Did someone already thought about how color highlight can be better? It's been 4-5 years now that I'm coding on a virtual console that run on iPad with a monochrome code editor. Despite the fact that's remind me the old days when I was 8 years old, that doesn't stop me for coding with it.

I mean, is it really important to know that strings are red and numbers are yellow? How does that help me? They are both literal and behave to the user-content categories.

I was talking with my friend, and he says he likes to know if something is a keyword or an identifier. In C++, a lot of common keywords to define stuff and control the flow are often the first word and easy to spot.

A couple of months ago, I tried Flutter, and the editor can highlight ident blocks and give them different colors, but with Flutter, it's easy to get 10 or more ident levels, Does the color help? Splitting the code does.

I think, there is so much stuff that is more important than coloring the grammar of a language. For instance: knowing if an identifier belongs to which Rust Crate because, It's easy to stack 10 or more dependencies in one file that as better chances of names collisions.

Knowing if an identifier was recognized, if it used, if it's a local, a member, a global, a compiled value or a macro seems more important.

I would like to color block of code that is important or sensible. That will help my coworker about the severity of a particular place in the code.

What do you think?

A couple of years ago, we decide to migrate our customer's data from one data center to another, this is the story of how it goes well.

The product was a Facebook canvas and mobile game with 200M users, that represent approximately 500Gibi of data to move stored in MySQL and Redis. The source was stored in Dallas, and the target was New York.

Because downtime is responsible for preventing users to spend their money on our "free" game, we decide to avoid it as much as possible.

In our MySQL main table (manually sharded 100 tables) , we had a modification TIMESTAMP column. We decide to use it to check if a user needs to be copied on the new database. The rest of the data consist of a savegame stored as gzipped JSON in a LONGBLOB column.

A program in Go has been developed to continuously track if a user's data needs to be copied again everytime progress has been made on its savegame. The process goes like this: First the JSON was unzipped to detect bot users with no progress that we simply drop, then data was exported in a custom binary file with fast compressed data to reduce the size of the file. Next, the exported file was copied using rsync to the new servers, and a second Go program do the import on the new MySQL instances.

The 1st loop takes 1 week to copy; the 2nd takes 1 day; a couple of hours for the 3rd, and so on. At the end, copying the latest versions of all the savegame takes roughly a couple of minutes.

On the Redis side, some data were cache that we knew can be dropped without impacting the user's experience. Others were big bunch of data and we simply SCAN each Redis instances and produces the same kind of custom binary files. The process was fast enough to launch it once during migration. It takes 15 minutes because we were able to parallelise across the 22 instances.

It takes 6 months of meticulous preparation. The D day, the process goes smoothly, but we shutdowns our service for one long hour because of a typo on a domain name.