A stack of stacks is a more natural representation of what SubLayouts
does: it packs information about the global focus as well as focus in
individual groups (sublayouts).
It doesn't carry information about the sublayouts themselves (but a
similar structure in X.L.Groups does), so we still use Groups and
fromGroups in some places, but future refactor can simplify that as
well, I'm sure.
My main motivation for this is that I need to expose the window groups
to the user config, and a stack of stacks seems to be a nice data
structure for that. The motivation for exposing the groups is that I
want to manipulate focus in a way that takes groups into account. As an
example, I want the following:
* mod-1, mod-2 to mod-0 switches to n-th group if not already focused,
and if focused, focus next in the group
* show these numbers and window titles in xmobar (like tmux/screen/vim
status line), like so:
1a weechat 1b browser 2 vim 3 mutt
Achieving this just using BoringWindows is quite tricky, but with the
ability to somehow (InspectLayout, which is work-in-progress, or message
with IORef) get the stack of stacks out of SubLayouts, this becomes
easy.
Turns out similar logic is already in updateGroup, and we don't even
need to worry about extra/missing windows and we can just differentiate
the result of reordering.
This makes the following sequence of operations idempotent, as it should be:
windows $ W.float w $ W.RationalRect 0 0 1 1
windows $ W.sink w
Previously, any window not visible to the SubLayout modifier (xmonad
invokes runLayout with tiled windows only) would be reordered to the end
of stack. This commit changes the reordering logic to only reorder
windows in Groups and keep all other windows where they were.
- We would really like a full template and not just a snippet where
we'll have to guess the context.
- It's generally nice to know which versions of xmonad and contrib a
users was using when encountering an issue.
We're not parsing anything (as opposed to the respective `P` functions)
and so there's no need to create a dummy map with units as values and
then take the difference; we can simply remove the relevant keys from
the map.
A new prompt that works like 'safePrompt', but is optimized for the
use-case of a program that needs a file as an argument.
This is necessarily a new function and can't just be achieved by using
the old `safePrompt`, as `getShellCompl'` does not at all filter the
files (compgen already does that based on the input), but only the
available commands. If we start the prompt with a single command then
the chosen `searchPredicate` becomes quite useless and we can't take
advantage of fuzzy matching for file finding. This, however, is quite
useful when having a program that explicitly expects a file as one of
its arguments, e.g. dragon [1].
What we have to do instead of to generate all available files with
compgen and _then_ filter this down to what we want via a given
function. In order to make this change backwards compatible we have to
introduce the rather ugly `shellComplImpl`, which takes a laundry list
of all of the different parameters that we need. Since the function is
not exported, this ugliness does perhaps not matter too much.
[1]: https://github.com/mwh/dragon
When `alwaysHighlight` is enabled and one immediately presses (by
default) Return after opening the prompt (because the highlighted
completion is what one wants) then the selection will not enter the
prompt history. Instead, an empty string will be entered because there
hasn't been any input yet and hence the `highlightedCompl` field has not
yet been filled in.
The fix is simply checking whether `alwaysHighlight` is set during
startup and, if yes, already setting the highlighted completion to the
first suggestion.
This builds upon the idea of 780360abf0
but it fixes the issue in a bit of a nicer way; instead of throwing an
exception and closing the prompt, we simply check if there is only a
single completion available inside `hlComplete` directly.
While what 780360abf0 said is true insofar
that we have incomplete information when inside `hlComplete`, this does
not actually matter in this case. We have access to the complete user
input (possibly consisting of things not from this round of completion),
which is enough here.
