Files
udpscope/tools/make-help-shots.sh
T
Martino Ferrari 6138e6bdce fix: the figures carried a chunk WebKit refuses, which is why a page never loaded
The user's report was precise and I misread it twice: opening trigger.page as a *file* rendered the
page with no image, while opening it through the app's `help:` URI spun forever. Both are one cause.
The three figures carried a `bKGD` chunk -- and the PNG specification does not permit `bKGD` on an
image with an alpha channel, which all three are (colour type 6). ImageMagick both writes and reads
it, so my "decodes fully" check proved nothing: I tested with a decoder more permissive than the one
that has to load the file. WebKit rejects it, so the image never appears; over `help:` the fetch
never completes either, so the page's load event never fires and Yelp's spinner never stops.

It came from `import`, which records the root window's background. Every figure now carries only the
chunks a decoder is guaranteed to accept (IHDR/IDAT/IEND), and `tools/make-help-shots.sh` normalises
its output at the end so a re-shoot cannot reintroduce it -- with the reason written where the next
reader will be, since "ImageMagick reads it" is exactly the trap.

`tools/check_help.sh` gains the rule, because the four yelp-check subcommands and the structural
rules are all blind to a figure no decoder will accept: a `bKGD` on colour type 4 or 6 is a finding,
checked directly rather than through a permissive tool. It is a whitelist because a decoder is one.

Verified: the rule fires on the committed figures (colour type 6, bKGD present) and passes on the
fixed ones; the figures still decode and are still 1280x800 / 938x420; check_help.sh is green on the
source and installed sets; make -C build test-help passes.
2026-10-15 00:01:52 +02:00

