lucaspickering / gdlk Goto Github PK

Instead of relying on GQL's untyped errors, we should use union types to make results, similar to what Rust does. More info

We use a bunch of different number types and some of the conversions are suscepitable to overflow/underflow. Some things to test include:

• Program size overflow (can't fit all instructions in a Vec)

We should prevent people from running an infinite loop that crashes the server. The easy fix is once we add CPU cycle counting, we can just add a hard cap on how many cycles you can run before it kills the program.

We want a composite score that can summarize all stats of a user_program. We should push this as the primary metric. It can be a generated column on the user_program_records table.

We should have a button in the IDE that executes the entire program without updating the UI step-by-step, so that it can finish instantly. Then it'll just update the UI at the end.

Actix already has a ResponseError type. It might cause less confusion if we rename ours to ApiError. Make sure to update any comments that reference it too.

When you first load the IDE, if there's source code in the editor, it doesn't properly compile it and display results. You have to make an edit first before it will do so.

It also might be that it is compiling, but that occurs after the render and it doesn't properly trigger a re-render.

1. Create first.gdlk
2. Open it in the editor
3. Open a second tab to the program spec page (where the solution table is)
4. In the first tab, make some changes to first.gdlk and save them
5. In the second tab, make a copy of first.gdlk
6. Open the copy, and it will be empty instead of having the contents of first.gdlk

This is because the second tab loaded the source code of first.gdlk before it was saved, and it never got refreshed. We could fix this by adding an explicit copy mutation to the API.

Our OpenID auth workflow doesn't use CSRF tokens

All the slug columns that we have are just generated from the name column at insert. We should use pg generated columns to clean that up. We'll need to make sure there's still an index on the new col, so that filtering can be turbo fast.

just make this in sql

Right now we init OIDC providers in series it'd be nice to parallize this before/shortly after we add more providers to prevent startup time from getting too long.

Once this juniper issue is fixed, we can go back to a stable version of juniper-from-schema.

We have some enums that we map to DB fields, which have string literals associated with each type. We could clean this up/automate it with the strum crate. I want to wait until their next release though, because that updates it to Rust 2018 which will make it a bit nicer to user.

We could use tracing for logging stuff in the app. This is going to be dependent on juniper support: graphql-rust/juniper#713

We should use a nonce in the OpenID workflow to prevent replay attacks.

• GitHub
• Steam

We want to restore the prod DB to development so we can get new programs etc. in development easily. We'll need to sanitize the DB though so that we don't leak any data. For now, we can just delete the user table and everything that references those rows. Maybe in the future we'll come up with a better solution so we can seed some user programs or something.

If you put a question mark ? in a user_program name, it causes a 404 on the SPA. Need to check other special chars too

The CI should run all down migrations to make sure they all work. This will be easy once this issue is closed: diesel-rs/diesel#2095

The functionality is implemented but they haven't done a diesel_cli release yet, so once that's done we can use it.