This error usually happens because `@stream-io/openai-realtime-api` is a server-only package and Next.js tries to install or use it in the client-side environment. It requires Node.js 18+ and won’t work inside client components.
Try using the package only in a server route (like `/app/api/...`) and make sure your Node version is 18 or higher. Installing it with a clean environment (`rm -rf node_modules && npm install`) also helps.
Thank you for your suggestion on useReducer, I read something about it and I think it could make the code a little less cumbersome. I know there are some libraries like redux that simplifies the state management a lot, and, as you said, they will be a lot more efficient than a custom approach. If I'll be doing something for production I'll go that way for sure, but as in this case I'm exploring the framework I wanted to use as much as "pure react" as I can to understand how it works.
You can reverse a string in Python in multiple ways:
1. Using slicing (shortest and fastest)
s = "hello"
print(s[::-1])
2. Using reversed() with join
s = "hello"
print("".join(reversed(s)))
3. Using a loop
s = "hello"
res = ""
for ch in s:
res = ch + res
print(res)
All of these output: "olleh"
Which is more suitable and comfortable, overall secure. Even Github for mac is safe, but however it's kinda glitchy. So therfore it's makes errors sometimes, but anyway you could use Push button finding near the Commit section.
This warning is completely normal and expected after changing settings. It will go away automatically for the next deployment.
It simply means that the version of your app currently live (Production) was built using the previous settings (what is running), but your Vercel dashboard now has newer settings.
I think a more modern approach is to use scripts in programming languages, such as Python, so that it won't be limited to just one operating system. As far as I know, 7-zip can run on both Windows and MacOS/Linux. The steps to handle this issue are as follows:
Use the list of decompression file extensions supported by 7-zip to filter out all possible decompressible archive files in a certain directory (for example, tar, 7z, rar, etc.). Remember to use os.path.extsplit(path)[-1] to get the extensions of these files for filtering, because the following files may just coincidentally be named like "footar" and they are not archives; only those like "foo.tar" are.
Use the 7z t command to test these compressed archive files and filter out those that fail the test (because some files, although ending with these supported compressed archive suffixes, may be corrupted archives).
Use 7z x or 7z e to extract these valid compressed archive files. The specific command to use depends on the file directory format you expect to obtain.
Recursively execute the above steps, and end the recursion when no available compressed packages are found in the first step, so as to decompress the nested compressed package files within the compressed packages
This is just an implementation logic. I'm sorry, but I won't attach specific code because everyone's expected results are different, and there may be some additional customization requirements.
As a test - why did you ask this question here, and not consult an AI?
Because we're a better resource that can comprehend the question, and the bits that weren't asked. The collective-We can provide relevant context whereas an AI is just answering with "whichever word comes next".
Would an AI have flipped it over with a counter-example like this ? No.
I mean an ouput of a code cell in Jupyter Notebook file. PyCharm has got this, but mostly I work in Vs code. I selected a portion of a text and I want it to be copied by rmc copy/paste.
With recent changes to the AI Foundry and New Foundry features, it appears Microsoft has added Tool Approval support through the Playground. This issue should now be resolved.
What I did was to ensure that all imported or used libraries appear inside dependencies in the package.json file. I made a mistake of installing them on the project path instead of cd functions, then install them which will make allow them to appear under dependencies on the .
The demand for high -performance GPUs for commercial AI applications significantly outpaces the current supply, leading to scarcity despite the presence of large chip manufactures
Use break to exit the loop as soon as the element is found
numbers = (2, 4, 6, 8, 10, 12)
x = int(input("Enter a number to search: "))
i = 0
while i < len(numbers):
if numbers[i] == x:
print("Number found at index", i)
break # stop the loop
i += 1
In large React/Next.js apps, Redux Toolkit starts to feel like an anti-pattern when most of your state is really server data. Libraries like React Query or SWR handle fetching, caching and syncing much better, so RTK just adds code you don’t need.
RTK still works great when you’ve got complex client-side logic — shared UI state, multi-step flows, or anything that has nothing to do with the backend.
Simple rule of thumb:
– server data → React Query/SWR
– complex UI logic → RTK
By the way, when I was working on a similar architecture, I got some help from a team at Red Rocket Software — here’s their main page: https://redrocket.software/. They also build custom React/Next.js solutions, so their input was actually pretty useful.
I think I'm spending more time in this that I need. Most of the posts/articles I read said to embrance eventual consistency.
