Forem: Thiago Massari Guedes

Bringing my old MacBook Pro 8,2 back to life (again)

Thiago Massari Guedes — Mon, 09 Feb 2026 03:01:44 +0000

(This is a continuation of the previous post "Bringing my old MacBook Pro 8,2 back to life")

I brought my old MacBook pro 2011 back to life, after years collecting dust in my garage, running Linux Mint beautifully. Fast and modern, but...

If I managed to get it working with Linux, I probably could get it working with OSX too. I decided, then, to get it back to life and using OSX Snow Leopard - the operating system it came with.

Installing using the original disc

When I tried to run the original installer, it access the GPU right away causing those weird rendering issues. With some searching, I found that I could set a nvram variable to disable the dedicated GPU. On Linux I ran (I am almost sure that was the command):

printf '\x07\x00\x00\x00\x01\x00\x00\x00' | sudo tee /sys/firmware/efi/efivars/gpu-power-prefs-fa4ce28d-b62f-4c99-9cc3-6815686e30f9

It worked and I got OSX Snow Leopard installed. I had a goal in my mind - being able to access my website using some web browser.

Did it work?

Of course not.

My WiFi network was not supported by that old mac version. After seeing that using that was going to be quite a lot of work, I decided to see what was the oldest version that would work.

Upgrading to the next release

When I started upgrading to Mac OSX Lion, instead of holding Option key (⌥), I pressed Command (⌘) + Option (⌥) + P + R and my nvram trick was gone.

After some more searching, I found that I could boot using Command (⌘) + S and boot to single user text mode. When there, I had to run:

nvram fa4ce28d-b62f-4c99-9cc3-6815686e30f9:gpu-power-prefs=%01%00%00%00

This would make the installer boot, but not the installed version.

To be able to boot into the newly installed OSX Lion - after some more research, I found that I could just delete the radeon drivers. As I deleted incorrect things and had to do this a few times again, I started to move the drivers and restore. Also, every update (Yes! Even Snow Leopard had Apple Software updates working), the drivers were back. Anyway, after many times, I found exactly what to do. I created a bkp directory and ran every software update and every OS upgrade:

This is the /System/Library/Extensions/ of the installed drive

cd /Volumes/Thiago-MBP/bkp
mv ../System/Library/Extensions/*Radeon* .

Eventually in OSX Mavericks (AFAIR), I got WiFi working, but Safari was still not recognising modern certificates

Upgrading jorney

At some point, I thought: I had everything working nicely with Linux Mint - and then I decided to go back to something outdated and full of issues as even new software does not support those old OSX versions and new OSX versions don't support my old Mac. Why do I do that to me?

Anyway, thanks to Internet Archive and lots of patience, I found all the OS installers I needed. I installed and upgraded to:

Snow Leopard
Lion
Mountain Lion
Mavericks
Yosemite
El Capitan
Sierra

And when I got to Sierra

FINALLY!!!

Is it better than the Linux Mint experience?

No.

Interesting, I found that Youtube is not flaky using this version of OSX + Firefox and it was with Mint. Everything else was better, much better on Linux.

Getting any tool installed was harder as everything dropped support for OSX Sierra. I had to compile most of the tools from source (Luckily not Firefox) and remembered how nice is TextMate - which I am using right now to write this post.

What's next?

I am trying to decide if I will stick with OSX here or try something different, such as FreeBSD or WindowMakerLive.

Let's see

Original post: https://thiagocafe.com/view/20260209_bringing_my_old_macbook_pro_82_back_to_life_again/

Bringing my old MacBook Pro 8,2 back to life

Thiago Massari Guedes — Sun, 08 Feb 2026 04:24:20 +0000

This is my old MacBook Pro 8,2 2011. It was in my garage collecting dust after it's dedicated GPU got toasted. I am not great with electronics, so replacing the soldered Radeon GPU was not something I thought I'd ever do.

(Sorry, there are way too many images. I added, however, a link to my blog this time)

Full content:
https://thiagocafe.com/view/20260208_bringing_my_old_macbook_pro_82_back_to_life

Compiling Window Maker in a Debian-based OS

Thiago Massari Guedes — Mon, 17 Nov 2025 00:12:12 +0000

I wanted to study Window Maker source code and change a few things. To do that, I need to compile it from source and change ligthdm to use my custom compiled version.

Unfortunately, there is not much documentation about compiling Window Maker from source, so I am adding it here so others (and myself in the future) can reuse this.

