freedesktop_entry_parser is needed to parse .desktop entries which it doesn't seem ironbar was doing until now. There's one other rust library that can do that but this one seemed lighter-weight and more stable.

unicode-segmentation is needed for truncating menu entries which is really needed because some common .desktop files (like the JDK) have really long names. As far as I can tell there's no actually safe way to truncate unicode strings in the Rust stdlib.

There is .desktop file parsing happening in the desktop_file module already. Did you not spot that, or does that not do what is necessary here?

In some other modules (music, focused) these are already using a (partially) common truncate option that hooks into GTK's in-built truncation/elipsize system. I'd prefer to use that here, although I'm open to integrating unicode-segmentation and having a words mode there.

I'm not actually sure this is correct, the way ironbar handles TOML is a bit confusing to me.

Ironbar is just using the standard toml crate, it's TOML that's confusing - there's two syntaxes for objects, and two for arrays.

Generally I write the corn config and then use its CLI to generate the other three formats, which isn't necessarily the most readable but at least ensures they're valid.

I think here, you actually want something like (shortened and untested):

[[start]]
type = "menu"

[[start.start]]
type = "custom"

This is the default menu that xfce has except that I picked the icons myself since I couldn't figure out how xfce does that (I think it might have a series of fallbacks due to lack of standardization in icon themes).

I looked through a number of icon themes and tried to identify icons that were available in the majority of them.

I'll road test this myself just to double-check before merging, but I'm sure that'll be fine. I need to rework some of the icon stuff soon anyway.

My code organization is pretty sub-par here, I've been struggling with the fact that refactoring into functions is not free in Rust since it introduces a borrow. I thought about just writing a bunch of macros but I don't know if that's idiomatic.

I do sometimes stick a macro in a function to deal with repeat work, although I'm not a tremendous fan. My biggest issue is that they result in a performance hit to the language server when editing (I do need to refactor some of the existing ones out into inline functions really).

An alternative is to use closures, although I'm not sure if they incur any runtime performance penalty.

You can also actually use .iter().map().collect() to collect into a HashMap (or IndexMap etc) directly, so long as you try to collect a tuple of (K, V).

Data flow is a little clunky because I was struggling to write code that avoided cloning.

You might be able to refactor to use into_iter instead of iter, which will give you the owned object instead of a reference. That'll allow you to move it.

That said, I don't think this is the end of the world if you do need to clone a few strings here as it's only init code.

For GTK objects I've just been spamming .clone() until the borrow-checker stops complaining which seems to be the approach used elsewhere in ironbar. My understanding is the these are just pointers into objects managed by GTK so this doesn't really cost anything but then I don't understand why they don't implement Copy.

It's a weird internal quirk of gtk-rs, they've got some pretty nuts macros going on, but it actually is Copy sort of.

Each widget is basically a struct made of two pointers. The struct does not implement Copy but each of the pointers do. This means the Clone impl in reality does just create a new struct and Copy the two fields.

As to why they don't implement it on the exposed structs, I've no idea. I guess it's more explicit?

This is all copied from popup.rs but I didn't see any way to hook into that existing leave notify event. I could do some refactoring here but at least in my initial PR I was trying to keep changes isolated to the new module.

I see, that's an annoying problem. You might be able to listen to the unmap event to get round that, which (in theory) will run when the window is hidden.

Reminder for myself to add an image

(no need to action anything)

If you set this, does it append to or replace the XDG entries?

It will replace the default entries. It might be worthwhile to list the default entries in the documentation in case you want some of them but not others.