Bartek Jaskulski

Speak (in your words) when spoken to

2026-10-05T22:54:43Z

I love that more and more voices online (at least those, who I happen to read by accident on Hacker News or Lobste.rs) emphasize that one should not simply be human proxy meat.

It truly dreads me, how on the course of few past years, with exponential degradation over less than last 12 months, people loose the ability to think critically. Actually, this seems to proceed inversly proportional to the technological advancements. I may sound like an old prick, praising the past, when everything was simpler, and scolding the rotten present. And maybe I am to some level. But I am definitely put up with reading verbatim citations from llm tools. Some stress, that if they wanted to read a language model's opinion on the matter, they would've promted it theirselves. For me this is not the point. I feel insulted too. Yet, the real reason it itches me, is the fact that you waste my time and neither of us learns anything.

Above may sound harsh, nonetheless, the AI sugarcoates everything, wastes tokens on weasel passages and forces me to dig through the mud to get to the essence. The most of us hates meetings, precisely because a poorly conducted gathering leads to inflation of words, voiding time of everyone present for a little value. You see the resemblance?

The learning part is the sad one. Back in the days, when I was occasionally playing on the stage, the most efficient mnemonic technique to learn by heart a few pages of text was to rewrite it by hand. Programming tutorials for absolute begginers follow the same pattern, actively encouraging to write down mundane code blocks, execute them, experiment. This is a valid technique for infusing one's mind with the knowledge. Currently being replaced by a shortcut which doesn't even require a minimal muscle flexing. Copy-paste teaches you nothing. Reading, then pasting, while may help you feel better about yourself (at least you know what you forward, right?), doesn't teach you anything either. The minimal effort you can put, that allows the knowledge to sink into your skin, is to write the thought in your own words. You don't have to understand it perfectly, that's the entire point. You are learning after all, constantly. When rewriting, you will find yourself skipping on some lines that are a pure fluff; you will find asking yourself follow up questions, that you can feed back to the llm to deepen the knowledge; you will remember a tiny fraction of what you've just written, then repeat it over and over, until you will grasp it and won't need an AI to express a valid point. And I won't be pissed at you.

I beat myself up to this day, for over a year, for submitting a fully generated internal architecture proposal (one about server management, when we had to decide stack to transition to). I wasn't simply asking the llm to hallucinate something barely holding the ground, just to check off my task. I really dove deep into servers' architecture, trying to strike a good balance between complexity, features and ease of management, having our team's level of expertise in mind. (I was, and still am, learning about servers; at that time with help of the ai, trying to uncover every nook and cranny). And I stood by every decision presented to the CTO. He gently told me how utter shit the document was, and that I musn't simply dump llm output into the internal docs.

I must admit, at first I was mostly sulking. I understood the form of the text could be better, as this one was full of obvious ai artifacts. Still, the content was something I spent 2 days of my work on and it was planned into the last detail. Or so I thought.

Nevertheless, from that day on, I decided not to use the AI in communication with other people. By no means I am against AI usage. This is my daily tool, effectively enhancing my brain power, speed and understanding. But I remain responsible for the final output, and I keep my quality gates high. I cannot embarras myself, if you'd like to ask me a follow up question and I'd have to put you on halt to wait for my llm response.

I expect the same from you. After all, I asked you a question because I want to know your opinion on the matter. You don't have to feel like you are the brightest mind in the room, especially when asked directly. It doesn't matter anymore. The AI already swept the field and can easily outsmart majority of the population (to some level). What I want is to be certain, I can rely on you, not as my stenotypist, but a person who considers available options, weigths the consequences and has the guts to call the shots.

Succumbing to the path of least resistance is what I fear the most and what frustrates me about passtrough copy of the llm output, not the fact, that I could type it myself in an identical chat window, using the same model make and receiving a response close to the one you just sent.

And if I ever use the AI, when writing to you, I'll inform you, along with an explanation on why I couldn't write it myself and sincerest apologies.

Using AI agents from mobile

2026-10-02T23:06:04Z

This one have been tempting, I keep thinking about how cool it would be to setup an agent ready to communicate with my phone in any way, like through whatsapp, ssh, either. And I know, I must never do it. For my sanity. Having always-on access to vibe coding would be like a doom-scrolling for programmer, a doom-vibing, really. I truly fear, I would succumb to filling up every minute of my free time (even going beyond that!) to tinker with projects, come up with new ideas, never stop working. And such vision scares me.

Even now, I installed github on my phone to have access to write down some of my thoughs whenever I feel like it, and here we are, 1 a.m. writing down a second post immediately.