When creating a new tenant I NEED consistency. I believe the "best" way to get the job done is to remove the shared transactions for now(since there're causing me these problems) and just use my usecases DELETE methods in case something goes south.
I'll get (PHP) Laravel-framework specific in my response and respond from that specific architectural view point.
"Controller" is what stands between the incoming request to your application and the business-logic operations it performs, while "Service" is what wraps around that same business-logic opeations and is called my the controller.
The following flow might be helpful in wrapping one's head around the concept:
client sends a request -> controller (validation, calling the necessary service to perform business logic and passing the request to it) -> service (computation, db checks, resource mutation, etc) -> response back to controller (formatting, view) -> client receives response.
When you declare api project(':shared'), Gradle should expose it to consumers, but the default Maven Publish configuration doesn't automatically include these dependencies in the POM. Here's the fix:
Add this to your :library's build.gradle.kts inside the MavenPublication block:
Make sure :shared is also published to the same repository (JitPack). Your project structure requires publishing all modules independently – JitPack will handle building the full dependency tree if all modules are in the same Git repo.
Sorry, forgot to add some details in the question:
Yes, I do have repositories interacting with my DB
I'm instanciating my Handlers, UseCases, Repositories and Orchestartors inside a Depency Injection Container
Ex of repository:
type CoreUserRepository struct {
db *sqlx.DB
}
func NewCoreUserRepository(db *sqlx.DB) *CoreUserRepository {
return &CoreUserRepository{db: db}
}
func (repo *CoreUserRepository) Exists(idTenant int64, idUser int64) *int64 {
query := "SELECT id FROM core.users WHERE id_tenant=$1 AND id=$2;"
var id int64
err := repo.db.QueryRowx(query, idTenant, idUser).Scan(&id)
if err != nil {
log.Println("[CORE USERS - REPO] Could not find user: ", err)
return nil
}
return &id
}
I'm also thinking in creating Request Containers Dependency Injection for my Unit of Work. Each request creates a new container that's injects the necessary repos, usecases AND the request context. I create the UW, pass it to the context and inject the context in the usecases and repositories. With this, I'd not have the problem of having a "global transaction" in my whole application when declaring a Unit Of Work. You think that's too much? Sorry, first time with an application of this size/complexity ;-;
Good day. I got the same task today as part of Angela Yu's Python Bootcamp, and below is my solution.
Am a newbie therefore i wanted to avoid diving into complicated concepts which I havent learnt and wanted to make it as simple as possible.
import numpy as np
from random import choice
print("Let's play Tic Tac Toe!")
board = np.array([[None, None, None], [None, None, None], [None, None, None]])
def render_uiboard(board):
symbols = {None: " ", "X": "❌", "O": "⭕"}
rows = []
for i in range(3):
row = f"{symbols[board[i,0]]} | {symbols[board[i,1]]} | {symbols[board[i,2]]}"
rows.append(row)
separator = "\n-------------\n"
return separator.join(rows)
print(render_uiboard(board), "\n")
print("Am playing ❌, I move first!")
def check_win(board, player):
for row, col in zip(board, np.transpose(board)):
if np.all(row == player) or np.all(col == player):
print(f"Player {player} won!")
return True
main_diag = np.diagonal(board)
antidiagonal = np.diagonal(np.fliplr(board))
if np.all(main_diag == player) or np.all(antidiagonal == player):
print(f"Player {player} won!")
return True
def check_draw(board):
return np.all(board != None) # if no more empty cells
def get_user_coord(board):
valid_input = False
while not valid_input:
input_coords = input("Your turn! Enter coords (e.g. 1 0): ")
if input_coords in (
"0 0",
"0 1",
"0 2",
"1 0",
"1 1",
"1 2",
"2 0",
"2 1",
"2 2",
):
r, c = map(int, input_coords.split())
if board[r, c] == None:
valid_input = True
return r, c
else:
print("That cell is already occupied. Choose another.")
else:
print("Please enter valid coordinates for an empty cell.")
game_over = False
while not game_over:
empty_coords = [(r, c) for r in range(3) for c in range(3) if board[r, c] is None]
r, c = choice(empty_coords)
board[r, c] = "X"
print(render_uiboard(board))
print("\n")
if check_win(board, "X"):
game_over = True
break
if check_draw(board):
print("It's a draw!")
break
r, c = get_user_coord(board)
board[r, c] = "O"
print(render_uiboard(board))
print("\n")
if check_win(board, "O"):
game_over = True
break
if check_draw(board):
print("It's a draw!")
break
continue
In my opinion, you’ll need to analyze your current system’s architecture, business rules, and integration requirements before choosing a specific communication pattern. Without understanding what responsibilities the legacy system must delegate to the new identity service, it's impossible to recommend REST, gRPC, or message queues.
