Can this be done for all runners? Both windows and macos seem to have Yarn 1.22.19

Not a fan of using *, from what I understand that's basically any. I think we could use generics instead (we could also replace the * that existed before if we are changing the code anyway). There's also the option of using empty in places where we aren't using the generic fields, which would be "safer" because we wouldn't be able to use them, but this doesn't feel very clear.

We could probably simplify error handling by passing sqliteQueryExecutor as an argument (that's never null) to the operation handling functions. And move all "Database not initialized" error handling to the main funciton

Can we also check if the FILE_PATH contains no file (with module.FS.stat or sth)?

Looks good to me

LGTM!

Is there a chance that older web clients will still try to fetch this file?

How does this interact with migrations?

Nice!

Probably the "better solution" is the refactor @kamil is doing right now, but it makes sense to me to sequence it later, so we have something working on web already

Makes sense!

This one is really interesting! BindingType and TypeID both already exist in the emscripten codebase, but they are generic definitions (+ some specializations for specific types). We can add our own specializations for vectors, as this diff is doing. And c++ will use the "most specific" definition, so in this case for BindingType<vector<int>>, our definition will be more specific, than the generic BindingType<T> provided by emscripten.

comm-query-creator.js filename doesn't match the class inside

Personally I don't like the FS$ namespacing and don't think it's needed (it's namespaced to the the module/file anyway)

Can you amend the test plan with checking if the cookie was migrated correctly?

Seems like we are using keyserverPrefixID not as a prefix, but just as the server id. It would be more descriptive if it was named something like ashoatKeyserverID. Could make a task/ implement it quickly in a separate diff?

Also remove convertToNewIDSchema from genesis.js

Followup task: https://linear.app/comm/issue/ENG-4402

Rebase

Inline getChatThreadItemSize

Seems like unsigned int is the official type, and uint is typedefs by some compilers automatically

Are we already explicitly using c++17 in native (not c++14 with extensions for 17)? If not it would be good to use it.

Looks good, just some questions:

• would it make sense to make these methods more inline with others? So e.g. setPersistStorageItem getting PersistItem as an argument, or getPersistStorageItem returning PersistItem or null instead of string?
• can we test these methods, and amend the test plan?

Refactored the code, used a similar approach as it's used on native (union with item type). Extracted search to a separate component.

We are changing the approach for this task, so I'm abandoning this diff. More info in the linear task.

We are changing the approach for this task, so I'm abandoning this diff. More info in the linear task.

We are changing the approach for this task, so I'm abandoning this diff. More info in the linear task.

Yeah, use of this any is contastrained to this one specific migration, and it's used as a function passed to other functions (that are typed correctly) so it shouldn't 'infect' other code

LGTM, but I'm not confident in db code, so adding for varun as blocking for this one

Replaced types with any as discussed in D8355

I've readded the types. MediaValidator needed to be special cased because of flow issues (that's the same issue as @bartek had). The types are replaced with any in D8386.

There's actually another problem: the generated functions can't be typed with our types because they will get out of sync. What I mean by that is that these functions operate on data that is stored in sqlite/redux at this point in time. If in the future someone modifies e.g. RawMessageInfo then this new type could conflict with the generated functions' body and make a flow error.

Use hasMinCodeVersion from the previous diff. Don't split NEXT_CODE_VERSION per platform as it's not needed anymore.