So if I limit my working to 40-50h a week, it's not because I am sloppy, lazy and cannot keep the pace of moder worlds. I choose to move at my own speed, as things need time to grow, and rushing them is not always a viable solution.

I am also afraid of losing control over the code. This happens even with regular coding agents, but in terminal the loop is closer, I can easily inspect recent diff, complain about specific word choices or overall code organization. It helps with code ownership, understanding and sometimes even learning, when I set out to tackle what otherwise might be close to impossible for me. The very premise of mobile agents seems to be the opposite of that flow.

#ai

Fed up with dotfiles tricks

2026-10-02T22:08:47Z

Really, what is so great about zsh, exa and other cool tools. I admit, I've been swept by the idea once, mingling with my dots all the time, but it can only take you so far.

Like that boasting for zoxide with insanely smart and fast navigation... Did you hear of CDPATH? Combined with native fish features (which is superior shell) gives you superpowers in navigation and else beyond comprehension. Really, try it, type letter or two and arrow right, out of the box.

What I really try to say is that I cherish my minimal config. It's really fish, fd, rg, neovim, lf, some custom code living in my dotfiles and we're good to go. And when I ssh into server, I don't miss my fancy setup, just changing suit into bash, find, grep.

Focus on what can speed you up without slowdowns, and chop off unnecessary load ruthlessly.

Checking git commits based on repo content

2025-01-09T14:33:57Z

Interactive way to look for all git commits containing a string with further examination.

git log --all -S "SubscriptionEnd" --oneline | fzf --preview 'git show {1} -- ./src' --bind 'enter:execute(git show {1})'

Simple static page with github gist

2025-01-05T13:33:32Z

Gists allow for static html preview, which may be useful for POC or quick prototyping. Just drop a HTML file inside (named index.html or whatever if you have only one file).

You can access such gist at gitstpreview.github.io/?gist_id

https://gist.github.com/zlargon/9de497580ac33180e6128c187e2a93bc

#github

WordPress REST API always jsonifies response

2025-01-05T12:59:55Z

Each time, you return something as WP_REST_Response, content of this response is passed through json_encode function. This way it's impossible to serve any other content, especially relying on raw string. Such value would be encoded from "string" to JSON form of ""string"". As long as we use objects and arrays it's fine, but for scalar types (and content types other than JSON-based) we cannot rely on WP REST API directly.

To mitigate this, you can hook into rest_pre_serve_request and handle REST server response yourself.

#wordpress #php

Why may I suck at learning/working?

2025-01-05T12:59:55Z

Pitfalls

I find myself often wandering into rabbit holes. E.g. spending 10h+ to investigate the best way to organize links (bookmarks) and trying to write a shell script for that.

This time may be efficient, yet it's generally a waste because, besides brief learning of bash scripting, I don't provide any value to myself.

Any idea on overcoming rabbit holes? At the moment, not really. Just stick to the plan and maybe investigate whether your plan is worth pursuing in the firsthand.

Links:

https://www.coursera.org/learn/learning-how-to-learn

Students often compare themselves to others who are farther along in their coding journey or have more experience. This is a recipe for depression and frustration. Solution: Only compare yourself to your past self. Have your abilities and knowledge improved from where you were last week, last month, or last year? Be proud of the progress that you've made!

Removing commits from github

2025-01-02T06:39:51Z

In case of leaking secrets in github, the only viable way of removing commit from web access is to nuke the repository.

Removing commit with git filter-repo is the first thing to do, but it only removes it from plainly available history, still dangling and waiting to be pruned by git gc. And github is known to keep dangling commits for a long time¹. If you don't have a time to contact github support, make repository private and/or remove it completely (you can always recreate it later, but you will miss all stargazers, etc.)

git filter-repo --path path/to/remove1 --invert-paths
git push --force --verbose --dry-run
git push --force

#git

https://stackoverflow.com/a/32840254 ↩︎

WordPress i Cloudflare

2024-12-31T15:01:32Z

Ustawienie DNS potrafi namieszać z redirect loop w WordPressie.

Szybkie rozwiązanie to

define('FORCE_SSL_ADMIN', false);

Można też skorzystać z wtyczki¹.

#wordpress

https://community.cloudflare.com/t/endless-redirect-with-wordpress/3914 ↩︎

removing lines with sed

2024-12-31T15:01:32Z

sed -i '<line_no>d' <file>
sed '1d' dump.sql > no_head.sql

#linux #sed

Collation and accents in database

2024-12-31T15:01:32Z

It is generally recommended to use utf8mb4_general_ci collation for columns, but sometimes it may be misleading, especially in conjunction with UNIQUE constraint on VARCHAR field.

