Re: [w3c/IndexedDB] Disallow starting readwrite transactions while readonly transactions are running? (#253)

I'm conflicted between the two. I'll explain my arguments for both sides, to help the decision.

**No ordering**

As a user, the tutorials and MDN guides I have access to don't describe the ordering guarantees. For example, purely based on the guides, I didn't know that I can rely on requests completing in order. For example, when I had to fire 3 put requests, I wrote code waiting for all 3 to complete before proceeding. I didn't realize that I can just wait for the last request. I claim that most folks writing IndexedDB code operate in a similar mode.

As an implementer, the request ordering costs us extra complexity and will cost us extra memory. AFAIK the Cache Storage API has ran into a similar problem, and they decided to sacrifice ordering for performance. I definitely wish we didn't have ordering constraints -- user code that needs ordering can add extra coordination, and code that doesn't could squeeze out a bit more performance.

So, I think we shouldn't add extra ordering and, instead, we should call out that the ordering is implementation-dependent, but must be consistent with some ordering where readonly and readwrite don't overlap. To be clear, I don't know of a concrete setup where this would be a net-benefit... I'm just thinking about not over-constraining the future.

**Ordering**

Having strict ordering requirements and associated tests means we'll have consistency across IndexedDB implementations, increasing the odds that content will work well across browsers.

On the testing front -- after Chrome removes its snapshotting system, we won't have any implementation that does snapshots. So, we won't have any guarantee that WPT tests won't accidentally start depending on ordering anyways. 

The Cache Storage API parallel isn't perfect. In their case, prioritizing performance is easily the right call, because Cache Storage reads generally block page loads. IndexedDB is used more often outside of critical paths.



-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/w3c/IndexedDB/issues/253#issuecomment-455674397