Changelog
Every release, and what is done since the last one (Unreleased). The state of the work is in the roadmap. The list is generated from the commit messages by git-cliff (git cliff -o CHANGELOG.md; scripts/release.sh regenerates it for each release).
Migration to v0.8.0
What a program written for v0.7.0 meets in v0.8.0, with a link to the section that has the details. Nothing a v0.7.0 program calls changed its meaning; the entries below are new optional arguments, new procedures, and pages that moved.
| Change | What to do |
|---|---|
Error bars: ip_error_bars takes horizontal before stat | nothing; a call passing stat by position (none in FIGURE's examples) must name it |
3-D plots: ip3_line takes colors, ip3_scatter colors and sizes, ip3_setup_axis format, each before stat | as above |
New public names of use figure: ig_color_picker, ip_pie_chart and ip_bubbles, ip3_setup_axis_ticks, ip3_push_style_color, ip3_pop_style_color, ip3_push_style_var, ip3_pop_style_var | nothing: a program with a name of its own that clashes renames it, or uses figure with only |
Data in place: a 2-D item, a drag tool or a plot query outside a plot is a usage error (FIG_ERR_STATE, <p>: no plot is being declared (...)), in every checks mode; in 'report' and 'recover' it crashed in ImPlot | nothing for a correct program; one that drew an item outside a plot now gets the status, or the message at the next app%next_frame |
Images: tex%destroy from a task is refused (FIG_ERR_STATE, the texture kept), as init, update and load are | nothing for a correct program; release textures on the main thread |
The tutorial has a new chapter 7, Building and running the studio; chapters 7 to 25 are now 8 to 26, and their programs tut07_* to tut25_* are tut08_* to tut26_* | a bookmark to a chapter, or a copy of a program, takes the new number |
The documentation's own tooling: scripts/check_docs.py also checks tables, links, generated pages, and that every public name is in a sentence of its page and called by a compiled program | nothing for a program; a contribution keeps them green (Contributing) |
Migration to v0.7.0
What a program written for v0.6.0 meets in v0.7.0, with a link to the section that has the details. Nothing a v0.6.0 program calls changed its meaning; the entries below are new optional arguments and what the build installs.
| Change | What to do |
|---|---|
Lists and trees: ig_selectable and ig_tree_node take two more optional arguments at the end, index and stat; list_clipper_object%begin takes keep | nothing; a call by position keeps its meaning |
Frame checks: three more pairs are checked, ip_begin_aligned / ip_end_aligned, ip_begin_legend_popup / ip_end_legend_popup (end only after a true begin) and ig_begin_multi_select / ig_end_multi_select (always paired) | nothing; a program that drives these through the raw layer (ImPlot_BeginAlignedPlots, igBeginMultiSelect) is not checked, as before |
Plots: FIGURE patches one line of ImPlot (native/patches/implot-0001-digital-rect-order.patch): its digital bars drew nothing with Dear ImGui 1.93 | nothing; a program that called ImPlot_PlotDigital_* through the raw layer now sees its bars |
Installation: the installer also writes a CMake package (lib/cmake/figure-<tag>/), and builds with Intel ifx (--intel); each release has a sources archive for machines without network (Building without network) | nothing; a CMake project that wrote its own pkg_check_modules recipe can use find_package(figure CONFIG) and figure::figure instead |
New public names of use figure: the file API (figure_make_folder, figure_remove, figure_set_file_time, figure_file_time, figure_file_exists), drag and drop (ig_drag_source, ig_drop_target, ig_dragging, ip_drop_target), plots across plots (ip_link_axis, ip_begin_aligned, ...), multi-selection (selection_object, ig_begin_multi_select, ...) and 3-D plots (ip3_triangles, ip3_image, ip3_mouse_ray, ...) | a program with its own procedure of one of these names imports figure with only: (as it would for any new name) |
Migration to v0.6.0
What a program written for v0.5.0 meets in v0.6.0, with a link to the section that has the details.