Some accents are cut off from general_ci, when comparing phrases and it may lead to constraint violation even though visually strings seems different. One example, I've stumbled upon is Hungarian: Kömlő and Komló. Take under consideration, that there are different languages, besides Hungarian in the database, so using specialized collation is not the case.

In such situation it's best to change column's collation to utf8mb4_bin, which compares characters by it's binary representation, but is unusable for any additional features provided by collation, like correct sort order (accents come last with _bin). Furthermore:

utf8_bin says to just compare the bytes. CHAR/TEXT utf8 with utf8_bin validates (on INSERT) that the bytes comprise valid utf8 bytes, but does not do anything useful for comparisions other than exact (no case folding, etc) equality. BINARY/BLOB should be usually be used instead CHAR+utf8; this stores the bytes without any checking.¹

#db

https://mysql.rjweb.org/doc.php/charcoll ↩︎

Adding existing user to a group

2024-12-31T15:01:32Z

sudo usermod -a -G groupname username

#linux

Get all links from website

2024-12-31T15:01:32Z

[...document.links]
    .filter(el => el.href.includes('https'))
    .map(el => el.href)

Later I can "Copy Object", and e.g. download the files, with

xclip -sel cli -o | jq '.[]' | xargs -n1 -P4 curl -O

#js

gzip tl;dr

2024-12-31T15:01:32Z

This will remove file after compression:

# Remove file after compression
gzip file

# Keep file
gzip -k file

# Decompress file, remove archive, store on fs
gzip -d file

# Decompress file, write to stdout
gzip -dc file

#linux

AI search

2024-12-31T15:01:32Z

There currently two interesting tools, when it comes to AI search engine: Phind and Perplexity.

The first one is positioned as coder's tool, and the functionality to chat with search is great for debugging and followup questions. Unfortunately, regular search with followup question doesn't keep a reference in memory. Additionally, they recently became aggressive about advertising pro features (yup, still using free version). I haven't been using it yet, but you can also add a part of "system-like" prompt, to force model to particular tone or so.

On the other hand, Perplexity is recommended by more users. At first I was thinking of using it only for regular search, not code related, but possibly their model surpasses Phind (although it messes with code formatting). Perplexity has similar customization availability as Phind, for providing user-defined rules, but there's additional feature, I've just found out: collections. Besides simple organization of your search, you can also add custom prompt, which will apply to all threads started in a specific collection.

#ai #search

WordPress repository runs basic PHP check upon plugin deploy, and it can be mischievous

2024-12-31T13:53:02Z

sometimes... Look at this:

***********************************
PHP error in: flexible-subscriptions/tags/1.0.0/vendor_prefixed/symfony/polyfill-php80/Resources/stubs/Attribute.php:
Parse error: syntax error, unexpected fully qualified name "\class_alias" in Standard input code on line 40
Errors parsing Standard input code
***********************************

Regular scoped code... I don't know which version of PHP can actually report such parse error, although I've been testing it with 8.0-8.3, 7.0, 7.1, and 5.6.

General recommendation – never scope polyfill code, as it makes no sense.

Differences between GPT-3.5-turbo and GPT-4

2024-04-05T09:03:14Z

As GPT-3.5-turbo is worse at following instructions, the most important stuff should be repeated in user prompt or at the end of system prompt. Any rules included at the very beginning of a prompt are likely to be forgotten, if are overloaded with additional context or instructions. This may be helpful when creating prompts which runs for both models, but only one of them has to take a specific action.

On the other hand, GPT-3.5-turbo is more likely to make mistakes in negation instructions, like "do not output anything except JSON". Using positive messages are to be likely more successful: "Respond only with JSON object, avoid any additional comments or wrappers like code fences"

Executing commands within `less`

2023-05-17T06:06:34Z

It was known to me, that vim can benefit from filters, but I've actually read some commands description in less and it brings multiple powerful features, like executing shell commands, etc.

Think of looking though git blame and executing git show with desired SHA. I cannot believe, I always exit less before, to span a new command, and repeat.

Ordinal numbers in PHP

2023-04-13T08:46:24Z

It's worth knowing that PHP's built-in NumberFormatter can format numbers in their ordinal form, i.e. 1st, 2nd, etc.

https://www.php.net/manual/en/class.numberformatter.php#numberformatter.constants.ordinal

Adding collaborative comments to code

2023-04-13T08:40:27Z

I miss document editor experience, when collaborating on new git repository, specifically adding comments to particular pieces of code. In GitHub or GitLab it is only possible to create a new issue, referring to line or attaching a piece of code, but for large amount of comments issues doesn't seem to be the best method.

