Hacker News Reader

On the first point, I don't have either of those books, but IMO the Wikipedia complaints aren't compelling.

On the second point, many SQL databases offer indexing [1] (as you said), clustering [2] (arranging a table according an index), and partitioning [3].

I think database designers are very conscious of the difference between the logical relational model and the implementation details. For example, this [4] conversation on the PostgreSQL mailing list about unique indexes vs. unique constraints. I also think that in practice there just aren't many other ways to abstract the logical model from the physical storage that gives you something substantial.

IMO, this is one of the successes of the relational model: the logical model and the physical implementations don't need to differ for you to have a really good situation all around.

[1] https://www.postgresql.org/docs/10/static/indexes-intro.html

[2] https://www.postgresql.org/docs/10/static/sql-cluster.html

[3] https://www.postgresql.org/docs/10/static/ddl-partitioning.h...

[4] https://www.postgresql.org/message-id/CAO8h7BJMX5V1TqzScTx2N...

the point is though, it's a little bit like a broken calculator. "The Relational Model" represents a logical model as a kind of pure piece of mathematics, and presents a system that isn't just dump disk persistence but actually a logic induction engine. However, this is not how SQL is used in practice. It is possible to implement the relational model in SQL, however if you assume SQL databases are supposed to be implementations of it in and of themselves, they violate the commutative, associative and distributive laws of the relational algebra- it's a bit like a calculator that sometimes gives you 2+2=5 (because it's faster to give wrong results sometimes). In addition, the ergonomics of a typical SQL database system do not make mathematical/logical soundness the easiest thing to achieve. Most developers don't know or care about logical induction or mathematical soundness, they just want a place to dump their objects.

That said, out of the open source SQL database systems, postgresql actually does try to implement the relational model properly. But I don't have enough direct experience with it to say definitively how well it does.

How many millions of dollars hss NULL != NULL cost the world?

> what is better?

well, obviously "worse is better". historically, SQL has been "Good enough" and was sponsored by IBM/Oracle that nobody has bothered to try and implement the relational model correctly, except for perhaps postgres (which used to have its own query language before bolting SQL onto its core engine)

what I hope people get out of my comments is not about selecting the right system for today- A better thing does not necessarily exist yet-- But at least understand the relational model well enough to see what its utility is. Functional programming and Logic programming are coming back in fashion, and maybe someone here will build the better thing.

I don't think "nosql" databases or graph databases are necessarily the better thing. (but they might be for certain usecases). Though I think relational db systems could learn something about developer ergonomics from them. Interfacing with a DB using a fortran inspired query language is archaic. Constructing the queries using string concatenation has been a security nightmare. You could build something waaayyy better off of prolog or mercury, full programming languages with the inductive logic engine at the core of their paradigm, with a database system integrated in, and expose a graphql api for your isomorphic react web app cos that's how we do in 2018.

Even microsoft's linq is a massive improvement over using SQL. Language integration, strong typing. These are things we like, right?

> well, obviously "worse is better".

> A better thing does not necessarily exist yet

Ah, of course.

> Interfacing with a DB using a fortran inspired query language is archaic. Constructing the queries using string concatenation has been a security nightmare.

Perhaps, but this this often for very ubiquitous technologies. See almost every HTML authoring tool.

You can write full application servers in Oracle PL/SQL now. [1] And with PostgreSQL's row-level security, you can easily have a GraphQL server with PostGraphile [2]. But those aren't popular solutions.

> ou could build something waaayyy better off of prolog or mercury, full programming languages with the inductive logic engine at the core of their paradigm

That doesn't really change the equation. You still either have to (1) choose your DBMS's query language to write your application server (like today) or (2) use one language to construct queries in another (like today).

I won't argue that a Prolog DBMS wouldn't be nice. I will argue that in terms of the hard parts of a DBMS, it's not going to fundamentally change things.

[1] https://docs.oracle.com/cd/B28359_01/appdev.111/b28424/adfns...

[2] https://www.graphile.org/postgraphile/

first off, thanks for the links. Secondly,

> That doesn't really change the equation. You still either have to (1) choose your DBMS's query language to write your application server (like today) or (2) use one language to construct queries in another (like today).

I don't know if you're misunderstanding me, or I'm misunderstanding you. prolog is a query language. The logical programming paradigm is that a program is a "list of facts" and you run the program by "querying" the state of the world. The only piece missing is making the fact storage engine scalable/robust/production ready. Or.. alternatively, there's the Eve[1] project which starts off with prolog to make a hypercard like environment with a built in data storage layer. Its aim is less production ready apps and more developer ergonomics.

