libvaxis: a TUI library for zig

[obsidian rover]

When spamming it, the terminal is mid-event in it's write

So it sends a bunch and the last one is \x1bO

And the next read will be B\x1bOB....

[old skiff]

i just pushed to this branch https://github.com/Rexicon226/osmium/tree/debug

[wooden lantern]

libvaxis doesn't currently support the style used in the post does it?

[obsidian rover]

Not natively no, you would have to make it your own

It's complicated to do well

[wooden lantern]

ahhh, I see. thanks for pointing it out though :)

[obsidian rover]

What do I run to get into the tui?

[old skiff]

ok so build it

then uh

i guess you could just run any old python file

do zig-out/osmium examples/calculator.py --debug

[obsidian rover]

hmm

it doesn't build

[old skiff]

:(

[obsidian rover]

what zig version?

[old skiff]

what's the issue?

13

[obsidian rover]

error: FileNotFound
/home/tim/.zig/versions/zig-linux-x86_64-0.13.0/lib/std/process/Child.zig:423:5: 0x1196f67 in waitPosix (build)
    return self.term.?;
    ^
/home/tim/.zig/versions/zig-linux-x86_64-0.13.0/lib/std/process/Child.zig:300:9: 0x1196fe3 in wait (build)
        try self.waitPosix();
        ^
/home/tim/.zig/versions/zig-linux-x86_64-0.13.0/lib/std/process/Child.zig:400:17: 0x1138b5f in run (build)
        .term = try child.wait(),
                ^
/home/tim/repos/osmium/tests/matrix.zig:30:20: 0x11387ed in addCase (build)
    const result = try std.process.Child.run(.{
                   ^
/home/tim/repos/osmium/tests/matrix.zig:21:26: 0x1139310 in addCases (build)
        const test_run = try addCase(b, test_file, exe);
                         ^
/home/tim/repos/osmium/tests/cases.zig:19:30: 0x1139878 in addCases (build)
        parent_step.dependOn(try matrix.addCases(b, dir, exe));
                             ^
/home/tim/repos/osmium/build.zig:97:5: 0x113a5e0 in build (build)
    try cases.addCases(b, exe, test_step);
    ^
/home/tim/.zig/versions/zig-linux-x86_64-0.13.0/lib/std/Build.zig:2117:24: 0x111caa7 in runBuild__anon_8821 (build)
        .ErrorUnion => try build_zig.build(b),
                       ^
/home/tim/.zig/versions/zig-linux-x86_64-0.13.0/lib/compiler/build_runner.zig:301:9: 0x1117dcf in main (build)
        try builder.runBuild(root);
        ^
error: the following build command failed with exit code 1:
/home/tim/repos/osmium/.zig-cache/o/ed52e6c312b90fa99360cb68c1cf0811/build /home/tim/.zig/versions/zig-linux-x86_64-0.13.0/zig /home/tim/repos/osmium /home/tim/repos/osmium/.zig-cache /home/tim/.cache/zig --seed 0xeda4b3bb -Zed9a954b4dc94e10

[old skiff]

uh, well what's further up the stack trace?

ah well, you're running the tests

they expect you to have a python3.10 binary in your PATH

i probably should make it use the python it builds instead, but i can do that later

just build with zig build osmium

[obsidian rover]

Ah, I was just doing zig build

Huh

Still the same error

[old skiff]

actually i dont have a step for osmium

lol

interesting

would you just comment out that try cases.addCases line?

[obsidian rover]

Ok it's building now!

[old skiff]

nice

[obsidian rover]

I assume this is on ghostty?

[old skiff]

nope

it's rio

lol

[obsidian rover]

ok

[old skiff]

hence me being 70% sure it's the terminal's fault

[obsidian rover]

I was wondering why it was getting SS3s for arrows, I know ghostty does the CSI variant by default

Both are valid

[old skiff]

does mouse wheel down just translate to arrow down?

[obsidian rover]

Sometimes, yes

There is a mode that when in alternate screen, mouse wheel sends arrow keys

vaxis doesn't enable this - so rio must have it on by default, or maybe only on if mouse reports aren't turned on

Ok I am in the tui and I can press arrows right away and it goes between the two

I can't trigger the panic

I will install rio

[old skiff]

you tried scrolling really hard?

[obsidian rover]

Yeah

I mean as hard as I can

Huh

rio ...doesn't like my input

[old skiff]

wdym?

lol

[obsidian rover]

I type and nothing happens

[old skiff]

i can confirm that the panic only happens in rio

so i must assume it's a rio skill issue

i'll just scroll really gently

[obsidian rover]

I am still going to add this if statement for a partial event

I didn't allow for partial events when I first wrote the parser

I just am not sure how you ended upa t the else arm

[old skiff]

cuz it doesn't handle the 'O' char

why is 'O' used here for scrolling any idea?

[obsidian rover]

\x1b O O is "/"

I mean, unless rio has opted for non-standard encodings

do you also have kitty installed?

[old skiff]

kitty doesn't run on windows

[obsidian rover]

kitten show-key would be interesting to see what you generate from your mouse

Ohh this is windows?

[old skiff]

mhm

well

it's a windows terminal that is sshed into a linux machine

i'm 99% sure that ssh isn't the issue

[obsidian rover]

ok in your shell there, type cat -v and press arrows and your mouse wheel

on the linux machine i mean

[old skiff]

left arrow then right arrow

mouse wheel doesn't do anything

[obsidian rover]

Ok, these are what ghostty sends as well

ok now try printf "\x1b[?1049h" && cat -v

Your mouse will probably send stuff there

[old skiff]

this is scrolling down

[obsidian rover]

ok

That all seems correct

Oh wait

I think this is a rio bug

[old skiff]

me as well :P

[obsidian rover]

Ok so you set it to return 0 bytes and you get an unhandled 1b

[old skiff]

that is O

yes

[obsidian rover]

1b is escape

[old skiff]

ah

[obsidian rover]

I would guess rio is writing all their events, and the last one fails to write completely - so they rewrite it at the beginning of the next syscall

So the sequential writes come in as \x1bOB\x1O....next write \x1bOB\x1bOB...

[old skiff]

ah

that makes sense

[obsidian rover]

they didn't writeAll 😉

[old skiff]

time to go dig through rio

im pretty sure it's in rust

[obsidian rover]

gl! I know zero rust

Yeah, i had to install rust to build it 🙂

man that will throw off my parser state really bad

[old skiff]

at least it isn't your bug :P

[obsidian rover]

That's why you are getting hte OB spammed

Because it eats \x1bO\x1b and the rest never gets parsed as an escape sequence

Becaaause it's now off by one

that is my bug 🙂

Need to recover from that nicely

[old skiff]

returning the N = length is the best way :P

just skip over it

that would probably mess up some further stuff tho

[obsidian rover]

That works in this case with rio, but presumably anyone could send an \x1bO\x1b sequence at anytime

The standard state machine handles this fine

I just made some shortcuts for input parsing

[old skiff]

ah

[obsidian rover]

And because - until now, i have never seen a borked escape sequence like this

[old skiff]

i am sort of an expert at breaking things

im about to suffocate in my room, brb while it builds :D

this cali summer is no joke

[obsidian rover]

Just pushed the fix

    const key: Key = switch (input[2]) {
        0x1B => return .{
            .event = null,
            .n = input.len,
        },

I also guard for the len < 3, but by this point we've already check that 0 and 1 are what we want, but if 2 is an escape, we have an invalid event - consume the bytes and move along

Ah...it should be input.len -1

Well, it should be 2 🙂

[old skiff]

it must be 2

because it would have paniced on the earlier index access

[obsidian rover]

Yeah

Ok it's fixed

You should be panic free

[old skiff]

yep

and the warning is gone of course

currently installing vs 2017 so i can build rio

gonna try tracking down this rio bug

[obsidian rover]

There is a chance this is not a bug on their end

I mean, it's kind of a bug but they could be delivering all of the correct events already

It's only a bug if the partial write results in a mismatch of the number of events sent

SO don't kill yourself on it 😉

[old skiff]

i will take that as an excuse to not hunt for it, and get back to designing this bad tui :P

this BOBOBO thing is kinda annoying tho

[obsidian rover]

It shouldn’t still be spamming that?

[old skiff]

tis is

[obsidian rover]

(╯°□°)╯︵ ┻━┻

[old skiff]

no panics tho!

