6 ms·

Ten-year-old severe SQLite bug fixed

36 points by turrini 10y ago

winteriscoming 10y ago
When I read the subject I thought the bug was reported 10 years back and got fixed now. But this looks like it was reported very recently and got fixed immediately. The bug report doesn't have any discussion except that it says the bug even affects versions which were released around 10 years back.
- baq 10y ago
  Which is typical for SQLite. And I mean this in the best possible way.
- dx034 10y ago
  It is also worth pointing out that in this case the result of a query wasn't returned, an error that would probably be noticed in most cases and bears no security implications (at least none I can see). That alone makes it far better than a 10y old security relevant bug. The fact that it was only reported now probably shows that it rarely occurs in real world applications.
  - colanderman 10y ago
    > The fact that it was only reported now probably shows that it rarely occurs in real world applications. And/or that those who encounter it assume they're misunderstanding left joins, and try something different.
pulse7 10y ago
This is why Larry Ellison is the richest Californian: nobody wants to experiment with important business data...
colanderman 10y ago
I encountered a similar bug to this (basic JOIN semantics error) about 4 years ago. Though it had been fixed in the head version by the time I found it, it had clearly been present for years. Turned me off to SQLite ever since -- it's painful not to be able to trust query results. Goes to show that even literal "millions and millions of test cases" don't catch bugs you didn't think of in the first place. This is a domain where a theorem prover could really shine -- checking transformation invariance in a non-Turing-complete language.
- PopsiclePete 10y ago
  I doubt "enterprise" databases like Oracle or MS SQL Server don't have similar bugs. Actually, what non-trivial software doesn't have bugs? I had the opposite reaction to you from reading this bug. They fixed it amazingly quickly, after it was reported, no fuss.
  - colanderman 10y ago
    I've used Postgres a lot more and for more complex queries than SQLite, and have never encountered nor heard of a query semantics bug with it (though I'm sure they exist). I know that is total anecdata, but my experience with Postgres is – whether thanks to more collective query hours, better testing methodology, or simply higher quality design – an experience that leads me to trust it more than SQLite.

36 points by turrini 10y ago

When I read the subject I thought the bug was reported 10 years back and got fixed now. But this looks like it was reported very recently and got fixed immediately. The bug report doesn't have any discussion except that it says the bug even affects versions which were released around 10 years back.

baq 10y ago
Which is typical for SQLite. And I mean this in the best possible way.
dx034 10y ago
It is also worth pointing out that in this case the result of a query wasn't returned, an error that would probably be noticed in most cases and bears no security implications (at least none I can see). That alone makes it far better than a 10y old security relevant bug. The fact that it was only reported now probably shows that it rarely occurs in real world applications.
- colanderman 10y ago
  > The fact that it was only reported now probably shows that it rarely occurs in real world applications. And/or that those who encounter it assume they're misunderstanding left joins, and try something different.

pulse7 10y ago

This is why Larry Ellison is the richest Californian: nobody wants to experiment with important business data...

colanderman 10y ago

I encountered a similar bug to this (basic JOIN semantics error) about 4 years ago. Though it had been fixed in the head version by the time I found it, it had clearly been present for years. Turned me off to SQLite ever since -- it's painful not to be able to trust query results. Goes to show that even literal "millions and millions of test cases" don't catch bugs you didn't think of in the first place. This is a domain where a theorem prover could really shine -- checking transformation invariance in a non-Turing-complete language.

PopsiclePete 10y ago
I doubt "enterprise" databases like Oracle or MS SQL Server don't have similar bugs. Actually, what non-trivial software doesn't have bugs? I had the opposite reaction to you from reading this bug. They fixed it amazingly quickly, after it was reported, no fuss.
- colanderman 10y ago
  I've used Postgres a lot more and for more complex queries than SQLite, and have never encountered nor heard of a query semantics bug with it (though I'm sure they exist). I know that is total anecdata, but my experience with Postgres is – whether thanks to more collective query hours, better testing methodology, or simply higher quality design – an experience that leads me to trust it more than SQLite.