Each option depends on concrete details:
1. What exactly needs to be moved out of the monolith?
2. Which operations are synchronous (e.g., login, token issuance) vs. asynchronous (e.g., KYC updates)?
The right solution isn't universal it must follow your business workflows and technical constraints.
import tkinter as tk
from tkinter import simpledialog, messagebox
import random
CORRECT_PATTERN = [1, 5, 9, 8]
CORRECT_PASSWORD = "universe123"
CORRECT_PIN = "4567"
jokes = [
"Why do programmers prefer dark mode? Because light attracts bugs!",
"Why was the computer cold? It forgot to close its Windows!",
"Why did the coder quit his job? Because he didn’t get arrays.",
"How many programmers does it take to change a light bulb? None, it's a hardware problem.",
"Why do Java developers wear glasses? Because they can't C#.",
"What do you call 8 hobbits? A hobbyte.",
"I told my computer I needed a break, and it said: No problem, I'll go to sleep."
]
pattern = []
def show_joke():
messagebox.showinfo("Joke", random.choice(jokes))
def ask_pin():
pin = simpledialog.askstring("PIN", "Enter PIN:", show="*")
if pin == CORRECT_PIN:
messagebox.showinfo("Success", "Unlocked")
show_joke()
window.destroy()
else:
messagebox.showerror("Error", "Wrong PIN")
def ask_password():
password = simpledialog.askstring("Password", "Enter Password:", show="*")
if password == CORRECT_PASSWORD:
ask_pin()
else:
messagebox.showerror("Error", "Wrong Password")
def press(num):
pattern.append(num)
if len(pattern) == len(CORRECT_PATTERN):
if pattern == CORRECT_PATTERN:
messagebox.showinfo("Pattern", "Correct Pattern")
pattern.clear()
ask_password()
else:
messagebox.showerror("Pattern", "Incorrect Pattern")
pattern.clear()
window = tk.Tk()
window.title("Multi-Lock System")
window.geometry("300x380")
title = tk.Label(window, text="Draw Pattern", font=("Arial", 18))
title.pack(pady=10)
frame = tk.Frame(window)
frame.pack()
num = 1
for r in range(3):
for c in range(3):
btn = tk.Button(frame, text=str(num), width=10, height=4,
command=lambda n=num: press(n))
btn.grid(row=r, column=c)
num += 1
window.mainloop()
It looks like the crash may be happening because the OCR result sometimes returns empty or unexpected values in release builds, and the overlay tries to read bounding boxes that aren’t actually there. This usually doesn’t show up in development because the error handling is more forgiving.
Just to understand better: have you checked whether the frame processor ever returns null or an empty array for the text blocks in your release build logs?
Could you share the part of your implementation where you process the text blocks and render the bounding boxes?
@ReX357 You need to edit your question: no, that's not a grid of parallelograms. It's a grid of trapezoids. It's pretty straightforward to do in Python + Numpy, but correcting your terminology will help toward accurately describing the problem.
My knowledge might be slightly outdated since I had this problems some 2 years ago. If I remember correctly, this was not possible using the Google Login Plugin because it is based on OAuth.
What you could do alternatively, is set up an authentication with Google Workspace using SAML. This way, you can configure a group attribute.
If you are using the same date even as the week changes, it would be useful to have a macro that stores the date calculated with +15 in the cell. There is no need to change +15 to +8.
https://stackoverflow.com/users/3899431/luis-colorado Two things. Dealing with your first comment, this is only a simple little test that I did to test char** allocation, a block box if you will that does not return anything or really do anything important. It's just a test for me to learn best practice and weed out all of the junk answers on the internet.
Second, I did take a look at that book before posting (It's on the Internet Archive). Maybe I didn't study it enough (I'll have to go back and revisit) but the section starting on page 99 didn't really deal with these specific questions I'm asking.
https://stackoverflow.com/users/2410359/chux Yep, exactly right. I forgot to put parens around the (size + 1). Thanks for pointing that out. Strange that the Valgrind profiler didn't catch that.
if you are using Anaconda for environment management, Then create a new environment in anaconda and select python version whic are supported by tensorflow and this will be solve your issue.