[obsidian rover]

Alright I will take a look this week at making the parser more robust

[old skiff]

aside and possibly silly question. i'm printing this line to seperate cli commands but i dont know how to correctly calculate how long it should be

right now i have:

try writer.writeBytesNTimes("⎯", state.text_width);

it's "" because that character is a utf8 and doesnt fit in ascii

text_width is just the width of the box

but it's a bit too long right now

[obsidian rover]

This is a child window?

It should be win.width

[old skiff]

yeah it's one level down from the vx.window(), where the width passed to the child function is .{ .limit = win.width / 2 }

[obsidian rover]

The child window has it's width set when you create it

[old skiff]

yeah

and i use that width after creating it

[obsidian rover]

you are writing direct to the tty?

[old skiff]

no

[obsidian rover]

Or is that in the scrollview?

[old skiff]

so writer is an ArrayList(u8) which buffers the input

then i create a Segment whos text is the items of that array list

[obsidian rover]

Ah ok

[old skiff]

all the logic there is fine i think, i just dont know how to correctly calculate the width of the character

it seems to be a bit larger than a cell

[obsidian rover]

win.gwidth

[old skiff]

maybe i just chose a stupid character

[old skiff]

o

uno momento

[obsidian rover]

Something seems off though

It should be one cell

Oh maybe not

yeah that is a weird one

[old skiff]

what character do you use for the borders?

i'll use that

[obsidian rover]

hopefully measuring with gwidth will fix it, otherwise use a box drawing character: ━

https://en.wikipedia.org/wiki/Box-drawing_characters

[old skiff]

using the box drawing character worked perfectly

[obsidian rover]

A lot of terminals will custom draw these to have no gaps (ghostty, kitty, foot, alacritty, etc)

[old skiff]

it is a bit thick though

[obsidian rover]

There is a thinner variety

"─"

[feral charm]

the shitty ass kernel console:

how has no one been able to make a decent-looking alternative to the kernel console

[old skiff]

why would you

[obsidian rover]

That hting has to work everywhere and be tiny!

and have an embedded font that is tiny as well

[feral charm]

tru

that's complicated

[old skiff]

well using gwidth isn't really useful because it obviously can't return a fractional width or something

probably rio is having issues with rendering

using ─ is perfect

[old skiff]

rio crash #1 😄

[fleet schooner]

@obsidian rover The unicode on windows fix is not quite as simple as I had hoped. It seems we need to decode UTF-16 first before we can encode to UTF-8.

which makes sense

[fleet schooner]

The PR is ready now.

[fleet schooner]

@obsidian rover it looks like bracketed paste is supported in conhost. Does this mean we can enable it somehow in libvaxis on windows?

[obsidian rover]

Yes - it should be enabled by default though.

We just blindly issue the enabling code

[fleet schooner]

ah ok. I will test it again. I thought it was not working.

[fleet schooner]

nope, it doesn't seem to work

It looks to me like windows\Tty.zig eats all escape sequences and never actually sends them to parser.parse anywhere. So I don't see how bracketed paste can work.

[obsidian rover]

Ohh right. Because it's using the win32 api only I didn't plumb in any escapes

Does it allow bracketed paste mode if there is no vt processing?

[fleet schooner]

I guess not. I tried passing all escape sequences to Parser.parse, but I didn't magically get bracketed paste support.

It might still be a good idea to actually parse escape sequences into events even on windows. Do you want me to PR my changes?

[obsidian rover]

Yeah - I think I remember reading a case where at least on the input side you can send VTs even without vt input processing turned on

So there's likely some case somewhere that happens on the output side

[fleet schooner]

Yes, I've definitely seen a few escape sequences arriving. That's why I assumed there would be bracketed paste sequences too.

[wooden lantern]

@obsidian rover any idea why when in -Doptimize=ReleaseSafe, debug stuff is still printed to the term?

is there a way to stop this without redirecting output

[obsidian rover]

Not sure why - that is odd. One of my plans for this week is to clean up logging and put it into a single log scope so it can be more easily filtered with a custom logFn

(Which is how I would stop it: custom logFn and filter the scopes you don’t want)

[wooden lantern]

sounds good, thanks for that man

any updates on the github issue I put up btw? It's all good if you haven't had the time, i appreciate your work :) just curious

[old skiff]

are you passing in the optimization mode to the libvaxis dependency

[obsidian rover]

Haven’t touched it again yet…I started but it was more complicated than I originally thought so needed to spend some more time on it

[wooden lantern]

no stress man, take your time :)

[wooden lantern]

:O just target, shit. let me try that now

same thing lol

fn build_targets(b: *std.Build, build_options_module: *std.Build.Module) !void {
    const targets: []const std.Target.Query = &.{
        .{ .cpu_arch = .aarch64, .os_tag = .linux },
        .{ .cpu_arch = .x86_64, .os_tag = .linux },
    };
    const optimize = .ReleaseSafe;

    for (targets) |t| {
        const target = b.resolveTargetQuery(t);
        const libvaxis = b.dependency("vaxis", .{ .target = target, .optimize = optimize }).module("vaxis");
        const fuzzig = b.dependency("fuzzig", .{ .target = target, .optimize = optimize }).module("fuzzig");

        const exe = b.addExecutable(.{
            .name = "sftm",
            .root_source_file = b.path("src/main.zig"),
            .target = target,
            .optimize = optimize,
        });
        exe.root_module.addImport("fuzzig", fuzzig);
        exe.root_module.addImport("vaxis", libvaxis);
        exe.root_module.addImport("options", build_options_module);
        b.installArtifact(exe);

        const target_output = b.addInstallArtifact(exe, .{
            .dest_dir = .{
                .override = .{
                    .custom = try t.zigTriple(b.allocator),
                },
            },
        });

        b.getInstallStep().dependOn(&target_output.step);
    }
}

[still hare]

fyi you don't need custom logfn to filter log scopes, std_options has log_scope_levels

pub const std_options: std.Options = .{
  .log_scope_levels = &.{
    .{
      .scope = .terminal,
      .level = .err,
    },
  },
};

i'd prefix the scopes in vaxis with vaxis_ though, it's good idea for libraries to do that imo

[obsidian rover]

You have to filter each scope then right, there is no globbing on the enum?

[still hare]

yeah you have to add all of them to the array

you can do it reverse too though, have default scope level be err and only list your own scopes to override

[obsidian rover]

I was thinking of dropping them all under a single .vaxis scope to make it easy. I'm not sure more granularity than that is needed

But prefixing would be a good idea too - we'll see where it ends up!

[still hare]

yeah as long as the scope names avoid name collisions was the point

[obsidian rover]

I guess you could filter with std.mem.startsWith(u8, @tagName(.scope), "vaxis") in a custom logFn

[still hare]

@tagName(scope) probably :P

[wooden lantern]

Hey @obsidian rover, is there planned mac support for pty?

[obsidian rover]

The terminal widget? Yes

[stone tendon]

any ideas on why the terminal would glitched and a weird offset after quitting ?

[obsidian rover]

If you didn’t call tty.deinit

[stone tendon]

that fixed it ! they were missing in some examples, thanks for the help !

[obsidian rover]

If you are willing to, a PR to fix the examples you found without it would be awesome!

[stone tendon]

yess planning to, just tinkering with the library and get the hang of it, it s very nice so far

[stone tendon]

when handling events, if the event won't change anything visually, is it a good idea to just continue instead of rerendering everything ?

[obsidian rover]

Yep! If there are no visual changes, then no reason to re-render.