| Change | What to do |
|---|---|
Tables: FIGURE patches Dear ImGui's 0 / 0 in a SizingStretchProp table's first frame (ocornut/imgui#8718); the columns start equal there. A program built with -ffpe-trap=invalid no longer stops at its own such table, and Markdown is drawn with the traps as the program set them (v0.5.0 held them) | nothing to change; a program that held the traps itself around such a table can stop doing so |
| Finding items by path: the items are recorded in every mode from the first frame; v0.5.0 recorded window and remote programs only from their first path call, which found nothing | nothing; a program that called test_record_items(.true.) for that first call can drop it. A window program that never finds items by path pays about 0.02 µs per item and frame; test_record_items(.false.) before app%init gives it back |
Dear ImGui at the docking branch's head (1.93.1 WIP, cf53467; v0.5.0 had 1.93.0): every arrow (tree nodes, combos, collapsing headers) is a stroked chevron, and the dark style's window, popup and child backgrounds are darker. The style has two more variables, SymbolArrowSizeFactor and SymbolArrowThicknessFactor (Styles) | a program's own golden pictures (test_compare_image) of its windows are recorded again, once |
Raw bindings: igTreeNodeGetOpen takes a second argument, flags (cimgui's generated API at the new Dear ImGui); FIGURE's own procedures are unchanged | a program calling igTreeNodeGetOpen(id) from figure_cimgui passes flags too (0_c_int for none) |
3-D views: while the user drags or turns the wheel (fast_moves), the half-size renders leave out the meshes' grid lines (edges); the render at the end of the move draws them. threads takes up to 16 (was 4); the vertex caches are freed once the view rests (render_info(cache_bytes=)) | nothing; a program that wants more speed on a fine grid can pass more threads |
Migration to v0.5.0
What a program written for v0.4.0 meets in v0.5.0, with a link to the section that has the details.
| Change | What to do |
|---|---|
Plot data in place: the 1-D items' array dummies are asynchronous, so nvfortran passes rows, strided and reversed sections in place instead of copying them; the plot is the same | nothing to change in a program; a test that checks test_plot_stride() or test_plot_copies() under nvfortran now sees the section's own stride, and FIGURE's copy of a reversed section counted |
figure_last_error: the message is kept per thread; a failure in a task no longer overwrites the main thread's message | nothing; a program that read, on the main thread, a message set by a failure inside a task reads it from the task (t%result gives the task's own failure, as before) |
Synthetic input: queued input gives the frames an event gives (three more, then idle), and in window mode a lazy loop no longer waits for events while input is queued; Dear ImGui is built with its item hooks (IMGUI_ENABLE_TEST_ENGINE), off unless items are recorded | nothing; a window-mode test runs its queued input without the max_wait pauses; a program built on the raw cimgui layer that tests ImGuiItemStatusFlags sees five more bits defined, which Dear ImGui passes to the hooks only, never to LastItemData |
Windows and layouts: ig_begin(name) without open gives a window registered with app%add_window a close button, which hides it; ig_dock_layout_begin returns .true. again after app%switch_layout or app%reset_layout | nothing: without add_window, switch_layout and reset_layout both behave as in v0.4.0; a program that registers a window it closes with its own flag passes open as before, or closable=.false. |
app%save_image works in every mode: headless and remote modes draw the frame with FIGURE's CPU renderer, where v0.4.0 refused them with FIG_ERR_STATE; its failure message is save_image: '<path>' cannot be written: <reason> (was save_image: cannot write <path>: <reason>) | nothing; a test that expected FIG_ERR_STATE headless now gets the file; a program that matched the old message matches the new one |
FIGURE_CAPTURE records headless and remote programs too (the CPU renderer's frames), where v0.4.0 ignored it outside a window; FIGURE_CAPTURE_HEADLESS=1 records a window program without a window | nothing; a headless test run with FIGURE_CAPTURE set in its environment now writes frames and ends after FIGURE_CAPTURE_FRAMES |
| Headless mode: Dear ImGui's font atlas gets ids with the top bit set (v0.4.0 gave every page id 1, which the first user texture also had), and the sampler callbacks are registered, so an image drawn headless uses its texture's own filter | nothing, unless a program read ImTextureIDs through the raw layer: the atlas's are no longer 1 |
The native layer is compiled with -ffp-contract=off: no fused multiply-add in Dear ImGui's, ImPlot's and FIGURE's C++ | nothing on x86-64 built for any processor (no FMA without -march); on aarch64, or with -march=native, a layout rounding at a tie may move by one pixel, now the same as everywhere |
Shortcuts: a fourth scope, 'always', not routed: every call of the chord fires, whatever has the focus, a text field being edited included. v0.4.0's migration note said 'global' gives v0.3's behaviour; it does not when two calls bind one chord (only the first 'global' call of the frame fires) | for v0.3's behaviour exactly, scope='always'; a single key that must work everywhere keeps scope='global'; ig_set_next_item_shortcut refuses 'always' (FIG_ERR_ARGUMENT) |
Input scripts: a capture script's <frame> wait holds the recording until the program's tasks and processes are done; the documentation's pictures are all the CPU renderer's, checked in CI | nothing in a program; a contributor whose change alters a picture of the site records it again (python3 scripts/docs_media.py --headless NAME, or --check then --update) and commits it with the change |
| Markdown: the native layer compiles imgui_rich_md and md4c (MIT) and embeds IBM Plex Sans Bold, Italic and Bold Italic (OFL-1.1): a program linked with FIGURE grows by about 1 MB | nothing in the code; a binary distributed with its licences ships the two new ones too (share/licenses/imgui_rich_md/, share/licenses/md4c/; Third-party software) |
Migration to v0.4.0
What a program written for v0.3.1 meets in v0.4.0. Each line links to the "Changed in v0.4.0" box of the section that has the details: the old behaviour, the new one, and why.