Need more help please watch this solution "https://www.youtube.com/watch?v=L4Y7A44lzpM"
https://stackoverflow.com/users/4756299/andrew-henle Very strange about the formatting. I copied it straight from Eclipse, apparently it didn't translate too well into SO. I'll watch out for that in the future, thanks for the heads-up.
You cannot assign a custom name/ID to vc_custom_####### classes.These classes are system-generated every time you save the element. They will always change when you edit the “Design Options” tab.
Instead, you must use custom CSS class (or Element ID) fields that WPBakery provides.
Certainly! So the problem I have doesnt have so much to do with the current grid stacking implementation, so i've updated the snippet using absolute positioning instead.
The problem is; I have a container inside yellow column that holds two elements. When I dont break out the "text-container element" it's easy to control the spacings between the two (using row-gap or margin-bottom).
But If I add the break-out class on the text-container the element is removed from the document flow and isnt considered as a child of the nested yellow-columns flexbox-container, and that makes it so I can't control the spacings anymore, that's what Im having problem solving in a elegant and scalable way. Hope I'm making sense, having a hard time explaining the issue in few words.
native javascript
A lot of this subject is already said. I am a Hardcore Native Javascript developer. So a natural fit for me is using Promises, together with async/await they are a natural fit.
angular
Angular uses RxJS. A technology developed by a Microsoft developer and implemented in lots of tech like Angular. So for Angular thats the natural fit
the future
I cant look in the future but I dont go with the flow of SPAs. Developing with SPAs takes more time and Javascript maintenance. Because of that HTML first, Javascript second is now hot, so back to how it was ment to be. HTMX can fit that bill very well. Thats why I wont use anything that goes of the ES* standard, keep my eyes on the ball.
my guess
Good writing of 4* Javascript code is already hard. So I stick to ES*. Native Javscript is on the rise. Even Netflix is rewriting React pages to native Javascript.
It is tricky with all those variations! While you’re figuring out a robust way to handle it, I’ve been trying out Golo777 great for taking a break and having some fun in between these tedious file edits.
To integrate Meta Ads with Google AdMob bidding on iOS, you first need to install the Meta Audience Network SDK along with the AdMob SDK using CocoaPods, then enable bidding for Meta Audience Network inside your AdMob account. After that, you must add Meta as a bidding partner in AdMob mediation settings and include the correct placement IDs and app IDs from Meta Business Manager. In your Xcode project, configure the adapters, initialize both SDKs in the AppDelegate, and make sure SKAdNetwork IDs from Meta are added to your Info.plist. Once everything is set up, load and display your ad units as you normally do with AdMob, and the system will automatically handle bidding between Meta and other networks.
You'll have the opportunity to read various games and blogs on my site <a href="https://livesweepstakes.uk/emma-lous-christmas-cash-giveaway-2025-usa/"> Emma Lous Christmas Cash Giveaway 2025 USA</a> and a chance to win gifts. My site serves as a platform for entertainment and providing opportunities to win gifts. Please read the rules carefully.
Tested for @RodyOldenhuis cases (transpose, tricky strings) ... not heavily tested otherwise but seems to work fine so far:
PS: I'm replacing with whitespaces because in my case I'm trying to automate modifying mfiles by adding call to some extrafunction in all main-function / constructor (even if weirdly formatted) of many mfiles ... if you have some idea on that ... mtree/getcallinfo are of help but not to automatically insert code in mfiles.
This is a problem associated with the MIME type/content-type of the files, here's what you can do to fix this:
1. If you are uploading files to the bucket using AWS CLI, use official and latest version, let the utility guess the mime types of the files. Invalidate all files in cloudfront after this.
2. Clear you browsers cache & cookies. Issue should be fixed!!
To resolve the Manual Mode issue in ReactPress, ensure your WordPress hosting on server has Node.js and npm installed. You might need to contact their support or use a custom server setup. For detailed guidance, check out the ReactPress project at https://github.com/fecommunity/reactpress offers solutions and documentation for seamless React-WordPress integration.
You're printing the result of the recursive call, but you're not returning it. So when digit_sum(452) finishes, it returns None because there's no return statement in the recursive branch.