[stone tendon]

any suggestions on how you handle borders ? to avoid this type of stuff 😂

Im using the box drawing characters you ve mentioned before

from the wiki

[obsidian rover]

The gap where they meet? If you know the exact coordinate then I would draw the lines and then at that cell draw the connecting glyph

[stone tendon]

yess

the way im drawing them is the following

[obsidian rover]

┬

[stone tendon]

header:

pub fn header_component(parent: vaxis.Window, opts: ?OffsetOptions) !vaxis.Window {
    const opts_or_default = opts orelse OffsetOptions{};
    const header_child = parent.child(.{
        .x_off = opts_or_default.x_off,
        .y_off = opts_or_default.y_off,
        .height = .{ .limit = parent.height * 3 / 100 },
        .width = .{ .limit = parent.width },
    });

    const segment: vaxis.Segment = .{
        .text = header,
    };
    _ = try header_child.printSegment(segment, .{
        .col_offset = header_child.width / 2 - header.len / 2,
        .row_offset = header_child.height * 20 / 100,
    });

    const h_cell: vaxis.Cell = .{
        .char = .{ .grapheme = "━" },
    };
    for (0..header_child.width) |i| {
        header_child.writeCell(i, header_child.height - 1, h_cell);
    }

    return header_child;
}

sidebar:

pub fn sidebar_component(parent: vaxis.Window, opts: ?OffsetOptions) !vaxis.Window {
    const opts_or_default = opts orelse OffsetOptions{};
    const sidebar_child = parent.child(.{
        .x_off = opts_or_default.x_off,
        .y_off = opts_or_default.y_off,
        .height = .{ .limit = parent.height - opts_or_default.y_off },
        .width = .{ .limit = parent.width * 30 / 100 },
    });

    const border_char: vaxis.Cell = .{
        .char = .{ .grapheme = "┃" },
    };
    for (0..sidebar_child.height) |i| {
        sidebar_child.writeCell(sidebar_child.width - 1, i, border_char);
    }

    return sidebar_child;
}

but for the sidebar i pass y_off eql to the header_component.height

[stone tendon]

lemme try

[stone tendon]

worked perfectly

  win.writeCell(sidebar_component.width - 1, header_component.height - 1, intersection);

thanks !

[stone tendon]

is there a way to get the []u8 inside the TextInput widget without clearing the inner slice ?

[old skiff]

just input.buf.items?

[obsidian rover]

You can directly access input.buf.items and get the second half (it’s a gap buffer) from input.buf.secondHalf()

[stone tendon]

but it mentions that the pointers might get invalidated by various functions of the GapBuffer still fine to use ?

[stone tendon]

same way it s used in the toOwnedSlice i would say without the freeing

basically this is what i am trying to do, the input takes a search term, and it would show the result as the files that contains the term (col, line etc) i don t see the need to allocate each time i ll send the search term when clicking enter

don t mind the result just for debugging purposes 😂

[obsidian rover]

Yeah I agree

I almost wonder if this shouldn't change into a TextArea widget (implying it is for large text editing, and give it proper wrapping etc) and make a simpler TextInput widget meant for small inputs

That way it's easy to access the full input value without an extra buffer or allocator

[stone tendon]

Oh text input is the one on the left, the right side is just a segment, just wanted to see the data that i get from ripgrep process

OHH now i get what you mean nvm

I ll try that

[obsidian rover]

I meant the TextInput widget - I think the gap buffer is nice for larger text editing needs but small inputs like a search box, etc really don't need that

And a simple ArrayList can make it easier to use

[stone tendon]

btw any reason on why the offset in the segment is only in the first line ? or should i split them by new line and make a segment for each is that how the Segment is supposed to be used ?

[obsidian rover]

For right now though, you could get the length from input.buf.realLength, then allocate a buffer and memcpy input.buf.items into it, then memcpy input.buf.secondHalf() into it

[stone tendon]

ahhh so realLength always takes in consideration of the size of the secondHalf beforehand ?

[obsidian rover]

Ah - the offset is only for the first line so you can have multiple chunks of text with different styles but still printed as if a single body of text

Imagine a paragraph with a single bold word in the middle

It would need three Segments

If you have "different" chunks of text (bulletted lists, etc), I would print those with separate function calls

[stone tendon]

Ohh now i understand how Segments work, it s even shown in one of the examples as well i think in table example i should ve just payed attention

thanks !

[obsidian rover]

Think of a segment as a singular run of constant styled text within a larger body of text

[obsidian rover]

What are you making?

[stone tendon]

something similar to this

search and replace on multiple files

https://github.com/yassinebridi/serpl

[obsidian rover]

Oh cool

[stone tendon]

seems to be fun, and so far libvaxis seems very intuitive, specially how you can handle both vaxis events and your own custom events in the same event_loop

[obsidian rover]

Glad to hear!

[stone tendon]

any tips and tricks on how to debug ? the only way im debugging with so far is just start the process attach lldb to it and breakpoints

[obsidian rover]

I’m a printf debugger so I print statements, redirect stderr and tail -f the logfile in another terminal: zig build run 2> foo.log

[stone tendon]

OHHH that s neat thanks ! thanks

[ember robin]

Depends on exactly what issue you need to debug.

[stone tendon]

Yess ! is just that the formatting gets corrupted sometimes and what rock suggested fixed it, that s why i was using lldb and printing the variables in certain breakpoints

[stone tendon]

when rendering a window that has some segments/cells written, is there anything specific that has to be done in case you want that window to be empty again ? like saving a ref of that window and call clear or something ?

[obsidian rover]

You want it clear in the same render cycle or the next?

[stone tendon]

the next !

    pub fn add_result_view_component(
        result_view: *ResultView,
        parent: vaxis.Window,
        result: []const u8,
        opts: vaxis.Window.ChildOptions,
        allocator: std.mem.Allocator,
    ) !vaxis.Window {
        const child = parent.child(opts);
        log.debug("result: {s}, len: {d}\n", .{ result, result.len });
        if (result.len == 0) return child;

        var lines = std.mem.splitScalar(u8, result, '\n');

        var row_offset: usize = 1;
        var line_number: usize = 1;

        while (lines.next()) |line| {
            defer line_number += 1;

            const line_number_str = try std.fmt.allocPrint(allocator, "{d}: ", .{line_number});
            try result_view.line_numbers.append(allocator, line_number_str);

            print_line_number(&child, line_number_str, row_offset);

            row_offset += try print_line(
                &child,
                line,
                row_offset,
            );
        }

        return child;
    }

even though i am sure i am not printing those line numbers in when there is no result

[obsidian rover]

I usually clear every cycle and draw everything fresh

Clear the root, redraw everything

[stone tendon]

doesn t this clear the root ?

            const win = app.vx.window();
            win.clear();

[obsidian rover]

Yeah

[stone tendon]

i think something wrong on my side if that s the case, thanks for the help !

[obsidian rover]

It seems like there might be a desync happening somehow…like those characters are leftover due to an incorrect width calculation

What terminal, and are there possibly control bytes in what’s being printed?

[stone tendon]

im using wezterm ! and no i don t think so !

as it shouldn't go through that loop in the first place to print the line_numbers

you can see in the logs here

last render of result_view is called with a result that has len: 0

tried in iterm2 same thing, it will be just something dumb i have done, i ll keep you updated in case it is not !

[obsidian rover]

Hmm, yes something odd is happening!

[inner otter]

I'm currently rendering in the event loop, but sometimes I just want to signal a re-render from other sources without an input coming in. What is the best way to do that? I guess I need to separate the rendering and event loop? Is there a way to simply add a timer event as well?

[obsidian rover]

Look at the terminal widget example for a timer based render. Otherwise you can send in any event you want from whatever source.

Ie a redraw event with no payload that gets posted from some thread on a timer

[inner otter]

Ah I just add it to my own event union and somehow add it to the event loop?

[obsidian rover]

Yep

loop.postEvent

[inner otter]

ah and that is thread safe?

[obsidian rover]

Yes

[inner otter]

neat

[fleet schooner]

@obsidian rover How do I get colored undercurls? I set cell.style.ul to an rgb color and cell.style.ul_style to .curly, but nothing happens. The undercurls are always in the cell.style.fg color. I tried it on a bunch of terminals

[obsidian rover]

woops

Just pushed the fix for that, didn't test though but that has to be it

[fleet schooner]

Doesn't seem to make any difference. Does ul apply to any understyling? Or just underline?

[fleet schooner]

@obsidian rover I pushed it to flow anyway. Even though it doesn't seem to have an effect. So if you want something to test style.ul you can use flow.

[obsidian rover]

It should apply to any understyle

[obsidian rover]

Works for me?

[feral charm]

Is there a reason in particular for why you chose the MIT license for this project?

[obsidian rover]

I wanted a permissive license but didn't want any user of the library to be forced to GPL

[feral charm]

There are licenses like the GPL that don't force share-alike tho?

[obsidian rover]

Is there a problem with MIT though?

[feral charm]

I'm just intrigued by why people use it so much, even in projects where community engagement is supposed to prime over proprietary adoption.

It doesn't make sense to me that people just use it as a go-to license even when the project is community-oriented

[obsidian rover]

I do this for fun - the more people that use the library the more fun it is for me

[feral charm]

Do you care mostly about individuals or companies using it? :v

[obsidian rover]

Makes no difference to me

TBH if tigerbeetle made some TUI app with this I would be honored

[feral charm]

I mean I sort of get that

But don't you think that if you provide a free, open-source work to someone, they should give you the favor back?

