I've found sites like Udemy/Khanacademy/Codecademy/Brilliant/Edx to be very useful — possibly more useful than expensive education.

But they still need:

1. Better correction/update mechanisms. Human teachers make mistakes and material gets outdated, and while online teachers are rectified faster than classroom teachers, the procedure is still not optimal. Knowledge should be a bit more like a verified wiki.

2. Some have great interactive coding environments, some have great videos, some have awesome texts, some have helpful communities. None has it all. In the end, I don't want to learn a new language by writing code in my browser. It could all be integrated/synced to the point where IDEs have plugins which are synced to online videos, with tests and exercises built in, up to a social network where you could send snippets for review and add reviews to other people's code.

3. Accreditation. Some platforms offer this against payment, but I think those platforms often feel very old school (pun intended), with fixed schedules, marks and enrollments. Self paced is a must.

4. Depth is important. Current online courses are often a bit introductory. We need more advanced courses about algorithms, theoretical computer science, code design, relational algebra, category theory, etc. I get that it's about supply/demand, but we will eventually need to have those topics covered.

I do believe that for CS, full online education will eventually win from the classroom — it's still in its infancy, but has more potential to grow into correct, modern education.

Mother fucking SQL, fuck mathematicians, fuck every thing!

So let's supose we'd only need the first char of a string. Every, and I mean fucking every (php, java, javascript, ruby, python, haskell) fucking language, uses something like `substring(input, 0, 1)` as it knows the input is nothing more than a fucking array of chars, otherwise known as motherfucking String. Logically the offset for the first char is 0.

Enter SQL, there you need to put `SUBSTRING(input, 1, 1)` because fuck every one! Fucking math guys who developed relational algebra on which (most) databases are based on (I love you for it, but come on you fuckers!), Decided that the first character should be at position 1...

Fuckers

Dear Microsoft Kusto Query Language (KQL)

Screw you. You suck like more than a sudden depressurization event in an airplane. Creating your own freaking query language is bad, the people who invented SQL based it on a the principles of mathematical relational algebra, which although confusing, and not suited for all use cases is at least consistent.

You were invented by a bunch of oxygen deprived halfwits based on the principles of sadism and incompetence.

The only situation in which I would voluntarily use KQL as my tool of choice is if my purpose was to extract a Dantesque style revenge on someone who had committed grievous harm to myself and my family members. In that case forcing them to work with you day in and day out would still border on cruel and unusual punishment.

Sincerely, A developer who has spent the past 2 hours dealing with your Lovecraftian madness.

P.S. I hope you choke on a raw chicken bone and no one gives you CPR.

Ok now I'm gonna tell you about my "Databases 2" exam. This is gonna be long.

I'd like to know if DB designers actually have this workflow. I'm gonna "challenge" the reader, but I'm not playing smartass. The mistakes I point out here are MY mistakes.

So, in my uni there's this course, "Databases 2" ("Databases 1" is relational algebra and theoretical stuff), which consist in one exercise: design a SQL database.

We get the description of a system. Almost a two pages pdf. Of course it could be anything. Here I'm going to pretend the project is a YouTube clone (it's one of the practice exercises).

We start designing a ER diagram that describes the system. It must be fucking accurate: e.g. if we describe a "view" as a relationship between the entities User and Video, it MUST have at least another attribute, e.g. the datetime, even if the description doesn't say it. The official reason?

"The ER relationship describes a set of couples. You can not have two elements equal, thus if you don't put any attribute, it means that any user could watch a video only once. So you must put at least something else."

Do you get my point? In this phase we're not even talking about a "database", this is an analysis phase.

Then we describe the type dictionary. So far so good, we just have to specify the type of any attribute.

And now... Constraints.
Oh my god the constraints. We have to describe every fucking constraint of our system. In FIRST ORDER LOGIC. Every entity is a set, and Entity(e) means that an element e belongs to the set Entity. "A user must leave a feedback after he saw a video" becomes like

For all u,v,dv,df,f ( User(u) and Video(v) and View(u, v, dv) and feedback(u, v, f) ) ---> dv < df

provided that dv and df are the datetimes of the view and the feedback creation (it is clear in the exercise, here seems kinda cryptic)

Of course only some of the constraints are explicitly described. This one, for example, was not in the text. If you fail to mention any "hidden" constraint, you lose a lot of points. Same thing if you not describe it correctly.

Now it's time for use cases.
You start with the usual stickman diagram. So far so good.
Then you have to describe their main functions.

In first order logic. Yes.

So, if you got the point, you may think that the following is correct to get "the average amount of feedback values on a single video" (1 to 5, like the old YT).
(let's say that feedback is a relationship with attribute between User and Video

getAv(Video v): int
Let be F = { va | feedback(v, u, va) } for any User u
Let av = (sum forall f in F) / | F |
return av

But nope, there's an error here. Can you spot it (I didn't)?

F is a set. Sets do not have duplicates! So, the F set will lose some feedback values! I can not define that as a simple set!

It has to be a set of couples, like (v, u), where v is the value and u the user; this way we can have duplicate feedback values in our set.

This concludes the analysis phase. Now, the design.

Well we just refactor everything we have done until now. Is-a relations become relationships, many-to-many relationships get an "association entity" between them, nothing new.
We write down on paper every SQL statement to build any table, entity or not. We write down every possible primary key or foreign key. The constraint that are not natively satisfied by SQL and/or foreign keys become triggers, and so on.

This exam is considered the true nightmare at our department. I just love it.

Now my question is, do actually DB designers follow this workflow? Or is this just a bloody hard training in Pai Mei style?

I once saw a DBA using two for loops in PL/SQL to join two tables. I wonder if he knew about the "alternative" way to do that...

Urgh.. the amount of things you have to know as a developer.. it can get stressful and frustrating sometimes when (in-depth) technology knowledge is demanded from you (for instance, for a job position)..
It's like being a doctor, being a lifelong student.

A few examples of what I had to know during my career:
Java, .NET, Python, PHP, JavaScript/HTML5/CSS3, Sass/Less, Node.js, ReactJS, AngularJS, Vue.js, Cordova, Ionic, Android, design patterns, SOLID, databases (design, implementation, administration, both NoSQL and relational,..), deployment tools (Octopus, Jenkins,..), VCS, CI/CD, HTTP, networking, security (OAuth2, CORS, XSS, CSRF,..), algebra, algorithms, software testing, profiling, Linux, Unix, Windows, MS Office (advanced mail filtering,..), ITIL, IT Law (licensing and its implications when choosing a product, distribution right,..), server architecture,..

Sure yeah, I know, I've studied all that at university but.. it's been too long (almost a decade now). I have to revisit that knowledge.

What do u guys prefer
Relational algebra or calculus...