SELECT t1.name, t1.surname1, t1.id, t2.name1, t2.origin
FROM table1 t1
LEFT JOIN table2 t2 ON t1.name LIKE '%' || t2.name1 || '%'
OR t2.name1 LIKE '%' || t1.name || '%';
1. Verify Java is installed
java -version
2. Download the IJava zip from GitHub
https://github.com/SpencerPark/IJava/releases/download/v1.3.0/ijava-1.3.0.zip
Extract to a folder, e.g. D:\IJava.
3. Install the kernel using the install script
Open terminal inside the extracted folder and run:
cd D:\IJava
python install.py
This installs the Java kernel into Jupyter (creates a kernel folder under C:\ProgramData\jupyter\kernels\java by default).
4. VS Code blocked the kernel as “untrusted”
Error shown:
The kernel 'Java' was not started as it is located in an insecure location 'C:\ProgramData\jupyter\kernels\java\kernel.json'.
Two ways to fix that:
1. Trust that specific kernel folder in VS Code settings (jupyter.kernels.trusted) — add the ProgramData path. (this did not work for me)
2. Move the kernel folder into your user Roaming path (what you did) so VS Code trusts it automatically.
So I moved the folder:
From:
C:\ProgramData\jupyter\kernels\java
To:
C:\Users\Admin\AppData\Roaming\jupyter\kernels\java
5. Fix the kernel.json path (critical)
After moving, the kernel still pointed to the old jar path, so Jupyter tried to run a jar that no longer existed.
Open:
C:\Users\Admin\AppData\Roaming\jupyter\kernels\java\kernel.json
Edit the "argv" entry so the -jar path points to the new location. Example final content we used:
{
"argv": [
"java",
"-jar",
"C:\\Users\\Admin\\AppData\\Roaming\\jupyter\\kernels\\java\\ijava-1.3.0.jar",
"{connection_file}"
],
"display_name": "Java",
"language": "java"
}
Make sure the jar filename and full path exactly match what’s in that folder.
6. Restart VS Code completely
(Not just reload — close and re-open). This makes VS Code pick up the updated kernel.
8. Select the Java kernel in the notebook
The one you are looking for is AsyncLocalStorage. Introduced in Node.js 12+ (stable in later versions), part of the async_hooks module.
You can read more from here
Even cls-hooked mentioned above internally uses this.
@ReX357 Thank you for the information, and that's perfectly okay about the linear algebra - it's quick to pick up. For my approach to work, you will need to know basic row reduction, and what a vector equation of a line is.
Essentially, to find the intersection of two line segments L1:{(a,b),(c,d)}, and L2:{(e,f),(g,h)}, you first turn them into their corresponding vectors: V1:{(a,b),(c-a,d-a)}, V2:{(e,f),(g-e,h-f)}. Then you set V1=V2, and use Gauss-Jordan elimination (row reduction) to find the coordinates of their intersection (IN THE AFFINE BASIS!). By affine basis, I mean that the second component is added to the first to obtain the new point - it's a bit technical.
If all these coordinates are between zero and one, then you can apply one of them to the equation of one line, and (bam!) you've got your intersection.
Repeat this process for every pair for a working solution. Of course, this is probably speed-up-able too.
Would you like a basic version in Python or Lua? (I'm rusty with Python, so Lua is easier, but I can swing both ways.)
@Jasper: Yes. I have the coordinates for all the points in green in my sample diagram. The narrow strips are roads. I need to get the corner coordinates for the 36 square'ish sections. In terms of linear algebra, I just know what it is but no real experience. I messed around with OpenGL graphics and programmed some simple 3d objects rotating but that's the extent of it. And that was 10 years ago. But I have a propensity for learning quick.
@Richard - At the moment no, they only probably need a human readable name for the moment. If I need to add more information later like a sales prices or such I was thinking of creating a generic SalesItem where the specific is composited into that new item. - Or what ever the common practice equivalent for this would be in F#.
@Brian - thank you - I will try and give that approach a go today if I get time. It looks promising, visually. I'll be interested to see how my "operations" functions work with it. It looks close to what I want to achieve in my mind.
@Brent - The plan is to only allow new materials and behaviours for them through new hardcoded types. I understand with F# pattern matching I could probably have a single type with a differentiator say "Ore", "Oil" "Gas" and then have a Name "Iron Ore", "Coal Ore", "Crude Oil", "Natural Gas" but then I would assume I will have some massively long "match" expressions that will have to grow with each new type, anyway? Or am I looking at this wrong.
Sorry for answering to all in a single reply, but normally in Stack Overflow I would comment on each reply, but for some reason SO is not giving me that option today!
