Why doesn't Sqlite have any viable alternative in the embedded RDBMS space?

Posted by pyeri@reddit | learnprogramming | View on Reddit | 4 comments

The general thumb rule in open source is that we always have alternatives. We have Debian and Ubuntu and Fedora and opensuse. We have mysql, mssql and postres for a reason. Even gcc has an alternative in clang. But when it comes to an embedded RDBMS, Sqlite is the only one available to the best of my knowledge. There is a Java experiment called H2 but I don't think it's used much. Then there are BerkelyDB and RocksDB too but they're key-value stores, not RDBMS. And it's not as if Sqlite is perfect, in fact far from it: 1. Strong Typing: Sqlite types are quite fickle, `datetime` doesn't really act like a datetime, you can store texts into numeric fields, etc. 2. Non-standard SQL: Instead of concat function, we have || operator. 3. Concurrency: Sqlite cannot handle too much concurrency (multiple simultaneous writes). And while the general advice is to use mysql or pgsql, what if I absolutely want to stick to an embedded or file-based database? Let me admit right away my lack of knowledge in system programming and what goes into the intricacy of building the low level nuts and bolts of an embedded RDBMS like Sqlite could be. I'm just curious why no efforts were made to work on any alternatives to Sqlite unlike most other software.

4 Comments

[-]

whoShotMyCow@reddit

Talk is very cheap, PRs are very costly 😔

nutrecht@reddit

> There is a Java experiment called H2 but I don't think it's used much. There's Apache Derby, HSQLDB, H2, that's 3 already in the Java space. And the reason these are not "used much" is simply because in-process databases just don't have that many usecases. > system programming These are just libraries, really not "low level" like an OS is.

hismuddawasamudda@reddit

DuckDB? Although it's in the analytics space (column store) it is SQL oriented.

spicy_tuna_code@reddit

I'm not sure that's actually a rule in open source. OpenSSL doesn't have any alternatives. glibc doesn't have any reasonable alternatives. Debian, Ubuntu, Fedora, and OpenSuSe aren't completely separate operating systems, though. They all use exactly the same kernel. They all run 99% of the same packages. They're different distributions of free software - different collections, in a sense. Each has some original code, but they all sit on top of the same foundations. That model is actually far more common in open-source, where a package forks and there are different variants of it floating around - like ffmpeg and libav, or MySQL vs MariaDB. That's the case for SQLite, in fact - there are several sqlite forks that add some interesting missing features. As for why there isn't a completely separate alternative to sqlite, it's because sqlite is really, really good and a competitor would have an enormous amount of work to do to beat it. SQLite is pretty much the gold standard in terms of testing. The library is 155k lines of code and 92 million lines of tests. [https://www.sqlite.org/testing.html](https://www.sqlite.org/testing.html) The amount of effort needed just to match the functionality and quality of SQLite is staggering. That's hundreds of person-years.

Reply to Post

4 Comments