The list below consist mostly of ideas noted down before they are requested/discussed by users (at which point they usually exist on the github issue tracker).
It's mostly a bunch of personal notes, probably incomplete. Feel free to query if you have any questions.
- window: preserve/restore relative focus ordering (persistent or not), and e.g. of multiple reappearing windows (#2304) -> also see docking reference to same #.
- window: calling SetNextWindowSize() every frame with <= 0 doesn't do anything, may be useful to allow (particularly when used for a single axis). (#690)
- window: add a way for very transient windows (non-saved, temporary overlay over hundreds of objects) to "clean" up from the global window list. perhaps a lightweight explicit cleanup pass.
- window: using SetWindowPos() inside Begin() and moving the window with the mouse reacts a very ugly glitch. We should just defer the SetWindowPos() call.
- window/size: how to allow to e.g. auto-size vertically to fit contents, but be horizontally resizable? Assuming SetNextWindowSize() is modified to treat -1.0f on each axis as "keep as-is" (would be good but might break erroneous code): Problem is UpdateWindowManualResize() and lots of code treat (window->AutoFitFramesX > 0 || window->AutoFitFramesY > 0) together.
- window/opt: freeze window flag: if not focused/hovered, return false, render with previous ImDrawList. and/or reduce refresh rate. -> this may require enforcing that it is illegal to submit contents if Begin returns false.
- drawdata: make it easy to deep-copy (or swap?) a full ImDrawData so user can easily save that data if they use threaded rendering. (#1860 see ImDrawDataSnapshot)
! drawlist: add CalcTextSize() func to facilitate consistent code from user pov (currently need to use ImGui or ImFont alternatives!)
- drawlist: maintaining bounding box per command would allow to merge draw command when clipping isn't relied on (typical non-scrolling window or non-overflowing column would merge with previous command). (WIP branch)
- widgets: add always-allow-overlap mode. This should perhaps be the default? one problem is that highlight after mouse-wheel scrolling gets deferred, makes scrolling more flickery.
- widgets: group/scalarn functions: expose more per-component information. e.g. store NextItemData.ComponentIdx set by scalarn function, groups can expose them back somehow.
- input text: support for INSERT key to toggle overwrite mode. currently disabled because stb_textedit behavior is unsatisfactory on multi-line. (#2863)
- input text: add discard flag (e.g. ImGuiInputTextFlags_DiscardActiveBuffer) or make it easier to clear active focus for text replacement during edition (#725)
- input text: display bug when clicking a drag/slider after an input text in a different window has all-selected text (order dependent). actually a very old bug but no one appears to have noticed it.
- input text: decorrelate display layout from inputs with custom template - e.g. what's the easiest way to implement a nice IP/Mac address input editor?
- input text multi-line: don't directly call AddText() which does an unnecessary vertex reserve for character count prior to clipping. and/or more line-based clipping to AddText(). and/or reorganize TextUnformatted/RenderText for more efficiency for large text (e.g TextUnformatted could clip and log separately, etc).
- input text multi-line: line numbers? status bar? (follow up on #200)
- input text multi-line: behave better when user changes input buffer while editing is active (even though it is illegal behavior). namely, the change of buffer can create a scrollbar glitch (#725)
- input text multi-line: better horizontal scrolling support (#383, #1224)
- layout: helper or a way to express ImGui::SameLine(ImGui::GetCursorStartPos().x + ImGui::CalcItemWidth() + ImGui::GetStyle().ItemInnerSpacing.x); in a simpler manner.
- layout, font: horizontal tab support, A) text mode: forward only tabs (e.g. every 4 characters/N pixels from pos x1), B) manual mode: explicit tab stops acting as mini columns, no clipping (for menu items, many kind of uses, also vaguely relate to #267, #395)
- layout: null layout mode were items are not rendered but user can query GetItemRectMin()/Max/Size.
- layout: (R&D) local multi-pass layout mode.
- layout: (R&D) bind authored layout data (created by an off-line tool), items fetch their pos/size at submission, self-optimize data structures to stable linear access.
- tabs: "there is currently a problem because TabItem() will try to submit their own tooltip after 0.50 second, and this will have the effect of making your tooltip flicker once." -> tooltip priority work (WIP branch)
- image/image button: misalignment on padded/bordered button?
- image/image button: parameters are confusing, image() has tint_col,border_col whereas imagebutton() has bg_col/tint_col. Even thou they are different parameters ordering could be more consistent. can we fix that?
- slider: allow using the [-]/[+] buttons used by InputFloat()/InputInt()
- slider: add dragging-based widgets to edit values with mouse (on 2 axises), saving screen real-estate.
- slider: tint background based on value (e.g. v_min -> v_max, or use 0.0f either side of the sign)
- listbox: make it easier/more natural to implement range-select (need some sort of info/ref about the last clicked/focused item that user can translate to an index?) (WIP range-select branch)
!- popups/menus: clarify usage of popups id, how MenuItem/Selectable closing parent popups affects the ID, etc. this is quite fishy needs improvement! (#331, #402)
- modals: make modal title bar blink when trying to click outside the modal
- modals: technically speaking, we could make Begin() with ImGuiWindowFlags_Modal work without involving popup. May help untangle a few things, as modals are more like regular windows than popups.
- popups/modals: although it is sometimes convenient that popups/modals lifetime is owned by imgui, we could also a bool-owned-by-user api as long as Begin() return value testing is enforced.
- tooltip: drag and drop with tooltip near monitor edges lose/changes its last direction instead of locking one. The drag and drop tooltip should always follow without changing direction.
- status-bar: add a per-window status bar helper similar to what menu-bar does. generalize concept of layer0 rect in window (can make _MenuBar window flag obsolete too).
