[KronisLV]

In my eyes, the following should be equal:

  - a binary file that's used for storing data, like an SQLite database
  - a text file that's used for storing data, like an XML or JSON file

Someone else linked a list of CVEs and how those could be exploited, the look of which isn't that bad: https://news.ycombinator.com/item?id=28690837

Those problems could be addressed in a pretty easy way, plus if you're security conscious, just run the importer in an entirely separate container which will basically be single use (possibly distroless, if you want to go that far), with resource limits in place.

But that's not my point. My point is that both of the data formats should be pretty much equal and them not being so in practice is just a shortcoming of the software that's used - for example, even spreadsheets ask you before executing any macros inside of them. There definitely should be a default mode of addressing such files for just reading data, without handing over any control of the computer to them.

> Zipping a SQLite database is not a good serialization method...

Therefore, with this i disagree. SQLite might be flawed, but zipping an entire dataset and sending it over the network, to be parsed and merged into a larger one is an effective and simple solution. Especially, given that your app can use SQLite directly, but probably won't be as easy to make while storing the state as a large JSON file, which will incur the penalty of having to do conversion somewhere along the way. Here's why i think it's a good approach: https://sqlite.org/appfileformat.html

Who's to say that JSON/XML/... parsers also wouldn't have CVEs, as well as the application server, or back end stack, or web server that would also be necessary? In summary, i think that software should just be locked down more to accomodate simple workflows.

[Dylan16807]

> My point is that both of the data formats should be pretty much equal and them not being so in practice is just a shortcoming of the software that's used - for example, even spreadsheets ask you before executing any macros inside of them.

Okay, but you need to defend against reality, not against what could in theory be possible.

Sandboxing is a pretty good solution, at least.

> Who's to say that JSON/XML/... parsers also wouldn't have CVEs, as well as the application server, or back end stack, or web server that would also be necessary?

Raw SQLite files are a huge attack surface that isn't directly designed to be secure. JSON is an extremely simple format that can be parsed securely by just about anything (though occasionally different parsers will disagree on the output).

[jameshart]

(Edit: Strange - this reply was intended to be attached to the grandparent of this post. Not sure why it ended up here.)

XML, a data format explicitly designed for interchange where parsing untrusted input was a design goal of the language.. contains ‘external entities’, which permits the person crafting an XML doc to induce a vulnerable reader of the document to pull in arbitrary additional resources and treat the data from them as if they came from the document creator.

There are all sorts of confused deputy attacks you can perform via this kind of mechanism.

If XML can have that kind of issue, when it ostensibly contains no arbitrary execution instruction mechanism at all, how can you expect a sqllite database file, which can contain VIEW and TRIGGER definitions, to be safe?

[smackeyacky]

It's been repeated a few times here that SQLite is a big attack surface - might be worth taking that discussion to a new submission rather than continuing to hijack this one:

https://news.ycombinator.com/item?id=28691759

[kortilla]

Your argument can be extended to claim just shipping executable binaries that output the data you want when you execute them should be equivalent as well.

It’s unsafe because the attack surface is so large and the use case of an untrusted attacker isn’t something strongly considered.

> In my eyes, the following should be equal:

In an ideal world maybe, but this hasn’t been true for the last 50 years.