First: Install dependencies

Start by installing deb packages needed to compile Window Maker

# Build tools
sudo apt-get install build-essential \
     autoconf automake libtool gettext git

# Base libraries
sudo apt-get install libx11-dev libsm-dev \
    libpango1.0-dev libbsd-dev

# Image libraries
sudo apt-get install libxft-dev libfontconfig-dev libfreetype6-dev \
libxpm-dev libjpeg-dev libpng-dev libtiff-dev libgif-dev libwebp-dev \

# Imagemagick library support
sudo apt-get install libmagickwand-dev

Let's create the alternative installation directory

sudo mkdir -p /opt/wmaker
# Optionally, change ownership to your user to simplify later changes
sudo chown $(id -u):$(id -g) /opt/wmaker

Compiling Window Maker

git clone https://repo.or.cz/wmaker-crm.git
cd wmaker-crm

./configure --prefix=/opt/wmaker \
    --enable-modelock \
    --enable-pango

make -j $(nproc)

# add sudo if you didn't change /opt/wmaker ownership
make install
# or if you don't want debug symbols (results are 5MiB binaries)
make install-strip

Configuring LightDM to use custom Window Maker

And last, let's create a custom session file to select our custom Window Maker

Edit with your favorite text editor:

sudo edit /usr/share/xsessions/wmaker-custom.desktop

Add the following content:

[Desktop Entry]
Name=Window Maker Custom
Comment=Use Window Maker from /opt/wmaker
Exec=/opt/wmaker/bin/wmaker
Icon=wmaker
Type=XSession

Restart with:

sudo /etc/init.d/lightdm restart
# or
sudo systemctl restart lightdm.service

That's all. Now you can enjoy your own compiled version of Window Maker

Post in the author's blog

Compiling Window Maker in a Debian-based OS

Fixing emoji support in the Linux terminal

Thiago Massari Guedes — Mon, 17 Nov 2025 00:09:12 +0000

Recently, tired of the huge accidental complexity of Gnome and KDE, I decided to switch back to Window Maker (again). Being a big terminal user, when I was setting up the terminal, I noticed that emojis were not being displayed correctly by Starship prompt.

First thing, I installed Nerd Fonts from their website nerdfonts.com. Guess what - no emojis.

Then, digging a bit more, I found that some emojis were being displayed properly, but curiously the ones being displayed by Starship prompt were all missing.

To my luck, there is a deb package that contains all the Emojis and can be used as a fallback for missing emojis:

sudo apt-get install fonts-noto-color-emoji

After installing it and restarting the terminal, I got all emojis being properly rendered.

Finally an easy one.

Post in the author's blog

Fixing emoji support in the Linux terminal

LuSH 0.15 with syntax sugar

Thiago Massari Guedes — Sun, 06 Jul 2025 22:01:52 +0000

This new release of LuSH brings some interesting and important improvements (not officially supported by LUA), being:

String interpolation

local name = "Thiago"
env.print("My name is ${name}")

String split

local items = string.split("string with spaces", " ")

Shell exec

print("Lush scripts:")
$> ls scripts | grep ".lush"

outputs:

Lush scripts:
mod1.lush
test2.lush
test3.lush

Subshell exec

print("Lush scripts, again:")
local scripts = $( ls scripts | grep ".lush" )
local tok = string.split(scripts, '\n', false)
for i, v in ipairs(tok) do
    print(i .. ': ' .. v)
end

outputs:

Lush scripts, again:
1: mod1.lush
2: test2.lush
3: test3.lush

Installing:

To install, simply run:

cargo install lush

OpenCV cv::Mat without copying and weird memory ownership

Thiago Massari Guedes — Thu, 01 May 2025 02:02:59 +0000

Recently I found an interesting OpenCV behaviour.

OpenCV cv::Mat is built to be easy to use. It will allocate, free and reallocate it's own internal data. That's fine for most of the uses, but what if you are sensitive to extra memory allocations?

As you might expect, C++ developers very often track the memory ownership of any allocated piece of memory, especially when copying memory is not an option.

To use cv::Mat without copying the buffer, you can use this class constructor that does not own the memory:

cv::Mat::Mat(
  Size size,
  int type,
  void *data,
  size_t step = AUTO_STEP
);

In the constructor documentation, you can read:

Pointer to the user data. Matrix constructors ... do not allocate matrix data. Instead, ... which means that no data is copied. ... The external data is not automatically deallocated, so you should take care of it.