[obsidian rover]

Not always, no. There are some things I do that I don't care and some that I do

For example - This is an investment in the terminal ecosystem. If someone can make a business off it somehow I'll still be pretty happy because I like to see the ecosystem move in a certain direction

[stone tendon]

providing for free, but when a lot businesses will use it, some features/bugs will have to be dealt with, meaning it will have to be maintained more frequently, meaning someone will have to get paid to do so

[stone tendon]

is there a way to higlight a part of a word or the whole line like on the sidebar ?

[obsidian rover]

Yeah there are a few ways to accomplish this

One way is to handle it via Segments:

window.print(&.{
  .{.text = "re"},
  .{.text = "main",
  .style = .{.reverse = true}},
  .{.text = "ing"},
});

You can do reverse, or set a bg color

Another is to do something like

var cell = window.readCell(row, col);
cell.style.bg = true;
window.writeCell(row, col, cell);

[stone tendon]

i can do this will writing that specific cell as well right ?

[obsidian rover]

Yeah

[stone tendon]

this is neat

thanks !

[fleet schooner]

what terminal is that?

[obsidian rover]

Ghostty

I tried both rgb and indexed underline colors and they both worked

[fleet schooner]

I figured it out. I was resetting the whole style on the next line.

[obsidian rover]

Still uncovered a vaxis bug though!

[stone tendon]

is this a working lsp in flow 👀

[stone tendon]

working good

[fleet schooner]

Yes it is. Flow has had LSP based goto definition and diagnostics support for a while now.

[stone tendon]

OHH i didn t know that the last time i tried it, neat !

[stone tendon]

im still facing the same issue of back then where i am having some already rendered cells not cleared after calling win.clear

allocated line numbers as a str

--------------------
len: 11
4: 
7: 
8: 
10: 
18: 
20: 
21: 
22: 
28: 
29: 
32: 
--------------------
len: 6
4: 
93: 
155: 
160: 
163: 
201: 

[stone tendon]

changing it from using a Cell to a Segment fixed it realized that i was using Cell in the wrong way i was passing []const u8 that was represented in more than one col which i guess Cell has to represent only one column ? which is the reason why it clears only one column in the screenshots above

[obsidian rover]

Yep, the value in cell should only be for one column

[inner otter]

What are good examples of how to use widgets?

Ah I found your comlink repo, that looks good 👍

[obsidian rover]

I was going to say that the examples directory has examples for each widget

But comlink uses a few as well

[inner otter]

I was trying to understand TextView, or basically any scrolling text widget with buffer. Didn't see an example for that sadly

[obsidian rover]

Check out @still hare ‘s spurdo editor

https://github.com/Cloudef/spurdo-editor

[inner otter]

Which Zig does it build with? 0.13 doesn't work for me

[obsidian rover]

Oh I’m not sure for spurdo. Libvaxis is pinned to 0.13 though

[inner otter]

Ah nevermind, I didn't update the submodules

Nice works well!

[dense turtle]

this looks great! i'm going to try it out for my next project :)

[dense turtle]

i'm loosely following the text input example, and i'm wondering what i'm doing wrong ... when i add this bottom line with the border, the program crashes with an integer overflow

[dense turtle]

after some poking around, i found out the reason for the crash: for some reason the parent window (which i got from doing vx.window()) has a width and a height of 0

i guess it seems like Screen never gets initialized with width and height values other than 0 ?

idk ;w;

[obsidian rover]

Are you initing a Loop?

When you receive a winsize event you have to pass it in to vx.resize. The resize function is what sets the screen size

It’s handled that way to prevent a lock on the screen which would be pretty expensive with how often it would need locking

[dense turtle]

oh oops, i forgot to add the winsize event to my loop

thats probably the issue, i test later

[dense turtle]

yep that was the issue, i stupid

thank you

[stone tendon]

is it normal for the program to crash if you don t call win.clear on each event iteration ? (win coming from vx.window() where vx is instance of Vaxis), want to see if there is any way to rerender only a child instead of the whole window

[obsidian rover]

No, that shouldn’t be happening. Do you have a stack trace?

[serene forge]

This library looks amazing. I think it would do wonders for the zig ecosystem to get some of this kind of zig tui work in front of a broader set of eyes with a site like https://terminaltrove.com/

I think zig is not yet registering with the wider cli community but I think it 100% should be

[obsidian rover]

I agree! I think flow is extremly polished and should definitely be put up on terminaltrove - as possibly the first zig project on there. @fleet schooner

[stone tendon]

yeah it turned out the way i was using vx.window() was wrong, is there any example that shows how to render individual child windows without rerendereing the whole thing ?

[obsidian rover]

No, I think all of the examples operate in immediate mode

[stone tendon]

but it s possible right ?

[obsidian rover]

Yeah - it should work just about the same

Do you have some code I could take a look at?

[stone tendon]

example on when the window have to resize all components should rerender

                .winsize => |ws| {
                    try app.vx.resize(app.allocator, app.tty.anyWriter(), ws);
                    win = app.vx.window();
                    win.clear();
                    header_view_component.update_win_component(&win, .{
                        .width = .{ .limit = win.width },
                        .height = .{ .limit = win.height * 3 / 100 },
                    });
                    sidebar_view_component.update_win_component(&win, .{
                        .width = .{ .limit = @min(50, win.width * 30 / 100) },
                        .height = .{ .limit = win.height - header_view.win.height },
                    });
                    search_input_view_component.update_win_component(&sidebar_view.win, .{
                        .width = .{ .limit = sidebar_view.win.width * 98 / 100 },
                        .height = .{ .limit = 3 },
                    });
                    result_view_component.update_win_component(&win, .{
                        .x_off = sidebar_view.win.width + 1,
                        .y_off = header_view.win.height,
                    });
                    selected_path_view_component.update_win_component(&win, .{
                        .y_off = 9,
                    });
                    // queue renders
                    try app.render_queue.writeItem(&header_view_component);
                    try app.render_queue.writeItem(&sidebar_view_component);
                    try app.render_queue.writeItem(&search_input_view_component);
                    try app.render_queue.writeItem(&result_view_component);
                    try app.render_queue.writeItem(&selected_path_view_component);
                    win.writeCell(sidebar_view.win.width - 1, header_view.win.height - 1, intersection);
                },```

and at the end of each iteration i rerender the components that need to rerender

...
            while (app.render_queue.readItem()) |component| {
                try component.render_component();
            }

            // Render the screen
            try app.vx.render(app.tty.anyWriter());

Component type:

pub const Component = struct {
    ptr: *anyopaque,

    render: *const fn (ptr: *anyopaque) anyerror!void,
    update_win: *const fn (ptr: *anyopaque, parent: *const vaxis.Window, opts: vaxis.Window.ChildOptions) void,

    pub fn render_component(component: *Component) !void {
        return component.render(component.ptr);
    }

    pub fn update_win_component(component: *Component, parent: *const vaxis.Window, opts: vaxis.Window.ChildOptions) void {
        return component.update_win(component.ptr, parent, opts);
    }
};```

update_win_component is ran when the window is resized

[obsidian rover]

                .winsize => |ws| {
                    try app.vx.resize(app.allocator, app.tty.anyWriter(), ws);
                    win = app.vx.window();
                    win.clear();
                    header_view_component.update_win_component(&win, .{
                        .width = .{ .limit = win.width },
                        .height = .{ .limit = win.height * 3 / 100 },
                    });

Does the component retain the reference to the window?

The window goes out of scope after .winsize block

[stone tendon]

oh yeah that s true is the reason why it segfautls, but the thing is, win width and height is set to 0 unless you get a new instance of Window from vx.window after calling vx.resize

[obsidian rover]

The windows hold a reference to the underlying screen, which is replaced when you call resize

pub fn resize(
    self: *Vaxis,
    alloc: std.mem.Allocator,
    tty: AnyWriter,
    winsize: Winsize,
) !void {
    log.debug("resizing screen: width={d} height={d}", .{ winsize.cols, winsize.rows });
    self.screen.deinit(alloc);
    self.screen = try Screen.init(alloc, winsize, &self.unicode);

So after resize, any previous window is no longer valid

You could reimplement Window in your application though, and instead of it holding a reference to *Screen, it could hold a reference to *Vaxis and then access vaxis.screen and always have an available screen

Window is really just a helper to make it easier to limit regions of printing to the Screen - nothing about it is internal to Vaxis. You could just copy the file and remove the stuff you aren't using

[stone tendon]

i should try that

i ll keep you updated !

