[randomdata]

> If you have two things that are roughly equivalent, it's always better to pick the more popular/standard option.

There is a good case to be made that, for better or worse, single page applications (SPAs) are the more popular/standard option as compared to generated HTML, even if roughly equivalent, for web applications. Does this mean that HN is making a mistake in not adopting that model? You did say always.

It is not like Postgres was a new project and they simply had to pick something. It had been with us for a decade before deciding to take the SQL direction. It could have much more easily continued down the path it was already going. There was a significant amount of cost involved to make the switch.

When Postgres was first created there was no clear winner, but by the time the switch was made it was apparent that SQL was becoming the more popular choice, I'll give you that. Is that the only reason for the move?

Thing is, I'm not sure SQL is even all that popular nowadays. I expect the vast majority of DBMS queries are only using SQL as a compiler target, more likely being written in a class of languages often referred to as ORMs. By your logic, databases should be adopting what best aligns with that programming model rather than clinging to SQL. You did say always, after all.

Granted, there are some half-hearted attempts to do exactly that by some DBMS vendors (e.g. recent Oracle releases support "3D" queries that return JSON structures rather than relations, more closely matching language semantics), but not in any way that appears to be all that serious and definitely not in any kind of consorted way. It seems the SQL crowd still can't get over that there might be an alternative. In fact, often they can't even envision anything other than SQL being possible. I don't know how many times I've heard someone claim that SQL is some kind of fundamental mathematical property.

> What was that?

The most common SQL bug I see in my travels can be distilled down to something like:

    SELECT t1.* FROM t1 INNER JOIN t2 ON t1.id = t2.t1_id

To be fair, I'm not sure the bug would go unnoticed with such a simple query - where one is able to keep the entire mental model in their head. But consider the basis of that query expanded into something complex, where one will struggle to maintain all of the minute details in active memory. That is where I see bugs be introduced all the time, even by experienced developers. A different design – including the originally designed database query language – has proven to eliminate this class of bug.

[wvenable]

> There is a good case to be made that

There is not. Counterpoint: Wikipedia. It's definitely not like this where SQL is used by 99.99999% of everyone and nobody has even heard of QUEL.

> It is not like Postgres was a new project and they simply had to pick something. It had been with us for a decade before deciding to take the SQL direction.

It wasn't even remotely popular until they made that change. In fact, it wasn't even remotely popular until years after they made that change.

Probably the interesting point is that they could make this change and not really lose any necessary functionality. Clearly the query language was not particularly important.

> Thing is, I'm not sure SQL is even all that popular nowadays.

A suggestion so laughable I have no response. SQL has never been as popular as it is right now in history. Even with the introduction of noSQL databases, SQL continues to gain. All boats are floating.

> By your logic, databases should be adopting what best aligns with that programming model rather than clinging to SQL.

Is there something they can adopt? If you presented me with 2 options, lets call one X and other SQL then we could have discussion about adopting one or the other. If this X was indeed more popular then maybe they shouldn't cling to SQL. But the dozens of ORMs across dozens of platforms is not an alternative that can be chosen.

> It seems the SQL crowd still can't get over that there might be an alternative.

If you've been around long enough you wait until someone can bring the receipts.

> The most common SQL bug I see in my travels can be distilled down to something like:

Yeah that is pretty common. Would it be nice if that wasn't an issue? Sure. I'm not arguing that an alternative might not be better, it's just not better enough. These are pretty minor annoyances. If you just made a language that fixes this one issue and left everything else alone, it would be better than SQL, but you wouldn't get anyone to switch. You'd be better off proposing this as a feature addition to SQL and maybe in 20 years you'll be able to use it.

[randomdata]

> SQL has never been as popular as it is right now in history.

As a compiler target, but even then the SQL is compiled to another representation, so if you want to take that angle it isn't SQL that is popular. Turtles all the way down.

> Is there something they can adopt?

What Oracle is doing was already mentioned. I am personally not entirely sold on their technical choices, and feel that they are trying to stay too close to SQL unnecessarily, but at least they are trying something which is more than we can say for a lot of the other vendors out there. I commend them on that.

> If this X was indeed more popular

It isn't what I would call popular, but it is also rather new. It is quite unusual for a technology to become immediately popular. Though, frankly, I'm not sure it could become popular if never adopted by any other engine. Oracle doesn't have a stranglehold on databases, and if anything is losing ground. Popularity can never come before wide distribution.

> If you've been around long enough you wait until someone can bring the receipts.

But now we circle back to the original posts – there is no good vector for anything else to bring said receipts. SQL has things completely locked down, effectively, which you suggest can only be overcome by becoming more popular than SQL, but, again, popularly is impossible before wide distribution. So the only possible way off of SQL is to convince the SQL crowd that an alternative is possible in hopes that they will consider opening things up.