- menus: would be nice if the Selectable() supported horizontal alignment (must be given the equivalent of WorkRect.Max.x matching the position of the shortcut column)
- tree node/opt: could avoid formatting when clipped (flag assuming we don't care about width/height, assume single line height? format only %s/%c to be able to count height?)
- settings/persistence: helpers to make TreeNodeBehavior persist (even during dev!) - may need to store some semantic and/or data type in ImGuiStoragePair
- style: a concept of "compact style" that the end-user can easily rely on (e.g. PushStyleCompact()?) that maps to other settings? avoid implementing duplicate helpers such as SmallCheckbox(), etc.
- style: try to make PushStyleVar() more robust to incorrect parameters (to be more friendly to edit & continues situation).
- log: have more control over the log scope (e.g. stop logging when leaving current tree node scope)
- log: be able to log anything (e.g. right-click on a window/tree-node, shows context menu? log into tty/file/clipboard)
- log: let user copy any window content to clipboard easily (CTRL+C on windows? while moving it? context menu?). code is commented because it fails with multiple Begin/End pairs.
- drag and drop: drag source on a group object (would need e.g. an invisible button covering group in EndGroup) https://twitter.com/paniq/status/1121446364909535233
- drag and drop: make it easier and provide a demo to have tooltip both are source and target site, with a more detailed one on target site (tooltip ordering problem)
- drag and drop: (#143) "both an in-process pointer and a promise to generate a serialized version, for whether the drag ends inside or outside the same process"
- drag and drop: feedback when hovering a region blocked by modal (mouse cursor "NO"?)
- text/layout/tabs: \t pulling position from base pos + step, or offset array (e.g. could be used in text edit, menus for simple icon+text alignment, etc.)
- font: enforce monospace through ImFontConfig (for icons?) + create dual ImFont output from same input, reusing rasterized data but with different glyphs/AdvanceX
- font: storing MinAdvanceX per font would allow us to skip calculating line width (under a threshold of character count) in loops looking for block width
- font: optimization: for monospace font (like the default one) we can trim IndexXAdvance as long as trailing value is == FallbackXAdvance (need to make sure TAB is still correct), would save on cache line.
- font: add support for kerning, probably optional. A) perhaps default to (32..128)^2 matrix ~ 9K entries = 36KB, then hash for non-ascii?. B) or sparse lookup into per-char list?
- font: add a simpler CalcTextSizeA() api? current one ok but not welcome if user needs to call it directly (without going through ImGui::CalcTextSize)
- font/opt: Glyph currently 40 bytes (2+9*4). Consider storing UV as 16-bits integer? (->32 bytes). X0/Y0/X1/Y1 as 16 fixed-point integers? Or X0/Y0 as float and X1/Y1 as fixed8_8?
- nav: some features such as PageUp/Down/Home/End should probably work without ImGuiConfigFlags_NavEnableKeyboard? (where do we draw the line? how about CTRL+Tab)
- nav: expose wrap around flags/logic to allow e.g. grid based layout (pressing NavRight on the right-most element would go to the next row, etc.). see internal's NavMoveRequestTryWrapping().
- nav: restore/find nearest NavId when current one disappear (e.g. pressed a button that disappear, or perhaps auto restoring when current button change name)
- nav: SetItemDefaultFocus() level of priority, so widget like Selectable when inside a popup could claim a low-priority default focus on the first selected iem
- nav: NavFlattened: broken: in typical usage scenario, the items of a fully clipped child are currently not considered to enter into a NavFlattened child.
- nav/popup: esc/enter default behavior for popups, e.g. be able to mark an "ok" or "cancel" button that would get triggered by those keys, default validation button, etc.
- nav/treenode: left within a tree node block as a fallback (ImGuiTreeNodeFlags_NavLeftJumpsBackHere by default?)
- nav/menus: when using the main menu bar, even though we restore focus after, the underlying window loses its title bar highlight during menu manipulation. could we prevent it?
- nav/menus: Alt,Up could open the first menu (e.g. "File") currently it tends to nav into the window/collapse menu. Do do that we would need custom transition?
- misc: idle: if cursor blink if the _only_ visible animation, could even expose a dirty rectangle that optionally can be leverage by some app to render in a smaller viewport, getting rid of much pixel shading cost.
- backends: glfw: could go idle when minimized? if (glfwGetWindowAttrib(window, GLFW_ICONIFIED)) { glfwWaitEvents(); continue; } // issue: DeltaTime will be super high on resume, perhaps provide a way to let impl know (#440)
- backends: opengl: rename imgui_impl_opengl2 to impl_opengl_legacy and imgui_impl_opengl3 to imgui_impl_opengl? (#1900)
- backends: opengl: could use a single vertex buffer and glBufferSubData for uploads?
- backends: opengl: explicitly disable GL_STENCIL_TEST in bindings.
- backends: vulkan: viewport: support for synchronized swapping of multiple swap chains.
- backends: emscriptem: with refactored examples, we could provide a direct imgui_impl_emscripten platform layer (see eg. https://github.com/floooh/sokol-samples/blob/master/html5/imgui-emsc.cc#L42)
- bindings: ways to use clang ast dump to generate bindings or helpers for bindings? (e.g. clang++ -Xclang -ast-dump=json imgui.h) (--> use https://github.com/dearimgui/dear_bindings)
- optimization: add clipping for multi-component widgets (SliderFloatX, ColorEditX, etc.). one problem is that nav branch can't easily clip parent group when there is a move request.
- optimization: add a flag to disable most of rendering, for the case where the user expect to skip it (#335)
