themightychris / db-to-git Goto Github PK

This spec may provide the features we need, letting us keep to a standard.

http://specs.frictionlessdata.io/json-table-schema/

Currently they're passed via environment variables, or via the .env file. Feels a bit unnatural, though the only other option seems to be CLI args (username, password, host, etc. or conn-string)

What if primary key values contain nasty characters like /? It should be safe to assume or just require that table and column names contain only characters safe for filenames in git, but primary key values will offer no such guarantees.

In most intended use cases, primary keys should be "slug"-like which would work well, but how can that be enforced as a constraint or something like a / coming up in a primary key value be handled?

Do we just need a lossless-but-still-readable encoding scheme?

1. Just not support and skip them?
2. Use the hash of a record's subtree as its name in the table's tree with a number appended for duplicates as needed?
3. Some sort of numbering or UUID scheme? It needs to be consistent though to minimize noise when diffing two disconnected dumps of the same data