[proud blade]

Once I finally rewrite zf (fuzzy finder) to use vaxis I’m considering submitting it to terminal trove… maybe 😅

[obsidian rover]

Oooh didn't know you were considering that!

[proud blade]

Although my current code works, it’s getting messy to add new features. I’ve been following vaxis from the day you first shared it and I started rewriting the UI of zf last week to use vaxis. It’s probably going to be a while before it’s done because I don’t have much free time these days

But I should say, thanks for libvaxis! It’s been really nice to use so far

[obsidian rover]

And thank you for zf, ziglua, great blog 🙂

[still hare]

the gap_buffer in build.zig.zon probably should be changed to point to a commit rather than the main branch

[obsidian rover]

Yeah. I was thinking about vendoring that one in and reducing it to the bare minimum that vaxis uses

In-band resize events for tracking purposes: kitty wezterm

[obsidian rover]

Sounds like Kovid is on board for implementing in Kitty

[obsidian rover]

Kitty has implemented!

[proud blade]

That was fast

[obsidian rover]

@proud blade When you are working on your zf rewrite, it would be cool if zf could be made easily into a widget. Basically have a couple easy ways to pass it events and draw to a window. Nearly everything I make has some usecase for a fuzzy finder

[proud blade]

I was working on zf this morning actually. This sounds like a great idea to me

Would that be weird for vaxis and zf to depend on each other?

[obsidian rover]

Haha - I was actually thinking of it as a third party widget...import zf and use whatever struct you provide as a widget

There are already a few applications using both libvaxis and zf

[proud blade]

That makes a lot more sense! I'll get the rewrite finished and then see about making it a widget

[proud blade]

Oh wow, that prompted me to do a search and I didn't realize how many people were using zf... feeling increased pressure to keep things working 😅

[obsidian rover]

Happy to help out with the effort, too!

[proud blade]

Thanks!

I actually do have a question on the zf vaxis rewrite. I don't want to use the alternate screen and instead draw the UI under the terminal prompt (like zf currently does). I'm not sure how I would best approach this though...

I'm just not finding a way to determine where the cursor is at the beginning so I can do all of my printing relative to that. Any thoughts on this?

[obsidian rover]

If you don't enter the alternate screen, it will print on the primary screen and all drawing is done with relative cursor motions instead of absolute motions like the alt screen

It's safe to call clear, but the screen will scroll to the furthest row printed that contains a non-default cell

My guess is you'll grab a vx.window(), clear it,, then create a child the exact size you want to limit all printing to ... win.child(.{.height = .{.limit = <h>}}). Print anything to this, then render

[proud blade]

I haven't tried making a child window yet... thanks!

Right now the relative printing works until I resize the terminal... then everything becomes relative to the top left of the terminal window, will a child window fix that?
Hmm it didn't...

[obsidian rover]

That's a good question

Currently on resize, we go to column 0 ('\r') then RI up the number of rows from where our cursor ended

This might conflict with what the shell prompt does

[proud blade]

Hmm I just tried in bash and zsh where I don't have any fancy prompt stuff setup and I still see the issue where resizing changes the relative drawing origin

[obsidian rover]

Yeah, I have some work to do on this part it looks like.

[proud blade]

🙏 thank you. Lmk if I can help with anything

[obsidian rover]

I won't be able to fix today, but I think it shouldn't be too bad. I need to add some linefeeds to prevent reflow from messing everything up too much and then change some logic in vx.resize

[proud blade]

No rush, I've got plenty of other projects to keep myself busy in the meantime

[obsidian rover]

Tagged a new release since we are at 0 issues and 0 PRs - thanks everyone for all your contributions and feedback! https://github.com/rockorager/libvaxis/releases/tag/v0.4.0

[obsidian rover]

I forgot to include a fix for @proud blade's resize-on-the-primary-screen issue. I've (mostly*) fixed that and tagged 0.4.1!

@proud blade It should work much better now. there is still some issues if you have a border. I can fix this by disabling autowrap in the terminal, but I'm not sure yet if I want to do that because it's pretty unexpected - since libvaxis requires library users to have a tty they can write to anyways people could do this on their own if it is absolutely needed

But I did a little demo of what i imagine zf will look like and the resize worked great now

[proud blade]

@obsidian rover thank you! I just tested really quick and it seems like everything works

[fleet schooner]

@obsidian rover is it possible to use the same zg as libvaxis? without adding it to my zon file

[obsidian rover]

Not currently - I will rexport it though

[fleet schooner]

is there perhaps a build system way to access sub dependencies of a dependency?

[obsidian rover]

Not that I know of :/

Since zg is broken up into modules, I will only export the modules vaxis loads (grapheme, code_point, and DisplayWidth)

Hmm

I think another way is that I can export the module from build.zig

[fleet schooner]

I guess I could just read the import_table field of the vaxis Module

[obsidian rover]

Oh yeah, that shoudl do it

[fleet schooner]

hmm, libvaxis doesn't actually use the module I need though. I guess I'll leave it as a local dependency then.

[obsidian rover]

Which module are you using?

[fleet schooner]

CaseData

[obsidian rover]

Hmm

Probably best to do a local dep then

[charred mountain]

idk if this is meant to be officially supported should be able to do something like b.dependency("libvaxis", .{}).builder.dependency("zg", .{})

[fleet schooner]

Perfect! That's exactly what I want, thanks. This prevents a duplicate download of zg, which is really slow. If it breaks in future I can always just add the local dep back, so I don't mind if it's not "official".

[uncut torrent]

hey, any chance this project just works with master? I saw that the target version is 0.13

[obsidian rover]

It doesn't work on master right now

[uncut torrent]

aight, thanks for the quick reply

[fleet schooner]

Are we going to get OSC 176 (set application ID) support in libvaxis too?

[obsidian rover]

I am for adding it. AFAIK, foot is the only terminal supporting it but it's a noop everywhere else

[obsidian rover]

I think that paste_allocator will need to be renamed allocator, and the library will hold onto the first value it receives from a query so it can restore it prior to exiting

[fleet schooner]

makes sense

[arctic cradle]

common foot win

[fleet schooner]

@obsidian rover I would have PRed a fix instead of opening #65, but I didn't have time and I didn't want to forget about it.

[inner otter]

How do I post my own event in the loop, postEvent?

[fleet schooner]

Yes, you should be able to use postEvent just like libvaxis does.

[inner otter]

Yeah I was putting a try before it and got the weirdest error message, but now it works!

[fleet schooner]

@obsidian rover Have you ever tried flow on windows? It looks pretty good, but there is a weird issue where cells are randomly rendered at half brightness. Have you seen this before?

It seems to be differential update related. Full renders are usually fine.

[obsidian rover]

It’s been awhile since I tried it on windows last. I wonder if there is some style reset not implemented by WT or maybe swallowed by conpty. I’ll check it out tomorrow

Is that whole line supposed to be red background?

Oh I see, it’s the imports too, hmm

[fleet schooner]

no, I have cursor line highlight off in this screenshot

its the imports

[fleet schooner]

It looks like I can query the default terminal background with queryColor(..., .bg), but not set it. I'd like to set it and have it reset automatically in Vasix.resetState. Shall I PR this? (I could also just do it in Flow directly)

[obsidian rover]

I would accept that PR!

@fleet schooner Have you seen XTPOPCOLORS / XTPUSHCOLORS? I'm thinking of implementing that at some point

kitty, xterm, and just recently foot added it

[fleet schooner]

Yeah, I saw it. I don't have a use case for it in flow though.

[feral charm]

How compliant is libvaxis with the X/Open Curses standard of the Single Unix Specification?

[obsidian rover]

I haven't heard of this - do you have a link?

[feral charm]

The SUS is like POSIX but with the X/Open Curses added for user interfaces

To be fair I also struggle to find copies of it :/

[obsidian rover]

Just briefly looking at it, I'd say "very little"

https://pubs.opengroup.org/onlinepubs/7908799/cursesix.html

[feral charm]

oh wow that was quick

[feral charm]

skill issue

[obsidian rover]

It actually has some similar concepts

Screen, windows, subwindows

[feral charm]

Well yes, you can't reinvent the wheel without making it round.

Does it matter tho

[obsidian rover]

But this is a very prescriptive specification regarding memory, function names, etc

And libvaxis does not comply with that

[feral charm]

want it to stay that way?

