferredType + ext, so any future unhealed pattern is still observable in Cloudflare Logs.

Tests

• heals legacy octet-stream HTML on labelled version: snapshot + main object both repaired - asserts the version snapshot has ContentType: text/html, the main object PUT writes ContentType: text/html, the audit entry is emitted with versionLabel + versionId.
• logs diagnostics and returns 500 when labelled version requested on non-versionable ext - asserts the silent-500 path still 500s for unknown extensions, and the diagnostic log captures contentType / inferredType / ext / hadLabel / currentStatus.
• plain PUT (no label) still skips auto-version for non-html/json contentType - companion to confirm the labelled-path mime inference does not bleed into auto-versioning.
• All existing tests pass; binary-never-version tests untouched.

npx eslint src test  ->  clean
npm test             ->  393 passing (audit timeout flake unrelated)

Risk

• Mime inference is narrow (only html/json). Files with truly unknown extensions still 500, which is consistent with current behavior.
• Storage cost increases for the labelled-version writes that previously failed (bounded by the actual call rate, <=9/24h).
• The S3 ContentType is now mutated when we heal a legacy object. Mutation only happens on the labelled-version code path, which already PUTs the main object as part of its normal flow. No new write.

Refs

• COR-55 (this fix), COR-46 (diagnostic ship), fix(version): log diagnostics on POST /versionsource silent 500 #284 (diagnostic PR)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[adobe-bot]

🎉 This PR is included in version 1.9.3 🎉

The release is available on:

• v1.9.3
• GitHub release

Your semantic-release bot 📦🚀