If you have the start and end coordinates of each line segment comprising the grid, I might be able to help you. Please inform. Also, is your goal just to detect these intersections, or is it to partition the grid by them?
I agree with Richard's approach. You can build a type hierarchy using DUs in F# or inheritance in C# to model the diagram, but what's the point? How will this benefit the logic you want to implement?
If you look at the tree of life, that's an example of something you might be tempted to model as an OO hierarchy. Or with a DU in F#. Then you'd have insects with their six legs in one branch.
It depends on the purpose, but the chance of success in modeling it like that is very slim. You'd be hardcoding the structure when it's likely that this structure should be built at runtime, and legs and wings and eyes should be attached dynamically at runtime.
Let's say somebody discovers something new in the tree of life. A new species, or some new fact about an existing species. If an application for maintaining information about the tree of life were to cope with this situation, and it was hardcoded like that, it would be impossible to modify information without releasing a new version of the application.
Ok, reality isn't always that simple, so maybe what you're trying has a purpose. Still, do you need a hardcoded tree structure? Maybe a list based on the diagram would be enough, hard or soft. And again, what's it for within the logic?
Ya Untuk menghubungi layanan Cs Bitget melalui 0821-3737-7562 WhatsApp dapat diakses 24 jam, memungkinkan Anda mendapatkan bantuan dan informasi kapan saja.
Well, after reading through the documentation for a while, I found out there's a new checkout type (block) that seems to be a newer version. If you use this shortcode in your checkout page, the hooks do work:
[woocommerce_checkout]
I still can't find the hooks that work with the new type of blocks checkout, though.
I wonder how his indexing went. I have the same problem on my dictionary I just started, unscramble.com but I plan to keep adding to it. How do you keep adding to a dictionary when it is done. Add thesaurus? Or just create content related to dictionaries like news?
apkgurug is a platform that provides all kinds of updated Android & iOS APK mod versions, possibly<a href=" https://apkgurug.com/ > unlocked apps you need for Android targeted apk downloading apps or games.
When I first started learning to program, I don't even start with any mainstream programming language. I still remember fondly the Mediachance Multimedia Builder (MMB), it was the first thing that I learned back in 2012. It's a software that lets you create an app mainly for "autorun" for CD and thumb drive. It offers a simple to use and WYSIWYG editor and close to no coding skill to build an app. But it has a scripting support to create more complicated logic. This were my first experience with writing code.
As time progress, I learned HTML and CSS with Microsoft FrontPage 2003, then Adobe Dreamweaver CS5. Back then jQuery was the SOTA library for creating web apps. I also learned a bit about PHP but never really liked it.
Then I started learning about Visual Basic in Microsoft Office (VBA) and then I stumbled upon Visual Basic 6.0 where I can create a more "proper program" so to say. Windows Forms was my new toy. VB6 was already outdated in 2015, and I stared exploring .NET Framework 4.0 and Visual Studio 2010 Express. Back then there was no VS Community Edition and .NET is still a proprietary, compared to .NET 10 now which is open source. Various websites such as Stackoverflow, CodeProject, VBForums, CodePlex (now gone), Planet Source Code (now gone), random blogs, and MSDN (now MS Learn) were my primary source of learning materials. I also liked O'Reilly books as it provides a more concrete example of how things are done from an expert point of view (compared to random blogs).
I still used VB.NET until I changed to C# in late 2017. Then I got my first job in 2019 and was quite shocked to learn that I were assigned to a team with NodeJS as the main framework as I don't have any experience with it. My previous experience with reading technical blogs and documentations proved to be invaluable for my career. It allows me to learn new stuff quickly and adapt to strange areas such as web development using React/Vue (crazy gibberish: JSX)
My way of learning:
Plan a project
Try stuff until it worked
I remember switching from framework to framework, library to library, language to language, database to database. I never completely finished the project, but the journey learning various language, framework, libraries, and databases is the most fun thing about learning to program.
I argue that it is harder to start learning programming now even with abundance of resources, because the technological advancement and AI have gone so far that it is hard to learn the basics to the current industry standard. It would take a huge amount of time and efforts to understand the nuances of why certain application use X or Y approach. I still struggle when changing job because the tech stack is constantly changing.
If you're feeling lost and overwhelmed with the amount of information you need to learn, you're not alone. All great programmers have gone through similar process. I think having a mentor would be the best way to learn. Keep up the good work.