[obsidian rover]

Without knowing the benefit of compliance, yeah...it would be pretty simple to write a compliance wrapper it seems

But I don't know what compliance gains

[civic condor]

Hey, I’ve been lurking here for a while and following Vaxis development for a while. It looks like a fantastic library!

I’ve got a similar library for c++ code (although less battle tested and well rounded) and I’m trying to develop a way of managing splits across the terminal, something like how vim splits or tmux works. I wanted to ask how Vaxis handles splits or panes or windows within a terminal window. What approach did you take here?

[obsidian rover]

Vaxis uses the concept of "Windows". So when you draw a widget, you are bound to only draw within that window. If you had two panes of text, you'd create two child windows off the root window...each full height and half width. When the text widgets draw, they are bound to their window and the internal model manages all of the state of where each character is

[civic condor]

Is there an example on the repo that shows what the interface is like here? How does the window link to the text it’s meant to draw? Does it take a (might use CPP terminology here) pointer to a vector of strings and just draw that?

[obsidian rover]

The two main datastructures you'll want to look at are src/Screen.zig and src/InternalScreen.zig. Both of these have a "cell" structure for each cell on the terminal window. When drawing, a pointer to the grapheme for that particular cell is stored in Screen, the pointer must live until render is called. Before rendering, Any widget can overwrite any particular cell's value (this makes for efficient / easy drawing and clearing of areas). Once the drawing is complete, the Screen represents the state we want the terminal to look like so we call render. InternalScreen is the state of our last frame that we drew. So render compares each cell for equality (grapheme, style, urls, etc). If the cell is different, it updates the terminal. the state of the cell is copied into the InternalScreen structure. The copy is so that you can compare frame-to-frame

[civic condor]

That’s pretty clever. I guess you can then use the difference between the two for unit testing as well?

[obsidian rover]

unit what? 😂

[civic condor]

I’ll take more of a look later on - I was skimming through src/Window.zig because I thought to start there before, but thanks for pointing me in the right direction

[obsidian rover]

I don't have much testing in the diff code.

[wooden lantern]

Hey @obsidian rover, is there anyway to have the image loading done in the background? that way instead of freezing, users can still interact with the application. I know Zig doesn't have async so would I need to spawn a thread in the background to do this?

[obsidian rover]

There are a few new apis for images. The best performing would be to use transmitLocalImagepath. This will be much faster, but doesn't work over ssh (and currently silently fails). Next would be to spawn a thread which loads the image, converts the pixels to an allowable format, and base 64 encodes the image. You can then call transmitPreEncodedImage from the main thread which will chunk up the data and send it. This should be quite a bit faster than current, but you should always send on the main thread since the tty writer is not thread safe

For the thread spawning one, you can look at the code in transmitImage - basically do that function in a thread, but then call transmitPreEncodedImage in the main thread

[wooden lantern]

oh awesome! I'll make sure to update libvaxis then. thanks

[obsidian rover]

The next option is to do all the sending yourself - you could chunk, send a chunk, check for input events, then repeat loop and send another chunk.

[wooden lantern]

right, that's also possible

thanks, i'll mess around with these options and see what works best :)

by "doesn't work over ssh", will it crash/error or simply not load? If I can catch on error, then I can just display an error over ssh

[obsidian rover]

Not load, but you won’t get an error. The terminal sends an error as an escape code by libvaxis doesn’t check or parse these currently

It’d be delivered async

[wooden lantern]

hmm, okay. i'll investigate into these options then. thanks man :)

[obsidian rover]

No problem!

[proud blade]

@obsidian rover it's been a while, but I'm making slow and steady progress with rewriting zf to use libvaxis! Got a nice preview pane working today. It is much easier to add new features now. Thanks for vaxis!

[obsidian rover]

Looks awesome!

[wooden lantern]

Hey @obsidian rover, is there a way to force a flush in instances like this where I need to free memory after writing?

[obsidian rover]

You should be able to just call render…or am I missing something?

[wooden lantern]

oh... I think i'm the stupid one lol

i'm calling render after the memory is freed

😭 sorry man, I need a rubber duck or something aye

[obsidian rover]

No worries! Happy hacking

[obsidian rover]

I'm doing some gardening in libvaxis. Lots of breaking changes are currently on the unreleased HEAD, but it will all be worth it 🙂

One of the major changes coming is the introduction of a TUI framework. The framework (named vxfw..."Vaxis framework") is by no means mandatory, and is basically just built on top of the Vaxis core that is out there today. There is a draft PR up https://github.com/rockorager/libvaxis/pull/105. Would appreciate any commentary if people have thoughts! Overall the idea is for vxfw to be the high level side of libvaxis. For those familiar with Rust, imagine vxfw as Ratatui and libvaxis as crossterm. Or in Go...vxfw is tview and libvaxis is tcell. Or something to that effect.

As a quick example, here is a very basic fuzzy finder implemented in 235 LOC with full unicode and full mouse support. This one uses Text, RichText, ListView, and TextField.

[wooden lantern]

I was a big fan of Ratatui and used it back in my Rust days. Very much looking forward to this :)

[fierce pond]

https://github.com/rockorager/libvaxis/pull/110

I made a PR for some unchecked errors I was running into

[frigid sundial]

So meaning we will be able to provide our own backends? eg: rendering TUI in web browser.

[wooden lantern]

Hey @obsidian rover, do you know of any libraries that render md content that work with libvaxis?

[obsidian rover]

No, I'm not aware of any

[obsidian rover]

Would be a good project though 😉

[wooden lantern]

haha, totally.

do you know in terms of the image library if there's a way to see what formats are currently supported? e.g. for my file explorer, i would love to know if the file i'm about to pass to the library is even an image type that it supports

currently i'm just passing it any file the user selects and seeing if it spits an error or not lol, not ideal

like, i can manually check the libraries readme and hardcode these,,, but ideally there would be something that updates automatically for this

[obsidian rover]

I’m not sure - I’d have to look through their docs, my guess would be to do what you are doing … zigimg has pretty good errors so you could probably do something by switching on the specific error you get

I think that table is out of date

JPEG is definitely supported

[wooden lantern]

only problem is sometimes things just die...

for example, this error occurs when you attempt to pass a .ico file into the function

[obsidian rover]

Hmm, that one seems like a zigimg bug, not an issue with unsupported formats

[wooden lantern]

jpg also seems unsupported?

[wooden lantern]

hmm, okay. i'll create an issue on their end then

[obsidian rover]

I wonder if it has to jpeg https://github.com/zigimg/zigimg/blob/master/src/formats/jpeg.zig

[wooden lantern]

I can do something like so which seems to fix a lot of the issues when passing in dud files

var match = false;
inline for (@typeInfo(vaxis.zigimg.Image.Format).Enum.fields) |field| {
    const entry_ext = std.mem.trimLeft(u8, std.fs.path.extension(entry.name), ".");
    if (std.mem.eql(u8, entry_ext, field.name)) {
        match = true;
    }
}
if (!match) break :unsupported;

[wooden lantern]

still unsure as to why jpg isn't working.. i'll try look into it and pass through a jpeg image and see if that helps

[wooden lantern]

Hey @obsidian rover, I was wondering if you had any further thoughts on issue https://github.com/rockorager/libvaxis/issues/79 in regards to sixel images and the latest two comments

• https://github.com/rockorager/libvaxis/issues/79#issuecomment-2438616652 : sixel performance
• https://github.com/rockorager/libvaxis/issues/79#issuecomment-2496185509 : supporting chafa instead

[obsidian rover]

I'm actually eyeing up this zig sixel implementation: https://github.com/daredemo/zixel

[wooden lantern]

that would be awesome, looks promising. excited to follow the development :)

[obsidian rover]

Hey everyone! So I've decided to use Github Discussions as a more general purpose support / Q&A place. I'll still respond to questions here but just wanted people to know that there is that available as well (and is a bit easier to follow than a giant discord thread).

Also there is a Show and Tell section, please post whatever you have built or are building with libvaxis there, it's always fun to see what people are working on!

[terse juniper]

Are there any examples and screenshots of what can be built using this library? Really interested in taking a further look.