| Change | What to do |
|---|---|
Usage errors reported: a wrong end, a scope left open or a drawing call with no frame open no longer abort inside Dear ImGui; next_frame(stat) reports them (mode 'report', the default) | pass stat to next_frame and act on FIG_ERR_STATE, or app%set_checks('abort') to stop at the faulty call; a loop without stat stops with FIGURE's message |
Shortcuts routed: ig_shortcut's default scope is 'focused'; a call inside a window fires only while that window has the focus, and no longer while a text field uses the key | calls outside every window behave as before; a key inside a window that must work everywhere takes scope='global'; two calls of one chord that must both fire, as in v0.3, take scope='always' (from v0.5.0) |
Numbers: ig_input and ig_slider take every kind, vectors, formats, scale, clamp; they moved from figure_imgui to figure_numbers; a slider with ends beyond half the kind's range is refused | nothing for use figure; a use figure_imgui, only: ig_input names figure_numbers or figure instead |
Tables: ig_begin_table with columns outside 1 to 511 is refused with FIG_ERR_ARGUMENT | pass a valid count (it aborted before) |
Plot axes: ip_setup_axes takes xscale/yscale; xlog/ylog are deprecated synonyms; new arguments come before stat; setup calls out of place are usage errors | name stat if it was passed by position; prefer yscale='log' to ylog=.true. |
Data in place: the plot items take arrays of any stride and integer abscissae; the dummies are no longer contiguous | nothing: every v0.3.1 call compiles and draws as before |
Drag tools: ip_drag_line_x, ip_drag_line_y, ip_drag_point outside a plot are usage errors | call them between ip_begin_plot and ip_end_plot (it aborted before) |
The tutorial's part three shows the new calls in the solver studio of part two.
All notable changes to this project are documented here. Versions follow Semantic Versioning. Format follows Keep a Changelog.
[0.8.0] — 2026-11-19
Added
widgets: ig_color_picker, the colour picker drawn in place
plots: Pie charts, bubbles and error bars along x
plot3d: Colours and sizes per point, axis formats and ticks, ImPlot3D's style by name
Documentation
The contradictions with the code, in one sweep
showcase: The solvers run from the window, recorded headless
tutorial: A build-along, a chapter on building and running the studio
One first step, Install reordered
Known issues as an answer, the roadmap's Next first
tutorial: Concepts explained where they are first used
Widgets, menus, docking, styles, frame loop and errors, in a learner's order
Plots, images, colormaps, 3-D views, editor and canvas, completed
Processes, remote, testing, capture, performance and architecture, completed
features: The feature map as a lookup
cookbook: Regrouped, seven programs made recipes, the rows in view
v0.8.0's roadmap section and migration notes
Every public name called by a docs program and named in a sentence
remote: The page called Remote mode, as the rest of the docs call it
The API pages built, not committed; two items deferred on the roadmap
Fixed
docs: The raw-binding policy sentence written by the generator
build: Declare the three v0.8.0 tests in fpm.toml
tests: plot_more writes its colormaps in the working folder
plots: A 2-D item outside a plot is a usage error, not a crash
images: tex%destroy from a task is refused, as init, update and load
colormaps: ip_colormap_stops returns complete arrays or none
[0.7.0] — 2026-11-16
Added
files: Make, remove, date and test files and folders on every platform
capture: A failed picture keeps the font atlas it was drawn with
install: A CMake package, find_package(figure) and figure::figure
install: The installer takes Intel ifx (--intel)
build: A sources archive per release, the native layer built without network
dnd: Drag and drop between items and onto plots
plots: Plots across plots, digital signals and grouped bars
select: Multi-selection in lists, tables and trees
implot3d: Triangles, quads, images, axis scales, the box, the legend and picking
Changed
- docs: The studio tutorials write their example cases without a shell
Documentation
known-issues: nvfortran's internal error on a local named size, reduced
v0.7.0's documentation pass
S3's additions read afresh, the known issues of v0.7.0 checked again
v0.7.0's pictures measured on the three platforms, the fixes after block D