> You'd be better off proposing this as a feature addition to SQL

It is...? I believe it has been since the very beginning, or at least since the early days. It's such a glaring issue, of course it had to be dealt with. But you have to remember to enable it – which most of the time it seems developers forget, at least in my travels.

Of course, by the same token, if you are able to remember such things, what do you need Rust for? You can avoid bugs and security issues in C just fine if you keep a clear head and never make mistakes. So, Rust doesn't do anything meaningfully different from C, except in the places you suggest don't matter because as long as you don't make mistakes you'll never need it, and yet despite that has gained a decent following. Why it but not some SQL alternative?

[wvenable]

> As a compiler target

Not as a compiler target. I doubt most projects exclusively use ORMs and never use SQL as a query language as intended. Also it's not just programmers using SQL inside of programs -- I teach SQL to accountants and they run their own queries. Any SQL alternative needs to be something accountants can learn.

I use an ORM all the time but I wouldn't use an ORM for reporting queries. Absolutely nobody abstracts over SQL for general purpose querying.

> But now we circle back to the original posts – there is no good vector for anything else to bring said receipts. SQL has things completely locked down, effectively, which you suggest can only be overcome by becoming more popular than SQL, but, again, popularly is impossible before wide distribution.

You've got it.

> So the only possible way off of SQL is to convince the SQL crowd that an alternative is possible in hopes that they will consider opening things up.

That's not how things work. You build something amazing and find a way to make it useful today. This is nothing new, it has been done through the history of technologies. This is how we get new programming languages, new protocols, and new standards. You could have said the same thing about x86 but ARM found a niche and expanded it. Or JSON vs. XML.

> It is...?

Is it? I don't know what you mean.

> Why it but not some SQL alternative?

I never said it couldn't be done -- it just hasn't been done. Any alternative to SQL has to have all the positive features of SQL and be a significant and obvious improvement. The majority of alternatives I've seen are intent on being radical redesigns. Rust, as you said, is not really radically different from C++ -- they just removed most of the footguns and design problems.

Any alternative to SQL needs to find a way to be immediately useful today. This is how NoSQL solutions became popular; the solved a real problem and I could use it today within an existing environment. But NoSQL is not a replacement for SQL -- it was just a replacement for tasks that an RDBMS was never good at in the first place.

But I think the idea of someone just building a better query language in their basement, posts it to Github, and we should all just move over is pretty unrealistic. I could build one of those tomorrow if I wanted.

[randomdata]

> Not as a compiler target. I doubt most projects exclusively use ORMs and never use SQL as a query language as intended.

If projects use 90% ORMs and 10% SQL, ORMs are still widely more popular. This exclusively you are trying to draw is nonsensical.

> Any SQL alternative needs to be something accountants can learn.

Try reading the thread next time. The entire discussion is about how SQL's biggest screw up is the lock-in. The better solution would clearly allow SQL to coexist. If accountants want to use SQL, there should be nothing holding them back. You can still write code in C too, if you want to. The existence of Rust didn't put an end to it, nor should it.

> Is it?

It is. You commonly see the problem as said earlier, but never deal with it, just leaving the bug there?

> This is how NoSQL solutions became popular; the solved a real problem

Indeed, they offered an alternative API which solved the real problem that is known as SQL – hence why the movement became known as NoSQL (i.e. literally "no SQL"). Unfortunately, most of the implementations that offered better APIs fell short in the actual database-y bits, souring the effort, which has ultimately also quashed any progress in improving the APIs over databases you can use.

> Any alternative to SQL has to have all the positive features of SQL and be a significant and obvious improvement.

Therein lies the problem here. Rust brings no significant and obvious improvement over C. It might be marginally easier to use, and might prevent a bug here or there, but in the end it doesn't do anything radically different from C. The same can be said of most SQL alternatives. Yet Rust "made it" regardless where the others haven't.

Of course, Rust had the power of marketing behind it. Being a Mozilla project originally, it had a built-in audience, and the flashy video they released extolling its virtues back in the early days won sufficient hearts and minds. Other than maybe Oracle who at least seems to be giving some minor effort, the big names are stuck on SQL.

Although credit to SQLite for recently documenting its bytecode format. While they have not committed to it being a stable API, and probably isn't suitable as an API in its current form, it does provide a framework in which to start considering a "WASM" for database engines. This will undoubtedly happen someday if we keep the pressure on the SQL crowd to change their ways. And, indeed, they are evolving. The "SQL is a fundamental mathematical property" claims have diminished greatly over time, and they are even starting to recognize that there have been alternatives, like QUEL. It's a slow road, but one that continues.

> I could build one of those tomorrow if I wanted.

And I'm sure it would be the greatest thing the world has ever known, but since nobody knows who you are, there is a fat chance of anyone ever coming to realize that it exits. It is not sufficient to just be significantly better.

[exnihilo21]

> Yeah that is pretty common.

What is the bug?