Hacker News Reader

Comment by akshayB | Hacker News Reader

SQL is around since the dawn of relational database and its hard to replace. The best option for mass adoption is to have drag and drop tools with visualizations like no-code ETL. Template like and markup language or framework are easier to adopt for new developers but majority of the population still tend of stick with the original language.

View on HN

> The best option for mass adoption is to have drag and drop tools with visualizations like no-code ETL.

No. I've worked with BI tools and when things get complicated you end up needing to go back to text to express the weird bits and every company has a few queries with weird bits in it.

I also will gladly agree that it's network effect is what makes it so hard to replace (as opposed to some perceived perfection of the language - it definitely isn't perfect) but SQL has evolved significantly over time. Core SQL hasn't - but Postgres in particular has pushed the envelope on what can be done with WINDOWs, CTEs, and aggregate modifiers. I think it's a bit misleading to say the majority of the population still tend to stick with the original language since, at a previous job we did attempt to write "neutral SQL" that would execute on MSSQL, Postgres and MySQL - but in most shops you'll have a chosen dialect and you'll be able to make use of more recent and advanced language features... So the majority of the population is using modern SQL just like the majority of programmers that'd describe themselves C/C++ programmers can't grok ANSI C.

If we still use math 100 years from now, we'll still use SQL. It's a fine way to query relational data, and relational data is a fine way to model reality.

I'd like to hear from people that think I'm wrong.

> If we still use math 100 years from now, we'll still use SQL.

Those are incongruent. Do you mean Western notation?

> It's a fine way to query relational data

It's quirky, but good enough for ad-hoc queries that I think it will be hard to overcome the momentum in that area.

It's not fine for application work, where you need things like composition. We've tried to solve those problems with ORMs, but the ORM is starting to fall out of fashion due to a number of problems of its own. SQL is not a great compiler target. I do eventually see something lower level built for programmers, not data analysts, rising up here. If SQL is compared to Javascript, something akin to WASM, perhaps.

ORMs, when best used (and we use them even though we're pretty SQL literate and maintain a lot of SQL) will survive forever, nothing beats an ORM for really dirt simple expressions that you want to be trivially testable. Never in my life do I want to see someone write an UPDATE query against a single table with no shenanigans with dynamic field support using string gluing to properly stitch in all the columns - this is something a known tool can do better, this is a great opportunity for an ORM.

A non-great opportunity for an ORM is anything I'd call a "report query" (some complex read-only query involving a lot of JOINs, a bunch of WHERE clauses and possibly some nested aggregation for funsies) - this is where you pull out the SQL (or alternative query language!) because an ORM will struggle to properly support all the functionality you need and because trying to tune a query being produced by an ORM (even just to make sure it's well aligned with logical indices) is a task that yields nothing but endless frustration.

Never in my life do I want to see someone write an UPDATE query against a single table with no shenanigans with dynamic field support using string gluing to properly stitch in all the columns

These are just convenient features that most ORMs provide and can exist entirely outside of ORMs, they are not the primary purpose of ORMs.

You are correct by their design. But by usage I've found that to be by far the most valuable thing that ORMs deliver. Making use of ORMs to power an ActiveRecord system in your codebase has only ever, to my observation, lead to pain. Querybuilders that are equipped with more advanced functionality around type security and response decoding are quite a valuable tool.

The GP is referring to SQL’s mathematical roots: https://en.wikipedia.org/wiki/Tuple_relational_calculus

Then the statement resolves to "If we still use math 100 years from now, we'll still use math.", which is a rather silly statement. SQL and Western notation are interesting to compare in that they are the dominant, but not exclusive, languages used to describe their respective mathematical domains.

I meant "still use math" as a proxy for "still use formal languages to communicate".

Maybe in 100 years AI will be so powerful that we'll just ask our question in natural language and get the answer we need. Or maybe in 100 years AI will have harvested us for the iron in our blood. Either way we wouldn't need SQL anymore.

SQL does not fully implement Relational Algebra or Calculus, which are isomorphic.

See Many of Chris Dates' books and things like Tutorial D which do meet the ALgebra.

SQL is near enough the theory to work and also has had so much effort put into making it work fast, reliably and scale for volume that a new language has too much to overcome even if it can deal with all cases. So it won't be replaced soon.

However 100 years is longer than SQL had been around so a proper relation server could come around, there is just too much uncertainty.

It's pretty obviously silly to have an unavoidable text parser in between code and data. We've reached the point where it's fairly low overhead, but it's still not nothing.

> SQL is around since the dawn of relational database

There was a decent amount of competition back then, though. It seems the industry eventually settled on SQL to be compatible with Oracle's dominance. Postgres didn't gain SQL support until about a decade in.

What were some of the alternatives? What did Postgres start with?

Postgres used QUEL in the early days, as did its predecessor Ingres. MRDS, the first commercially available relational database, used a language known as Linus. Alpha was the language originally envisioned by Codd to describe his relational model.

Code source control is a vital aspect of software development in the modern era, and no-code tools are incompatible with that unless they are also able to output their representations as plain code so that tools like "diff" work as expected, in which case you might as well stick to SQL.

That is so true! Have you ever used SSIS? A really powerful tool, but even small changes can cause hundreds of changes in the underlying XML, which makes change control a nightmare. Forget branching and merging anything other the most minor changes.

That is a limitation of text based tools.

There have been code source control tools based on the AST see Envy for Smalltalk.

Hopefully eventually we will dump the limitations of text based tools and use one based on the structure of programs. I don't want to know line 123 has changed I want to know that function fn in module m has changed or that function X was added on this date.

Because of this reason, vast majority of new generation query languages are translated into SQL but in fact it is not a great language as a target language. I think SQL should more focus on features as an efficient intermediate language rather than adding more and more ad hoc "convenient" features that don't really play well with other language features...

Reply on news.ycombinator.com