To be honest with you, I’ve never had a bad experience with PHP.

Yes, it’s “dirty” compared to something like Haskell, but it’s not a bad thing. Dirty things usually bring simplicity and allow implementing the intended case super quickly, at the cost of breaking apart at scale. There are no bad tools, there are wrong tools for the job.

Premature optimization is the root of all evil. The more I launch new projects for me/other companies, the more I come to the realization that the vast majority of the projects out there will never see scale. They will be proven non-viable/impractical and deemed obsolete way before they outgrow the $20 VPS they were hosted on.

Sometimes (all the time, really) launching quickly like there is no tomorrow is the most viable business strategy. If (yes, “if”, not “when”) your project outgrows PHP and gets to the point when PHPs abstraction model is the bottleneck, you’ll have the money to rewrite the project in any language out there, trust me.

As someone said on biking subreddit to a person that asked how to buy the newest super-aero helmet, “if the aerodynamics of the old helmet is what holds you back, someone will be sending you the new one for free”.

Sometimes people want to be too smart. If you want to consume a handful different restful API, it might make sense to abstract away some common functionality in your client implementation — yet to assume they follow the same convention in how their URI is built is borderline insane.

All I wanted to do was to change one API to a newer version, and now the implementation breaks for at least two other because it was done in an Abstract class and now I have to untangle that mess.

In some cases code duplication wouldn't be that bad. Even if an otherwise unrelated API seemingly share the same contract, still assume it has its own contract. You never know how those API evolve and I proclaim they will evolve towards breaking your assumptions.

I go to add a method call in a business logic class that's used exclusively in a particular service, and get blocked in the PR by some other guy-

Other Guy: refactor this into the shared framework referenced by all our microservices

Me: it is only and would only be used in this service

OG: what about the other business logic class in this service?

Me: it's not used there, and if it does end up used there then we can refactor it into a class that they both reference then

OG: I need to know when the abstraction of this function will be done. is it going to be delivered next sprint?

Me: YAGNI - better to avoid doing extra work when we don't know if we'll even need it

OG: tbh you can still abstract it with some generics and lambda magic, but im not gonna enforce that

Me: premature abstraction is the root of all evil (tongueout)

OG: not really, its the root of not having a million miles of tech debt in 2 years

I just can't win for losing with the anti-YAGNI yogi.

I'm in a big fat fucking stinking rut, as in progress on this project has absolutely stagnanted.

Gonna rubber face your duck now **UNZIPS** excepts I don't have zippers, as joggers are the one true way; fake Adidas til I fucking drop.

Brain damage aside, I understand both how I've layed out the data and what I'm supposed to do with it. We have a virtual machine, an array of instructions and arguments for a given process within it, and we need to walk this array and map values to registers.

We also need to spill values inside registers to stack, IF they are required at a further point within that block. This also isn't terribly complex. We simply look forward in the array and see if the value is an argument to any instruction that *needs* this value to be loaded (ie, within a register).

So this implies multiple iterations; we need to better understand how one particular value is used throughout an F before we can make a final decision on how many registers and stack space are actually needed for the whole block.

Here's where it gets tricky. If there's a call, we need to be certain that the symbol being invoked has already been fully processed. Besides the obvious fact that recursion fucks me up, there's another matter: say a private method gets invoked by another private method. We can take advantage of this, by which I mean, sacrilege incoming so put on this toga.

Looking at the output for C compilers, it would seem this is not done in practice, I would assume because it's a pain in the ass. But when you have the guarantee that F will only be called internally, as that's what "private" means, there's two ways it can go:

0. It's well below the 13-20 cycle threshold, so you inline the fucker. No suprises there.
1. It's a more involved affaire, and invoked in more than one place, so you don't inline it. Codesize matters.

Recursion and [1] are the big deal things holding me back. Not because it's too hard, like I said this is kindergarten level abstraction. I'm just slow and fanatical, which is how I prefer to spell "constant obsessive paranoid delusions". I can see the potential optimization I can pull here, so I'm stuck trying to figure it out.

Idea would be, handling the register allocation and stack spill for an internal-internal (or deep internal; what we like to call a "guts" method) in synchronization with the *calling* processes. This is, fundamentally, violating all conventions -- but so under the hood no one will notice.

Let me give you an example. If we were to pass some value to a function, expecting to mutate it and get a different value back, in a lot of cases it'd be stupid to make an implicit copy by using two registers, one for input and another for the output. Dude, it's one cycle. Multiply it by a million, say sixty times per second, for every time you __needlessly__ make a copy of a value that we've already stated is mutable.

Clearly unacceptable. This is, in the strictest sense, everywhere in every single codebase. Premature micro optimization is the root of all goodness, God is great and praiseworthy. So how do we go about it?

Answer is I know and I don't know. By which I mean to say, this very thing I've done by hand. Assembly is fun. Now the issue is teaching a calculator how to do it. Not so fun.

There is a dependency chain between processes, as I believe I've kind of alluded to. I'm trying to make decisions on the side of the caller depending on the details of the callee, which is why recursion is rawdogging my soul. This is the same situation, it's inverting the direction of one or more links in the dependency chain, which makes no fucking sense.

And yet it does.
Brain, explain yourself.

How do *you* handle this without crashing?
Brain?

<<ME STEWPED; BEEP-BOOP>>

Alright then, that was a useless attempt at fuckery. Let's have a nap then, maybe it'll come to me in the morning. That's what I've been saying to myself for almost a month now.

Perhaps it is a hardcoded fuk.