For the time being what seems to be the best option is to create a new, empty branch and requesting a merge into it. With such solution you create a diff against void, and you are able to rather easily add multiple notes, when working from scratch on some software.

To create an emtpy branch:

git switch --orphan <new branch>
git commit --allow-empty

Em dash in vim

2023-04-02T07:00:50Z

I'm always frustrated about writing --, instead of – for en dash in vim, but using digraphs can be bizarre. Why haven't I thought earlier about making it a keybind, just as in most today's editors (LibreOffice, MS Office, etc)?

inoremap <buffer> --<space> –<space>

https://vi.stackexchange.com/questions/2199/what-is-the-easiest-way-to-insert-en-dash-in-vim

`$wpdb->prepare` gotcha with placeholders

2023-03-23T05:07:40Z

$wpdb->prepare is a function which works similarly to sprintf, but is targeted for sanitizing database query. I'm putting aside the superiority of using prepared statements.

Most commonly used placeholder is %s to insert a string value, but it may get more complicated when you try to prepare values inside WHERE IN statement beforehand and later dump them as one value. Whether it's correct may be another topic, but similar code will fail due to %s always being wrapped in quotes.

$wpdb->query($wpdb->prepare( "SELECT * FROM wp_posts WHERE ID IN (%s)", implode(', ', [1, 2, 3])));

Result is SELECT * FROM wp_posts WHERE ID IN ('1, 2, 3').

You can workaround this behavior using positional placeholders, but then you need to remember about wrapping with quote yourserf it may be needed.

Max index length in older MySQL

2023-03-22T19:18:22Z

A reminder: max index length in MySQL was 767 bytes. I learned that the hard way, as unique constraint on BIGINT and VARCHAR (255) resulted in exhaustion of this space. In general, anything with VARCHAR, when supporting UTF-8 is considered will probably be too big to put a valid index on it because max size of such column would take up to 1020 bytes (255 * 4).

Loading vim quickfix list from grep

2023-02-17T10:11:03Z

One of things, I always forget about, but seems cool.

vim -q <(grep -RHn <query>)

Malware detection in Automattic

2023-02-08T12:59:04Z

Each plugin submitted to woocommerce.com store has to pass their malware detection which is just php-malware-finder with some custom YARA rules.

You can easily check your plugin violations with following command:

docker run --rm $(pwd):/data ghcr.io/jvoisin/php-malware-finder

Custom rules from Automattic can be found in their GitHub repository¹.

https://github.com/Automattic/php-malware-finder ↩︎

The WordPress way of resolving update packages

2023-02-01T06:39:32Z

Plugin updates in WordPress can behave in rather indeterministic way, dependent on the way, plugin is archived and the source it comes from.

To begin with distribution package, as plugin author you have two possibilities to pack your plugin:

Inside single parent directory, which then becomes target name for plugin
Use flat directories architecture, i.e. after unpacking, your files lie in top-level directory

I used to thought that name of your zip package matters (e.g. my-plugin.zip), but in the first case cases it's not the truth.

$ unzip -l my-plugin.zip
Archive: /home/user/my-plugin.zip
awesome-plugin/awesome-plugin.php

In such case, WordPress will infer top-level folder name (awesome-plugin), making it my plugin slug, thus target update folder. Simple as that.

And in most cases it's preferable method of packing a plugin, as the only reliable and predictable.

Although, second method (flat-package) may lead to manipulation of plugin slug and deactivation. It only may because it depends on update method and differs between direct plugin upload and fetching remote package.

Following case will influence our plugin slug by zip package name -- slug becomes my-plugin.

$ unzip -l my-plugin.zip
Archive: /home/user/my-plugin.zip
awesome-plugin.php

Yet, this is only true while you upload your plugin directly to WordPress, e.g. through admin panel (or any package which is local to server).

When we deal with remote fetch of plugin package (and we deal with that 90% of time), package is saved on our server as temporary file, which later has great implications, as our zip is stored with 6 random chars suffix. You can imagine by this far, what will happen when we try to unzip such file.

$ unzip -l my-plugin-yUi4dk.zip
Archive: /home/user/my-plugin-yUi4dk.zip
awesome-plugin.php

Once again, WordPress grabs our destination plugin's slug from zip file name, and our slug becomes my-plugin-yUi4dk.