So far, so good, let's see if this is true

cv::Mat original_image = cv::imread(image_path);
std::vector<uint8_t> image_data(original_image.rows * original_image.cols);
cv::Mat gray_image(original_image.rows, original_image.cols, CV_8UC1, image_data.data());

cv::cvtColor(original_image, gray_image, cv::COLOR_BGR2GRAY);
printf("AFTER: cv::Mat data: %p\n", image_data.data());
printf("AFTER: image_data data: %p\n", gray_image.data);

It prints

AFTER: cv::Mat data: 0x138168000
AFTER: image_data data: 0x138168000

Ok, that's great. The data is the same. However, what happens if the given data array is not big enough?

Let's change the example to remove one byte of the size.

cv::Mat gray_image(original_image.rows, original_image.cols - 1, CV_8UC1, image_data.data());

Surprisingly it works. Why?

void cv::cvtColor(InputArray src, OutputArray dst, /* ... */);

I found that OutputArray, when the allocated data is not big enough, allocates a new array - and that's where I found the surprise. If the data is not owned by cv::Mat, the pointer is silently replaced by a new allocated one, owned by cv::Mat

If I run the example again, let's see the addresses:

AFTER: cv::Mat data: 0x160168000
AFTER: image_data data: 0x1601e0000

In conclusion, if you are using cv::Mat providing the memory yourself, you need to be sure the allocated buffer is enough, and preferably also check the image.data to verify if the provided buffer was used.

Original post: thiagocafe/OpenCV Mat without copying and weird memory ownership

Using flutter with native resources on apple silicon processors

Thiago Massari Guedes — Fri, 20 Dec 2024 05:53:07 +0000

If you are building a flutter application for apple processors using Silicon processors (M1+) and dealing with native packages for things such as playing a sound, you maybe encountered this error:

Error output from CocoaPods:
↳
    /System/Library/Frameworks/Ruby.framework/Versions/2.6/usr/lib/ruby/2.6.0/rubygems/core_ext/kernel_require.rb:54:in `require': cannot load such file --
    ffi_c (LoadError)
        from /System/Library/Frameworks/Ruby.framework/Versions/2.6/usr/lib/ruby/2.6.0/rubygems/core_ext/kernel_require.rb:54:in `require'
        from /Library/Ruby/Gems/2.6.0/gems/ffi-1.17.0-x86_64-darwin/lib/ffi.rb:5:in `rescue in <top (required)>'
        from /Library/Ruby/Gems/2.6.0/gems/ffi-1.17.0-x86_64-darwin/lib/ffi.rb:2:in `<top (required)>'
        from /System/Library/Frameworks/Ruby.framework/Versions/2.6/usr/lib/ruby/2.6.0/rubygems/core_ext/kernel_require.rb:54:in `require'
        from /System/Library/Frameworks/Ruby.framework/Versions/2.6/usr/lib/ruby/2.6.0/rubygems/core_ext/kernel_require.rb:54:in `require'
...

Error: To set up CocoaPods for ARM macOS, run:
  sudo gem uninstall ffi && sudo gem install ffi -- --enable-libffi-alloc

You probably realised that running sudo gem uninstall ffi && sudo gem install ffi -- --enable-libffi-alloc will not work. You are going to get the same error.

The solution I found was to use cocoapods from brew.

For that, run:

sudo gem uninstall cocoapods
brew install cocoapods
brew link cocoapods

Last thing, as I am quite new to flutter and some of the Apple ecosystem, if anyone has a better suggestion, I am all ears.

Link to the author blog: Using flutter with native resources on apple silicon processors

sgrep 1.0 released

Thiago Massari Guedes — Wed, 27 Nov 2024 05:19:56 +0000

Have you ever wanted to see the top line when running:

❯ ps -ef | grep bash
  501 56821 56820   0 17Nov24 ttys000    0:00.27 /opt/homebrew/bin/bash -l
  501 95638 56821   0  8:19pm ttys000    0:00.00 grep bash
  501 76214 56820   0 Mon09am ttys002    0:00.44 /opt/homebrew/bin/bash -l
  501 82952 56820   0  7:03pm ttys003    0:00.23 /opt/homebrew/bin/bash -l

You can with sgrep

❯ ps -ef | sgrep -t 1 -p bash
0:   UID   PID  PPID   C STIME   TTY           TIME CMD
642:   501 56821 56820   0 17Nov24 ttys000    0:00.27 /opt/homebrew/bin/bash -l
644:   501 95648 56821   0  8:19pm ttys000    0:00.00 sgrep -t 1 -p bash
645:   501 76214 56820   0 Mon09am ttys002    0:00.44 /opt/homebrew/bin/bash -l
646:   501 82952 56820   0  7:03pm ttys003    0:00.23 /opt/homebrew/bin/bash -l

The main idea of sgrep is to have simple command line options, simple to use.