I'm mostly interested in building some kind of basic dashboard with separate table sections

[obsidian rover]

There are some minimal examples in the repo (specifically look at table). But also flow, zf, zfe, todui, and comlink are all built with libvaxis

[terse juniper]

Thank you

[fierce pond]

bit late, but you can also take a look at fancy-cat

[fierce pond]

pinging here in case you didn't see my question in the Github Discussions tab (my apologies if u did)

[obsidian rover]

Oh I saw it and forgot about it! Ill respond on github

[fierce pond]

No worries! Thank you

[wooden lantern]

Hey @obsidian rover, does Libvaxis support all of these image formats? https://github.com/zigimg/zigimg?tab=readme-ov-file#supported-image-formats

I noticed GIF is technically supported but won't actually render. Is there a way to know which image types will actually be rendered to the terminal?

e.g. I'm using this to do a simple check to see if zig-img supports the image but this doesn't seem to mean libvaxis supports the rendering

var match = false;
inline for (@typeInfo(vaxis.zigimg.Image.Format).@"enum".fields) |field| {
    const entry_ext = std.mem.trimLeft(
        u8,
        std.fs.path.extension(entry.name),
        ".",
    );
    if (std.mem.eql(u8, entry_ext, field.name)) {
        match = true;
    }
}
if (!match) break :unsupported;

[obsidian rover]

I’ll have to look at how I’m using it. I think it’s using zigimg to decode and then encoding it as a png to upload to the terminal - so I would expect them all to be supported. I’ll take a look tomorrow

[wooden lantern]

sounds good, thanks for that :)

here's where i'm using it if it helps you debug in any way https://github.com/brookjeynes/jido/blob/main/src/drawer.zig#L143

[obsidian rover]

It should be working with anything supported by zigimg. When you load the file, zigimg decodes the image to pixels - the vaxis transmit function will then convert those to either png, rgb, or rgba.

I'm not sure exactly how this interacts with gifs because those are a sort of container of separate frames

Maybe we only get the first frame of a gif?

[wooden lantern]

getting the first frame of a gif would be fine. it just seems that it wouldn't load the image at all but maybe it's my usage of the library

[wooden lantern]

just did some further testing, seems gifs are working correctly with Kitty. I failed to realise my friend who reported the issue was using Iterm. sorry about that

[candid rock]

For allocs during draw I have an arena allocator i init before I call draw and I deinit after it is rendered. What is the standard way of doing this?

[still hare]

Do not deinit, use the reset function with retain_capacity

You want to reuse the same arena between draws

[candid rock]

Cool, thanks

[flint gale]

@obsidian rover does the mouse scroll event listner work ? couldn't find any example. as well

[obsidian rover]

Yep, if you enable the mouse you’ll get mouse wheel events.

[flint gale]

do you have any example code there. I couldn't find any documentation or anything for it so looking at example code is all what i can do.

[obsidian rover]

I don't see an explicit example...here is how you can handle these events (they are delivered as a .mouse event)

To enable mouse, you to call vx.setMouseMode(writer, true)

[tranquil pasture]

Hello !
I have a background thread working, and I'd like to display data it generates.
With the "low level API" it's easily done, as I have access the the loop and its postEventthread-safe function.
Yesterday, I played around using the vxfwhigh-level API. While its usage is simpler, and gives access to more widgets, it seems it has no capability to inject user events.
Yes, I can eventually update the model from the thread with atomics, and continously redraw instead of waiting for events, but still, I'm wondering if I missed something.
tl;dr -> Does vxfwallows user events in its loop ? If yes, how ?

[obsidian rover]

I don't have a way to plumb that through nicely yet. I've held off because I haven't come up with a nice design for it. I am trying to keep the backend app runner completely decoupled from the widgets (IE a widget should never depend on what the event loop is)

So...in some way we need to asynchronously wake up the loop with a posted event, or maybe a wakeup that the app has events available (in it's own queue, for example).

I get around this in comlink currently by having a timer callback and I check for events. Not event based, which is icky

[tranquil pasture]

Hi,
Thank you very much for your detailed reply.
No worries, I understand the situation and I’ve already a workaround (continuously redrawing)
Thanks again !

[obsidian rover]

If you have any ideas on how you would like the API to look for injecting custom events or waking the loop, I am all ears!

[tranquil pasture]

Sure ! I think there is a quick easy fix first: the possibility to simply ask for a redrawing.
So a thread update some atomics and inform the app to trigger a redraw.
It’s not as clean as user events but get the job done without constantly redrawing.

For instance, rust cursive has a noop event

I admit it’s not clean nor sustainable on big projects

[obsidian rover]

Yeah I was basically thinking of a noop event / wakeup to tell the app to wakeup and deliver a noop to some widget

For that route, I think it would be delivered to the widgets as an object on the EventContext:

pub const Wakeup = struct {
    ptr: ?*anyopaque,
    wakeupFn: *const fn (?*anyopaque, Widget) void,

    pub fn wakeup(self: Wakeup, widget: Widget) {
        self.wakeupFn(self.ptr, widget);
    }
}

Something like that

Where ptr is the event loop implementation

All the pointers would be stable, so this object could be held onto by the application to call from whatever thread it needs

[tranquil pasture]

Also, I know we don’t have access to the loop, and it’s on purpose. But maybe we can try to find a way to configure the loop Event enum at comptime and expose a postEvent public function ?

But that will make things more complex for only a special case (external threads)

Another approach: do nothing and wait for async back so we can select()

[obsidian rover]

The problem with this is plumbing that event enum through to every widget. And ideally widgets will not all be within the libvaxis repo, but people can make their own widget libraries as well

[tranquil pasture]

Oh ok I understand. That was my main approach, but I didn't really understand the context.

Regarding my last sentence, I know it's confusing, but maybe it's too early to solve this. Workarounds exist, and async could solve this problem (for instance, Rust Ratatui can use Tokyo's select!)

[wooden lantern]

Hey @obsidian rover, am i misunderstanding how the postEvent function works?

I have some code which spawns up a thread to do some background work, then posts an event if all goes well

{
    const load_img_thread = std.Thread.spawn(.{}, loadImage, .{
        app,
        self.current_item_path,
    }) catch break :unsupported;
    load_img_thread.detach();
}

fn loadImage(
    app: *App,
    path: []const u8,
) error{ Unsupported, OutOfMemory }!void {
    var image = vaxis.zigimg.Image.fromFilePath(
        app.alloc,
        path,
    ) catch {
        return error.Unsupported;
    };
    defer image.deinit();

    if (app.vx.transmitImage(app.alloc, app.tty.anyWriter(), &image, .rgba)) |img| {
        app.image.data = img;
        if (app.image.path) |p| app.alloc.free(p);
        app.image.path = try app.alloc.dupe(u8, path);
        app.loop.postEvent(.{ .image_ready = app.image.path orelse "test" });
    } else |_| {
        if (app.image.data) |img| {
            app.vx.freeImage(app.tty.anyWriter(), img.id);
        }
        app.image = .{};
        return error.Unsupported;
    }
}

This is what my polling looks like

self.loop.pollEvent();
while (self.loop.tryEvent()) |event| {
    switch (event) {
        .key_press => {
            // ...
        },
        .image_ready => {
            std.debug.print("\nooooopppppps\n", .{});
        },
        else => {},
    }
}

The problem is, it seems that i have to key press a couple times for the event to be posted, i.e. the image to be shown

EDIT: disregard, this is working correctly. I simply had an incorrect if statement surrounding the logic

[wooden lantern]

Hey @obsidian rover, any clue why this is breaking images on subsequent loads from "cache" (hashmap)?
https://github.com/BrookJeynes/jido/compare/main...bj/2025-05-23/feat/cache-images

[obsidian rover]

what’s the panic that’s happening? My guess is there is some index reuse…vaxis will assign an id and maybe those are getting corrupted?

[wooden lantern]

Here's the log

terminal messy but here's another log from what i presume to be the same panic (did the same actions as above)

[obsidian rover]

Doesn't seem obvious, do you have that in a branch I can test locally?

