Merged PRs
dolt
- 8723: If a JSON document contains strings that can't fit in a single chunk, use the naive Blob chunker instead of the smart JSON chunker.
The JSON chunker never creates a chunk boundary inside of a string.
Originally, this PR added functionality to allow the JSON chunker to split JSON document inside a string. This was supposed to be safe and backwards compatible, because older versions of Dolt reading documents written by newer versions of Dolt are supposed to fall back on ignoring JSON document metadata if they don't understand it and treat the document like a blob.
However, tests revealed that older clients were not checking for this in enough places and would hang when trying to read documents written with this fix. This PR also contains fixes to check the JSON metadata in more places... but this doesn't do anything for existing Dolt servers running older versions.
So instead, this PR detects when a document contains strings that exceed some limit, and instead the writer falls back on writing the document as a plain blob without metadata. The limit is currently 32KB, but can be raised in the future.
I chose to keep the logic for splitting JSON documents inside a string, although the chunker doesn't currently use it, since we may decide to enable it in the future.