[1]: ( http://witheve.com )

I don't think I misunderstood.

You yearn for the day when a DBMS uses Prolog for its query language.

Users of that DBMS have the choice to either (1) query seamlessly because they use prolog for the entire application (2) query by "gluing" Prolog to their application language.

The main problem that I see is that relational model / sql is ALWAYS coupled with RDBMS, even with the proponents of the relational model!

The thing is, sql is much less capable then even forth. You can't do:

    query = where a = ?1
    columns = [name, id ]
    console.write(columns ?rename name as a |> query)

not even

    console.write

not even

    Window.open

neither store table in a field (not serializate it, store) create your btree, make open gl game, make a osx app, create a rdbms and so on. (yes, i'm aware some rdbms could, hacking around, kind of do it witth pain)

In short, the relational model is not see as something that can be made for a full language WITHOUT RDBMS.

That is what I wish to change:

http://tablam.org/

because I work before as foxpro developer and see how easy and logical was to work with tables without sql.

I agree the relational model can be very useful.

I will often keep a lot of queries in the DMBS, or even create a :memory: database with sqlite when I want to use relational logic.

Thanks for the link.

I agree that the integration of domain programming and database programming/querying in FoxPro/xBase created a smoother CRUD-app shorthand that current tools can't match. Less converting of sh8t between the layers (and less need for layers). The newest generation must like to type and read a lot of code, I guess. Or, it's a conspiracy from Carpel Tunnel and eye doctors to get more revenue.

> Interfacing with a DB using a fortran inspired query language is archaic.

Possibly amusing aside: Logica's RAPPORT, one of the earliest RDBs, was written in Fortran and was queried (or could be) via pre-processed Fortran. I don't know whether that was more faithful to the model than SQL.

Jaffer's Scheme RDB (https://people.csail.mit.edu/jaffer/slib/Relational-Database...) aimed to implement the model, possibly uniquely so at the time, he thought.

Actually, SQL seems more COBOL-inspired than FORTRAN-inspired to me.

As far as a "generic" query language not tied to relational databases, that's a rather tall order. Logic-based query languages are difficult to troubleshoot when you don't get the results you expected. It's kind of like a formalized version of Google: you get results, but you don't know if they are the best results, you don't know what you are missing, and you often don't know why you got what you did.

You have to have some kind of discipline and structure to the data in order to understand and troubleshoot results (at least for mortals and non-Sheldons). Some say relational is "too stiff" for some uses, but there is the Dynamic Relational proposal that gives it flexibility without abandoning relational and SQL. (Some argue it's not technically relational, which I don't agree with, but that issue doesn't change the similarity to existing RDBMS & SQL conventions.)

As far as relational-based alternatives to SQL, consider these discussions:

http://wiki.c2.com/?HowOtherQueryLanguagesAddressSqlFlaws

http://wiki.c2.com/?QueryLanguageComparison

http://wiki.c2.com/?ExperimentalQueryLanguageComparison

I agree with all you said, except this one:

> they violate the commutative, associative and distributive laws of the relational algebra

I follow the idea of relational set, but implementing a relational language (http://tablam.org, alpha btw) I see why bags surely are better. Is not a mistake rdbms take it. Make everything sets demand more memory and runtime costs. You need to couple a index or pay O(N) for each mutation and get only a few upsides.

If exist a way around this, truly I wanna know.

However, exist another reason for what sets are wrong (as being default) them fail to cover the most simple and yet more powerful capability of the relational model: Model data.

Is posible to argue than [a, 1; b, 1] must be a set and order not mater. But [a, 1] & [b, 1] are both relations, right? also are [a, b] & [1, 1]. Is wrong to make [1, 1] to [1] just to get a math purity.[1, 1] is valid. Say that duplicates are wrong is non-sense. Also say that "order not matter" is non-sense too.

Ironically, the solution for this? add another column just to keep thins as set.

> Most SQL systems provide a 1-to-1 mapping of tables

That is a bit weird criticism, since the whole point of relational databases is that the physical storage structure is abstracted away. SQL does not prevent databases from using different storage structures.

> Most SQL systems provide a 1-to-1 mapping of tables to underlying storage structure which places limits on performance

Is that the fault of the SQL language?

Also, serious DBMSs are generally engineered by pretty serious people, no? Are they really implementing such a mapping for no good reason?