showcase: Six of the showcase's pictures from FIGURE's renderer
Fixed
ci: release.yml back to the scaffold's; the sources archive in its own workflow
ci: The installer's CMake check finds macOS's gfortran-N
ci: The sources archive's runner gets SDL3's build packages
test: Ctrl+click in the multi-selection test is cmd+click on macOS
process: A detached child's end seen where init does not reap it
[0.6.0] — 2026-11-15
Added
testing: Items recorded in every mode from the first frame
view3d: Up to 16 render threads
imgui: Dear ImGui at the docking head, cimgui generated by FIGURE ⚠ BREAKING CHANGE
loop: FIGURE_DEBUG_WAKE logs what ends each wait of the loop
Documentation
names: The name map follows the native glue, and tells what FIGURE uses internally
view3d: Mesh ids go on after remove and clear
known-issues: The pictures on macOS, measured
changelog: v0.6.0's migration notes for Dear ImGui, the raw layer, 3-D views
known-issues: The entries v0.6.0 touched, checked and dated
frame-loop: What ended a wait, read afresh
roadmap: v0.6.0's section and a real "Next"
Fixed
native: Dear ImGui's 0/0 in a SizingStretchProp table patched, the Markdown trap hold goes
editor: A fresh code editor's first frames are finite under traps
tests: The editor canary's Linux flag is not named linux
native: A program built with traps opens its window
capture: Capture scripts know the shortcut key; recordings play alike
native: fig_time keeps its C linkage under clang
Performance
view3d: The vertex caches are freed when the view rests
view3d: The render's work shared out among its threads
view3d: No grid lines while the camera moves
[0.5.0] — 2026-11-11
Added
imgui: ig_shortcut scope 'always', the unrouted chord of v0.3
clock: figure_time, a monotonic clock in seconds
fonts: Noto Sans Math, the fallback font after Plex and Lucide
images: fig_read_image and tex%load, PNG and JPEG files through stb_image
testing: Items by path, a frame driver, id and label checks
app: A window registry with its View menu, layouts, preferences, message boxes, splitters
render: Pictures without a GPU, a CPU renderer of Dear ImGui's draw data
editor: A code editor with highlighting, marks, completion and a diff
view3d: A view cube, gizmos and a gradient editor (ImGuizmo)
implot3d: Small 3-D figures, ImPlot3D (ip3_*)
markdown: Markdown in the window, imgui_rich_md (ig_markdown, markdown_view_object)
Documentation
The version strings at v0.4.0, and the roadmap at v0.5.0
Every picture from one renderer, checked in CI; tutorial part three grows
Performance
- implot: ASYNCHRONOUS plot dummies, nvfortran passes sections in place
[0.4.0] — 2026-11-10
Added
app: Usage errors reported, not aborted: scope record, frame check, app%set_checks
imgui: One snapshot of what happened to the last item
imgui: Scroll to an item and give the keyboard focus
imgui: Sizes in em, groups, alignment and the layout metrics
imgui: Context menus, tooltips with content and links
imgui: Tables with a frozen header, sorting, row colours and trees across columns
imgui: Numbers of every kind, vectors, drags, log sliders, angles, ranges, radio buttons and list boxes
canvas: An area that takes the mouse, and drawing primitives on the window, the foreground, the background or a plot
implot: Plot items read strided arrays in place and take integer abscissae
implot: Six plot axes with scales, fixed ticks, constraints and the legend's place
implot: Read a plot under the pointer, keep a box selection, tags, infinite lines, text, stairs, stems and histograms
implot: A field through a colormap in one call, and a legend the user drags
imgui: A player for snapshots: previous, play, next, a slider, a speed and a loop mode
Documentation
- The documentation complete and checked: doc-comment and example checks, the Dear ImGui name map, tutorial part three
Fixed
- imgui: Shortcuts routed to the focused window, not fired in text fields
[0.3.1] — 2026-11-09
Documentation
- showcase: xall_gui's grid in the 3-D view, overset legacy's editor and the solvers run from the window
Fixed
task: Quit and request_frames safe from a task, and app%wake to start the next frame from any thread
implot: Printf formats checked against the value's kind before Dear ImGui or ImPlot uses them
view3d: The 3-D view owns the mouse wheel while the pointer is over it
