Investigate AI-agent analysis report on the gridfs specification. This ticket tracks findings identified during a systematic review of the spec: missing test coverage, ambiguous normative language, and spec/test inconsistencies.

Missing Tests

• Index creation is idempotent: MUST check whether indexes already exist before creating them
• Numeric index values MUST be compared by value across types (1 vs 1.0) — type normalization not tested
• Indexes MUST be created in foreground mode — mode not verified
• Zero-length file upload MUST NOT create any chunks — verify no chunks document inserted
• Drivers MUST NOT run the filemd5 command — negative test
• Orphaned chunks from a failed upload MUST NOT be cleaned up by the driver
• Each chunk received during download MUST be validated as the next expected chunk (sequence check)
• Extra chunks behavior: drivers MAY ignore them — not tested (neither error nor silent ignore)

Ambiguities

• Bucket name with dots: If a user specifies bucket name fs.backup, are collections fs.backup.files or
  fs.backup.files.files? Tests only use simple names.
• chunkSize as non-integer type: A chunkSize stored as double with fractional part (e.g., 262144.5) — what
  happens during chunk processing? Spec requires numeric comparison but doesn't address fractional chunk sizes.
• uploadDate boundary: "MUST be the datetime when the upload completed, not when it was begun." For streaming
  uploads, is "completed" when close() is called or when the server acknowledges the files collection insert?
• Zero-length file download: Spec says MUST NOT query chunks collection — but what is returned? Empty stream?
  Exception? Spec is silent.
• start == end in partial retrieval: start > end is an error, but start == end returns no data. Tests don't
  clarify this boundary.

Inconsistencies

• metadata: null vs. omit metadata: Spec says "If not provided the driver MUST omit the metadata field."
  options=null creates ambiguity: omit the field or set to null document?
• Index creation ordering assumption: Spec says check if files collection is empty with findOne before creating
  indexes. But findOne on a non-existent collection implicitly creates it. Was the write "before the first write"?
  The ordering assumption is unclear.

Notes

• 8 unified test files covering upload, download, delete, rename, and disableMD5 variants.
• MD5 is deprecated; disableMD5 option tested separately.
• Tests don't distinguish Basic API (MUST support) from Advanced API (MAY support).
• DRIVERS-2062 (Backlog): modify GridFS spec to support sessions,
  transactions and causal consistency.
• DRIVERS-3442 (Blocked): increase timeoutMS values in CSOT GridFS
  tests.