Additionally, during the upgrade there's a cleanup process which deletes old plugin. The flow goes like:

Click update plugin button.
New plugin release is downloaded and extracted as my-plugin-yUi4dk.
Previous plugin release is discarded.
WordPress cannot execute non-existing my-plugin/awesome-plugin.php, thus deactivates plugin.
Updated my-plugin-yUi4dk/awesome-plugin.php doesn't fit it's predecessor name, so it's considered separate plugin, but it's never been requested to activate it.
You end up with inactive plugin after update.

There's more to that when you consider private, paid plugins which authenticates through some API key, usually bond with plugin slug (as you should safely expect, this value will never be changed). When client tries to reactivate plugin, the API key is gone because what client activated is not our original plugin anymore.

Avoid saving not-string data to WordPress database

2023-01-18T19:18:43Z

Saving data to options (update_option) or post meta (update_post_meta) may be unpredictable at first when using booleans or integers, especially when you rely on data type.

WordPress supposedly serializes data before saving to database but that's not the case for scalar types, which are just casted to string.

Serialized false produces b:0, but casted to string is '' (empty string)

Getting such data from a database with standard functions (get_option, get_post_meta) returns any value as string. Don't you even think about strict type comparison!

When you try to save bunch of connected data values (e.g. plugin global settings) it may be the best to group those elements and save it as one option after json_encode.

Vim quickfix list on current word

2023-01-18T19:10:16Z

:vim <cword> **/*.php

<cword> works just as intended!

Remove interval within called callback

2023-01-18T19:08:47Z

Due to JavaScript's way of passing variables to closures you are able to remove interval or timeout created with native function within it's callback.

const i = setInterval(
    () => clearInterval(i),
    1000
);

#js

JSON in SQL (mysql)

2023-01-18T19:01:58Z

Since mysql 5.7 (it's important, I mostly use 5.6, eh... WordPress) tables can have JSON data type. It's convenient instead of serializing data to string. Also comes with a bunch of functions (didn't care about performance yet), capable of querying JSON data inside column.

ID | json_column

1 | {query: "my query"} 2 | {query: "not query"}

SELECT * WHERE JSON_EXTRACT(json_column, "$.query) = 'my query'

Run SQL query in one line CLI command

2023-01-18T18:44:01Z

mysql -uuser -ppass -e "SELECT 1;"

-e runs query provided as argument.

Create new user in Unix

2023-01-18T18:36:02Z

useradd -m <username>

-m option creates home folder for new user.

Afterwards it's needed to set up password

passwd <username>

Why or why not to write a blog

2021-03-19T22:00:19Z

Thought about starting a blog keeps me itching and bugging for some time, but I always procrastinated execution. I've been finding many reasons to do so, yet now I want to surpass them.

10 reasons why not to start a blog...

It's in fashion to create appealing posts which put something into the quantified matter. That's just easier for us, humans to perceive something in a limited scope, I think. Then, play a game and list some reasons against writing a personal blog.

You certainly don't have a time

Who does actually, when we all are running for betterness (at work, in life). There's just not much time to sit down and write some thoughts. Oh, maybe there would be a minute or two to write a tutorial, but definitely not something demanding deeper thinking.

You aren't good enough

There are many blogs around, yours wouldn't make a difference, so why bother? Especially when you don't have the skills to write about technical stuff or something that could bring peoples' attention. Leave it to experts.

You can't plan ahead

Seriously, even that post came straight from the head, without an outline or at least the slightest intention about a conveyed message, just free-fall. I guess it is obvious when looking at those short sentences.

You don't speak your voice

Everything you know is just someone's knowledge which you 'borrowed', and any idea you came up with is just another instance on the same matter, without freshness.

Cut the bullshit

Those few points (not 10 apparently) may seem harsh, but imposter syndrome, the flood of alike blogs, and many various reasons can keep you (it keeps me) from writing and producing your content. Well, I guess I can't focus on my gains, though I can list those too. Some of the great reasons for starting a blog would be:

improving language skills (I'm not native in English)
improving programming skills (after all, that's why I'm here)
contribute to the community (who knows, maybe I will come up with some useful tutorial or thought about programming)
have something to show in front of a potential employer

Fall for it

I think there's one message to myself (but not only myself): to fall freely into writing, especially when I feel, that's something not for me. It's great in many ways beyond short post can cover and one day it may become something meaningful, at least to myself.

And stop comparing yourself to ones, who are now on top, it took them many years to get there and they definitely earned it. If you follow, you would too.

Right now idea about writing a blog itches me even more with a variety of post ideas. We'll see if I follow them.