testing: The tsan check runs on a fresh clone; wake latency allowed 1 s on CI runners
[0.3.0] — 2026-11-08
Added
style: Looks and palettes, colours derived from twelve roles
app: The window's frame drawn by FIGURE, with the menus in its title bar; the window's icon
imgui: Trees, with a click on the label told apart from the arrow
process: Programs started from the program, captured or detached, and a log view of their output
task: Background tasks with progress and cancel, and the rule that only the main thread draws
app: Progress bars, notifications drawn by FIGURE, and a status bar recipe
files: A file browser drawn by FIGURE, the same in a window, in remote mode and in tests
imgui: Multi-line text in a buffer of any length, and the monospace font plex-mono
install: A one-line installer for FIGURE, per compiler, with pkg-config and uninstall
app: Files and folders dropped on the window, and the clipboard
implot: A colormap registry, with scientific colormaps bundled
implot: Error bars, annotations, drag lines and points, and contour lines with holes where the field is NaN
imgui: Colour editing with a picker, and colour-scale legends with discrete levels as bands
view3d: A 3-D view of meshes coloured by a field, drawn by a software rasterizer in every mode
app: prefer='figure' for the file dialogs, a title bar height in looks, the notification fade tested
remote: An image changing frame after frame travels as JPEG, exact once it is still
Documentation
api: The API pages regenerated for looks and palettes
api: The API pages regenerated for the window's frame and icon
api: The API pages regenerated for trees
api: The API pages regenerated for processes and the log view
api: The API pages regenerated for background tasks
api: The API pages regenerated for progress bars and notifications
api: The API pages regenerated for the file browser
api: The API pages regenerated for multi-line text, drops and the clipboard
api: The API pages regenerated for the colormap registry
api: The API pages regenerated for contours, legends, colour editing and the 3-D view
tutorial: Part two begins: the studio's own window, its shell, its case
tutorial: The studio edits its input, runs its solver and shows the result
tutorial: The studio shows the field in 3-D, reads large results off the frame, and fails well
tutorial: The studio tested headless and served to a browser; the tutorial in two parts
api: The API pages regenerated for set_icon and the ini file during a capture
task: Why Dear ImGui's context is not made thread-local, measured
api: The API pages regenerated for the open questions closed before v0.3.0
Fixed
The fpm package embeds looks and palettes; two tests portable to macOS
task: Task procedures recursive; headless keeps its own clipboard
test: The log view's copy pressed with the platform's shortcut key
task: The entry points that refuse a task's drawing are recursive
process: The captured lines no longer overwrite the heap
ci: The window check accepts a window clamped by a small display, and the close button on macOS's side
app: The bar's icon is set where the system has no window icons; a capture ignores the ini file
Names taken without trailing blanks, band edges labelled exactly, drops refused from a task
remote: The page's notices stay readable beside the frame counter
style: FIGURE's palettes derive every colour from their roles; paper's MinorAlpha in effect
task: Every drawing procedure FIGURE wraps refuses a call from a task
app: The app procedures that drive Dear ImGui or the window refuse a call from a task
[0.2.0] — 2026-11-06
Added
implot: Logarithmic axes in ip_setup_axes (xlog, ylog)
imgui: Text fields, integer fields and the next item's width
imgui: Drop-down lists and tooltips
imgui: Collapsing headers, tabs, id scopes and disabled blocks
imgui: Tables
testing: test_click and test_drag let the pointer rest before pressing (F1)
imgui: Selectable lines, wrapped/disabled/coloured text, child regions (F2)
imgui: Menus and keyboard shortcuts (F3)
Popups, modals and a confirmation before quitting (F4)
app: The system's file dialogs (F5)
implot: Tick formats, subplots, docking a window that appears later (F6)
app: The window's title after init (set_title, title)
app: Save a rectangle of the window as a PNG (save_image)
app: Save and load the arrangement of the windows (layouts)
imgui: A bar along one side of the window (ig_begin_side_bar)