[wooden lantern]

yep - https://github.com/BrookJeynes/jido/compare/main...bj/2025-05-23/feat/cache-images

[obsidian rover]

This is the panic I am getting...invalid free

I think what is happening is that you are caching the data of the file. On first transmitImage, vaxis calls zigimg.convert to get it to the right pixel format. zigimg frees the old pixel buffer. Then, the next round you go to transmit and it's bogus data and an invalid free when you try to convert it

What I woudl do to make it all better - you don't need to cache the image data, the terminal does that for you. Once you have transmitted an image once you can show it by referencing it's ID (which in vaxis terms, just means draw it)

So on first load of image - transmit it and cache the vaxis.Image object (which is just an id, width, and height)

Subsequent draws will be very fast since they won't do any conversion or transmission

[wooden lantern]

when does the terminal clear this "cached image"

is something like this to do "safe" with the terminal caching the image data (the data i fetch via the id). this seems to be working fine from my testing however
https://github.com/BrookJeynes/jido/compare/main...bj/2025-05-23/feat/cache-images#diff-face6d58eec5a8a75042e86c55da013110761229f0824c7fdd2156e1f63e2012R203

[obsidian rover]

Either if you explicitly tell it to (this would be image.destroy in vaxis) or I believe most do a LRU cache for them

Within your app it should be safe

There is a query to see if an id is still in mem (I haven’t implemented it)

[wooden lantern]

right, makes sense. I've handled all possible errors i can on my end so i'll jsut hope its enough :) thanks for the help

[wooden lantern]

Hey @obsidian rover, do you know why when running something like cd $(jido), the alt screens seem to break completely?
https://github.com/brookjeynes/jido/blob/main/src/events.zig#L421 here i'm handling opening up $EDITOR on a file, and jumping back when that editor closes.
doing something like this in cd $(jido) seems to just present me with a blank screen that still accepts input but doesn't show anything

[obsidian rover]

I'm not sure I get it. I tried locally and it seems to behave the same if I do cd $(jido) or just jido

Your gif looks like a blank screen momentarlily and then it loads the dir?

[wooden lantern]

ah sorry, maybe the gif wasn't obvious with what was happening

when i attempt to open a file, which enters altscreen and spawns $EDITOR (in my instance neovim), the screen remains blank but still accepts keyboard input

this only occurs when running jido in something like cd $(jido)

i'm wondering if this is something i'm handling incorrect with my alt screen usage or if this is something on libvaxis end

@obsidian rover here is what openFile is doing after the alt screen is left https://github.com/BrookJeynes/jido/blob/bedb34417b1bc96ad5c0767ef822c0f305b2f778/src/environment.zig#L51

[obsidian rover]

So you did a suspend and then call that?

[wooden lantern]

https://github.com/brookjeynes/jido/blob/main/src/events.zig#L421

That's the flow there

[obsidian rover]

I wonder if somehow the subshell is messing with it

[wooden lantern]

Hmm, interesting. Is this something I can fix on my end or a libvaxis implementation

[obsidian rover]

I'm not sure - what editor are you using?

[wooden lantern]

neovim, were you able to repro?

[obsidian rover]

Yeah I can repro

Hm

EDITOR=flow works fine

EDITOR=helix doesn't, neither does nvim

Which makes me think its a subshell thing? I know for sure that flow opens /dev/tty...helix and nvim might be checking stdin/stdout/stderr and the subshell messes with it? I don't think that should be happening though

I'll keep thinking about it. On the mend from covid and a bit brain foggy today

[wooden lantern]

that's okay, take your time! I hope you recover well and quickly ~

thanks for helping out on this one

[wooden lantern]

how have you been feeling lately? hope you're recovering well. just wondering if there were any further developments into this?

[main gate]

Hi, is there any up-to-date tutorial or examples for using libvaxis with Zig 0.14.1 in order to get started with the library? The libvaxis-starter repository example seems to be broken with the latest libvaxis master version (I guess tagged releases don't work with Zig 0.14.1 as they are ancient?)

[fierce pond]

https://github.com/rockorager/libvaxis-starter/pull/4

updated the libvaxis starter for zig version 0.15.1, hopefully it helps

[fierce pond]

https://github.com/rockorager/libvaxis/pull/260

Started adding tmux support for kitty images, would appreciate help from someone more familiar with kitty unicode placement

[wooden lantern]

Hey @obsidian rover, is rendering libvaxis via a buffered writer no longer an option in 0.15.0?

[obsidian rover]

It is, it accepts the new std.io.Writer which has buffering built in

[wooden lantern]

oh neat, so no need to do this

var buffered = self.tty.bufferedWriter();
try self.vx.render(buffered.writer().any());
try buffered.flush();

just this now?

try vx.render(tty.writer());

[wooden lantern]

hmm,, I seem to be getting some not so great errors when trying to render an image after upgrading to 0.15.1. any help would be appreciated 🙏
https://github.com/brookjeynes/jido/blob/0b1d69cded56dc2677993d9b62461650b6de4fc8/src/drawer.zig#L650

[willow river]

You can run your program as your_program 2>your_program.err to write the errors into a file so it's more readable and easier to debug.

[wooden lantern]

So true 🤦‍♂️ I'll do that when I get to my laptop next.

[wooden lantern]

here you go, hope that's a bit easier to see

[restive grove]

i was working on a TUI library as well https://github.com/adxdits/zigtui

[wooden lantern]

Hey @obsidian rover, is there a reason we now need a buffer when reading image from path? Is there a way for this to be allocated instead of a hard-coded buffer size like in the example? I am currently caching loaded images for quick access later but I'm now needing to keep a new buffer per image alive?

fn loadImage(app: *App, path: []const u8, buffer: *[1024]u8) error{ Unsupported, OutOfMemory }!void {
    _ = buffer;
    var buf: [1024 * 1024]u8 = undefined;
    const data = vaxis.zigimg.Image.fromFilePath(app.alloc, path, &buf) catch {
        const message = try std.fmt.allocPrint(app.alloc, "Failed to load image '{s}' - error occurred while attempting to read image from path.", .{path});
        defer app.alloc.free(message);
        app.notification.write(message, .err) catch {};
        if (app.file_logger) |file_logger| file_logger.write(message, .err) catch {};
        return error.Unsupported;
    };

    app.images.mutex.lock();
    if (app.images.cache.getPtr(path)) |entry| {
        entry.status = .ready;
        entry.data = data;
    } else {
        const message = try std.fmt.allocPrint(app.alloc, "Failed to load image '{s}' - error occurred while attempting to add image to cache.", .{path});
        defer app.alloc.free(message);
        app.notification.write(message, .err) catch {};
        if (app.file_logger) |file_logger| file_logger.write(message, .err) catch {};
        return error.Unsupported;
    }
    app.images.mutex.unlock();

    app.loop.postEvent(.image_ready);
}

[wooden lantern]

I believe there to be an issue with the deallocation of images, im unsure what i'm doing wrong unfortunately - https://github.com/brookjeynes/jido/blob/0b1d69cded56dc2677993d9b62461650b6de4fc8/src/app.zig#L95

Segmentation fault at address 0x7f4c8e8f1010
/home/bjeyn/.zvm/0.15.1/lib/std/mem/Allocator.zig:428:26: 0x13d7168 in free__anon_58795 (std.zig)
    @memset(non_const_ptr[0..bytes_len], undefined);
                         ^
/home/bjeyn/.cache/zig/p/zigimg-0.1.0-8_eo2vUZFgAAtN1c6dAO5DdqL0d4cEWHtn6iR5ucZJti/src/color.zig:1030:45: 0x13663d1 in deinit (zigimg.zig)
            .rgba32 => |data| allocator.free(data),
                                            ^
/home/bjeyn/.cache/zig/p/zigimg-0.1.0-8_eo2vUZFgAAtN1c6dAO5DdqL0d4cEWHtn6iR5ucZJti/src/Image.zig:137:23: 0x12e1ec3 in deinit (zigimg.zig)
    self.pixels.deinit(allocator);
                      ^
/home/bjeyn/projects/jido/src/app.zig:103:21: 0x127b557 in deinit (main.zig)
            d.deinit(alloc);
                    ^
/home/bjeyn/projects/jido/src/app.zig:207:29: 0x1226561 in deinit (main.zig)
        img.value_ptr.deinit(self.alloc, self.vx, &self.tty);
                            ^
/home/bjeyn/projects/jido/src/main.zig:116:25: 0x12189bb in main (main.zig)
        defer app.deinit();
                        ^
/home/bjeyn/.zvm/0.15.1/lib/std/start.zig:627:37: 0x1219db9 in posixCallMainAndExit (std.zig)
            const result = root.main() catch |err| {
                                    ^
/home/bjeyn/.zvm/0.15.1/lib/std/start.zig:232:5: 0x11c32f1 in _start (std.zig)
    asm volatile (switch (native_arch) {
    ^
???:?:?: 0x0 in ??? (???)

[wooden lantern]

Seems this was actually an issue from within Libvaxis - PR here https://github.com/rockorager/libvaxis/pull/293

Hey rockorager, in what instances would cursorLeft() and cursorRight() fail to work? Running the textinput example, left and right works but in the usage within my app it seems it fails to actually move the cursor left and right.

I'm a little confused as to why as i'm not doing anything particularly different to that example