A Relational Model of Data for Large Shared Data Banks by E.F. Codd (1970) [pdf]
seas.upenn.edu
seas.upenn.edu
> The true importance of relational theory is highlighted by the title of the original (and considerably shorter) version of Codd’s first paper. That version predated the published version by a year, and the title was “Derivability, Redundancy, and Consistency of Relations Stored in Large Data Banks.” The title of this unpublished version emphasizes that the real importance of relational theory is that it provides a rigorous method of asserting arbitrarily complex consistency constraints that must be satisfied by the data within the database. Strange as it may sound, Oracle Database did not enforce referential integrity constraints until Version 7 was released in the 1990s (by which time Oracle Corporation was already the world’s largest database company)
https://iggyfernandez.wordpress.com/2013/04/24/whats-so-sacr...
Extending the Data Base Relational Model to Capture More Meaning by E.F. Codd IBM Research Laboratory published in Volume 1 Number 1 (March 1979) issue of AUSTRALIAN COMPUTER SCIENCE COMMUNICATIONS, which incorporated papers from the 2nd Australian Computer Science Conference 1-2 February 1979 in Hobart, Tasmania.
http://www.retrocomputingtasmania.com/home/projects/history/...
Taking Codd serious suggests using relations inside programs as well.
There would still be a bit of a mismatch between relations held locally - a kind of disconnected data set - and relations in the persistent store. Synchronization, local-remote joins, transactions - there are lots of details to get right.
It's not something trivially added on via a library either, not if you want to do it right (properly typed, abstracted at the right level, e.g. creating ASTs to represent the computation).
And not only type system, but also the AST ideas you are talking about. Check out higher-order abstract syntax.
> One problem is that making use of relations involves very high level descriptions of computations, and computing them efficiently requires a lot of smarts - you'd risk baking half a database engine into the runtime.
I wonder if a variant of Greenspun's tenth rule applies? "Every sufficiently complex program (without formal relations) contains a half-baked database engine."
It has alread been done (and, unlike Standard Chartered's version, no weird and non-composable type-level programming required)
https://hackage.haskell.org/package/opaleye
Disclaimer: I wrote Opaleye, but most of the "embed statically typed relations in Haskell" leg work was done by HaskellDB.
Hmm, it says:
> An SQL-generating DSL targeting PostgreSQL. Allows Postgres queries to be written within Haskell in a typesafe and composable fashion.
Genering Postgres queries sounds very different from using relations as a datatype inside my program. Or is that feature buried somewhere deeper?