Installing

Using brew:

brew tap thiagomg/texted
brew install thiagomg/texted/simplegrep

Or using cargo

cargo install sgrep

Deserialising binary data in Rust

Thiago Massari Guedes — Thu, 15 Aug 2024 04:25:10 +0000

One common way to deal with serialised data in C++ is to map to a struct, but in Rust with memory ownership and some smart memory optimizations, it is a little more complicated.

For example, let's say a producer in C++ serialises this message:

struct book_msg {  
    uint8_t major_version, // byte 0: message major version  
    uint8_t minor_version, // byte 1: message minor version  
    uint8_t msg_type,      // byte 2: message type  
    uint8_t title[20]      // byte 3-23: title of the book  
}
// ....
auto msg = create_msg();
comm.send(&msg, sizeof(book_msg));

How can we deserialise that in Rust?

Creating the struct to be filled with the data

Rust does not guarantee that the order of the struct arguments is maintained, so we need to use #[repr(C, packed)] to tell the compiler that the order of the arguments has to be maintained

#[repr(C, packed)]
struct BookMsg {  
    pub major_version: u8, // byte 0: message major version  
    pub minor_version: u8, // byte 1: message minor version  
    pub msg_type: u8,      // byte 2: message type  
    pub title: [u8; 20]    // byte 3-23: title of the book  
}

Mapping the bytes to a struct

Let's say we received:

// 013854546865205369676e206f662074686520466f7572 in hex
let msg_vec: Vec<u8> = comm.recv();

to map this vector to a struct, we need to violate rust safe guarantees as rust cannot verify in compile time that the struct maps to the vector size and data types.

Explanation of those 2 lines:

We're casting the pointer from *const u8 to *const Msg
And then we are accessing it as a value *
And finally, getting a reference to this value &

let msg_bytes: *const u8 = msg_vec.as_ptr();  
let mapped_msg: &BookMsg = unsafe { &*(msg_bytes as *const BookMsg) };

Beware. When accessing parts of the message, it's better to copy the value to an external variable to avoid memory aligning issues.

let major = mapped_msg.major_version;  
let minor = mapped_msg.minor_version;  
let msg_type = msg.msg_type;  
let title = msg.title;

// Result:
// msg version=1.56, type=84
// msg title="The Sign of the Four"

Summary

Use structs with #[repr(C, packed)] to ensure the layout is preserved
Use pointers to map to this struct
Copy the value from the struct to an external variable to avoid memory alignment issues

Source: Deserialising binary data in Rust

RSS support added to Texted

Thiago Massari Guedes — Sun, 04 Aug 2024 21:35:59 +0000

A Rust developer on twitter posted:

"It would be nice to reed updates to your blog in an RSS feed."

Indeed that is a great idea and I implemented on Texted 0.3.6!

Texted is the platform powering both my blogs, Thiago Café and Blogaro!

Happy reading

Texted 0.3.4 released

Thiago Massari Guedes — Fri, 07 Jun 2024 06:18:34 +0000

Texted 0.3.4 released

Very close to release version 1.0, it comes with:

Simple Markdown support
Markdown with images
HTML support with and without images
texted-tool:
- creation of posts
- removing special characteres from post url
- bootstrap of a new blog with a simple command

Full Changelog

Enjoy and reach out for any questions.

Quick tip: Type of tokio spawn return

Thiago Massari Guedes — Tue, 04 Jun 2024 05:07:27 +0000

When I was implementing the metrics task using tokio, I wanted to save the result JoinHandle in a struct and I saw the type being displayed by the IDE: JoinHandle<?>

What does it mean?

When I looked the definition of the function spawn, that's the code:

pub fn spawn<F>(future: F) -> JoinHandle<F::Output>  
where  
    F: Future + Send + 'static,  
    F::Output: Send + 'static,  
{
// ...

In the Future trait, F::Output refers to the return value of the input function.

Now, the piece of code I have is a long running task. That is the piece of code.

let receiver_task = tokio::spawn(async move {  
    println!("Starting metrics receiver");  
    while let Some(event) = rx.recv().await {  
        if let Err(e) = metrics.add(&event.post_name, &event.origin) {  
            error!("Error writing access metric for {}: {}", &event.post_name, e);  
        } else {  
            debug!("Metric event written for {}", &event.post_name);  
        }  
    }  
});

As this function returns nothing, the type is Unit, hence for this lambda I can declare my struct as:

pub struct MetricHandler {  
    receiver_task: JoinHandle<()>,  
    //...
}

And now you can do:

let receiver_task = tokio::spawn(async move {
// ...
MetricHandler {
    receiver_task,
    // ...
}