style: Lucide icons in every font, by name (ICON_*)
imgui: Workspace buttons, window focus, list clipper, text level with fields
imgui: The size of a text in the current font (ig_text_size)
Changed
- imgui: ig_text_size's result named extent
Documentation
Forms in the widgets guide, a cookbook editor, roadmap v0.2.0
showcase: The tools built for Xall with FIGURE (R1)
api: The API pages regenerated for F7 (figure_library, set_title, title)
showcase: The run checker's navigator and comparison
showcase: The run checker saves its plots (image, data)
The overview pages catch up with v0.2.0 so far
showcase: xall_gui, a case's input and run in one window
showcase: xall_gui's case files, checked as the solver reads them
api: The API pages regenerated for the new wrappers
The pages say v0.2.0, released
Fixed
fpm: Give the package commit an author on CI
native: Capture scripts accept every key name test_key accepts
test: Select a field's text with keys every platform reads alike
testing: A 'shortcut' key name for the platform's shortcut key
build: Ship figure_testing in the static library
app: set_font's size reaches the next frame
app: File dialogs start from the program's working directory
[0.1.0] — 2026-11-05
Added
native: Add FIGURE_DEBUG_REDRAW diagnostics and an idle-CPU probe
bindings: Generate the raw cimgui/cimplot layer from their metadata (M2)
bindings: Generate the C++ glue for struct-by-value functions
implot: Add scatter, bars and shaded items and item styles (ip_spec)
implot: Plot matrix rows in place; no imgui.ini unless asked
Add GPU textures, heatmaps and colour scales
Add a Mandelbrot explorer and a 4K redraw benchmark (M3 done)
remote: Serve the UI to a web browser (M5, remote)
Brand the docs, add styles and frame capture (v0.1.0 docs gate)
dock: Turn docking on; dock spaces and program-defined layouts (S2)
style: Every Dear ImGui and ImPlot style colour and variable by name (S3)
style: Themes as files, embedded and parsed natively (S4)
style: Paper and high-contrast themes, Okabe-Ito lines, contrast test (S5)
style: Load and save theme files; the style editor in a window (S6)
style: IBM Plex Sans by default; fonts by name or file, scoped (S7)
style: Window flags and scoped style changes for both libraries (S8)
testing: Public synthetic-input module figure_testing (S10)
remote: Pace frames for the browser: a cap, backpressure, a real wait (S13)
remote: 12-byte vertices, and deflate when the link is the bottleneck (S13)
remote: An access token in the address, random by default (S14)
Changed
- implot: Take real64 arrays in ip_line, keep C kinds in figure_c_native
Documentation
plan: Record the native layer rebuilt with Wayland and DBus
Document GALLIUM_DRIVER=d3d12 for WSLg and record the driver check
Add the VitePress documentation site
manual: Point the example links at the main branch
plan: Record the name collision check
Record M4 as done, all platforms green
Retire plans/plan.md; issue #1 is the plan, the docs the reference
style: A theme gallery, recorded from one program in every style (S9)
performance: A performance guide built on the measurements (S15)
features: List every public procedure; check it with scripts/check_docs.py (S17)
media: Regenerate every recording for v0.1.0 (S17)
changelog: Generate CHANGELOG.md with git-cliff; the docs page includes it (S17)
readme: The mark, an animation, a compiled example, install with FoBiS and fpm (S17)
Check every page against the code for v0.1.0 (S17)
site: Fixes from a review of the built site (S17)
Fixed
build: Anchor build-product paths in .gitignore
base: Give Fortran-side failures their own error message
ci: Build SDL with clang on macOS, run rules through bash, free disk for nvhpc
build: Order the fpm link line; link clang's runtime on macOS
build: Add the platform libraries after fpm's objects (figure_platform)
native: Bound the glyph advance clamp so fonts above 15 px do not overflow
remote: Draw lists over 65536 vertices in remote and headless modes
ci: Retry the remote test's verdict open; Apple clang for the macOS fpm dependency
ci: C++17 for the macOS fpm dependency (Apple clang defaults to C++98)
test: The remote test's wrong token always differs from the token
Performance
remote: Measure redraw and bandwidth, locally and through a tunnel (S11)
remote: Skip frames equal to the last one sent (S12)