237 lines
10 KiB
Bash
Executable File

#!/bin/bash
# SPDX-License-Identifier: MIT
# SPDX-FileCopyrightText: 2026 Martino Ferrari
#
# Regenerates the screenshots the Mallard manual carries (help/C/<id>/figures/*.png).
#
# It runs udpscope headlessly under its own Xvfb, against tools/mock_udpstreamer.py, drives it
# into a state with the same `UDPSCOPE_*` probes the self-check uses, grabs the window with
# ImageMagick's `import`, and draws numbered annotation boxes on a borderless crop with
# `magick`.
#
# Two things about a headless box, both measured rather than assumed:
#
# * GTK4 picks the Vulkan renderer by default, and under Xvfb that surface is not readable
# back through X11 -- `import` sees a window with no pixels (standard deviation 0). Forcing
# the Cairo renderer (`GSK_RENDERER=cairo`, with `LIBGL_ALWAYS_SOFTWARE=1`) makes GTK paint
# with XPutImage, which `import` can read.
# * `xdotool search --name udpscope` finds nothing: GTK4 sets no WM_NAME on the X window it
# creates, so windows are found by their geometry, not their name.
#
# Box provenance: the sidebar width is read from the app's own probe dump (`side=NNNx`), so the
# side-panel box tracks the real widget; the header and status bands are the table in
# `window_boxes` below, each named where it is used, because the dump reports neither.
#
# Usage: tools/make-help-shots.sh [OUTPUT_DIR] (default help/C/<id>/figures)
set -euo pipefail
here=$(CDPATH= cd -- "$(dirname -- "$0")" && pwd -P)
root=$(dirname -- "$here")
app=${UDPSCOPE_SHOTS_APP:-$root/build/udpscope}
mock=$here/mock_udpstreamer.py
help_id=org.udpscope.app
out=${1:-$root/help/C/$help_id/figures}
port=18391
for tool in Xvfb import magick xdotool python3; do
command -v "$tool" >/dev/null || { echo "make-help-shots: $tool is required" >&2; exit 1; }
done
[ -x "$app" ] || { echo "make-help-shots: $app is missing; build it first" >&2; exit 1; }
[ -f "$mock" ] || { echo "make-help-shots: $mock is missing" >&2; exit 1; }
mkdir -p "$out"
# --- the window's own geometry, and the boxes it is annotated with -------------
# main-window.ui asks for 1280x800, and the sidebar is a fixed 320 px
# (sidebar-width-unit=px, min=max=320). The header bar is the window's titlebar and the status
# row sits at the foot of the plot column; neither reports a pixel height, so their bands are
# the third measurement -- a pixel scan of a grab, 51 px and 43 px on the current theme -- and
# are refreshed by re-measuring if the styling changes.
WIN_W=1280
WIN_H=800
HEADER_H=51
STATUS_H=43
PLOT_W=952
boxes_main="header bar:0:0:$WIN_W:$HEADER_H
side panel:SIDE_X:$HEADER_H:SIDE_W:706
plot:0:$HEADER_H:$PLOT_W:706
status bar:0:757:$PLOT_W:$STATUS_H"
boxes_trigger="event held, axis pinned to the event:0:757:$PLOT_W:$STATUS_H
the trigger marker at x = 0:0:$HEADER_H:$PLOT_W:706"
boxes_math="the graph editor:8:8:EW-24:EH-24"
display=
xvfb_pid=
app_pid=
mock_pid=
cleanup() {
[ -n "$app_pid" ] && kill "$app_pid" 2>/dev/null || true
[ -n "$mock_pid" ] && kill "$mock_pid" 2>/dev/null || true
[ -n "$xvfb_pid" ] && kill "$xvfb_pid" 2>/dev/null || true
}
trap cleanup EXIT
# A display of our own, so nothing on the developer's screen is grabbed or disturbed.
for n in $(seq 210 240); do
if Xvfb ":$n" -screen 0 1600x1000x24 -nolisten tcp >/dev/null 2>&1 & then
xvfb_pid=$!
sleep 1
if kill -0 "$xvfb_pid" 2>/dev/null; then display=":$n"; break; fi
fi
done
[ -n "$display" ] || { echo "make-help-shots: no free Xvfb display in :210..:240" >&2; exit 1; }
export DISPLAY="$display"
export GDK_BACKEND=x11 GSK_RENDERER=cairo LIBGL_ALWAYS_SOFTWARE=1
# A throwaway config dir: the app restores the last session for a source, and a restored
# layout would change the very arrangement the screenshot is meant to show.
tmp=$(mktemp -d)
trap 'cleanup; rm -rf "$tmp"' EXIT
export XDG_CONFIG_HOME="$tmp/config" XDG_CACHE_HOME="$tmp/cache" XDG_DATA_HOME="$tmp/data"
python3 "$mock" --port "$port" --rate 10000 --batch 100 >"$tmp/mock.log" 2>&1 &
mock_pid=$!
sleep 0.5
# The largest window that is not the 1600x1000 X root; that is the app's 1280x800 one (a math
# editor opened on top is smaller, and X may also expose unnamed GTK helper windows).
pick_window() {
local id geo w best= best_w=0
for id in $(xdotool search --onlyvisible "" 2>/dev/null); do
geo=$(xdotool getwindowgeometry "$id" 2>/dev/null | awk '/Geometry/{print $2}')
[ "$geo" = "1600x1000" ] && continue
w=${geo%x*}
case "$w" in ''|*[!0-9]*) continue;; esac
if [ "$w" -gt "$best_w" ]; then best=$id; best_w=$w; fi
done
[ -n "$best" ] && echo "$best"
}
# The math editor is a window of its own (a `GtkWindow`, presented on the main one), so it is
# the largest window that is neither the X root nor the app's 1280x800 one and that has a real
# size (X exposes 1x1 helper windows too).
pick_other_window() {
local id geo w best= best_w=0
for id in $(xdotool search --onlyvisible "" 2>/dev/null); do
geo=$(xdotool getwindowgeometry "$id" 2>/dev/null | awk '/Geometry/{print $2}')
case "$geo" in 1600x1000|"${WIN_W}x${WIN_H}") continue;; esac
w=${geo%x*}
case "$w" in ''|*[!0-9]*) continue;; esac
[ "$w" -le 50 ] && continue
if [ "$w" -gt "$best_w" ]; then best=$id; best_w=$w; fi
done
[ -n "$best" ] && echo "$best"
}
start_app() { # LOGFILE ENVS... ; extra args via APP_ARGS
local log=$1; shift
# shellcheck disable=SC2086
env "$@" "$app" --server 127.0.0.1 --port "$port" --buffer-mb 16 ${APP_ARGS:-} >"$log" 2>&1 &
app_pid=$!
}
stop_app() {
[ -n "$app_pid" ] && kill "$app_pid" 2>/dev/null || true
wait "$app_pid" 2>/dev/null || true
app_pid=
}
side_of_dump() { # LOGFILE -> the sidebar width the app reports, else the table's 320
local s
s=$(grep -o 'side=[0-9]*x' "$1" 2>/dev/null | head -1 | grep -o '[0-9]*' || true)
echo "${s:-320}"
}
# annotate NAME RAW LOG BOXES W H -> the figure, with a numbered box per line of BOXES.
# A box is `label:x:y:w:h`; `SIDE_X`/`SIDE_W` are the sidebar's own numbers, `EW`/`EH` the
# grabbed window's width and height.
annotate() {
local name=$1 raw=$2 log=$3 boxes=$4 win_w=$5 win_h=$6
local side_w side_x i=0 draw=() line label x y w h
side_w=$(side_of_dump "$log")
side_x=$((win_w - side_w))
while IFS= read -r line; do
[ -n "$line" ] || continue
IFS=: read -r label x y w h <<<"$line"
x=${x/SIDE_X/$side_x}; w=${w/SIDE_W/$side_w}
w=${w/EW/$win_w}; h=${h/EH/$win_h}
i=$((i + 1))
draw+=(-stroke '#e01b24' -strokewidth 3 -fill none
-draw "rectangle $x,$y $((x + w)),$((y + h))"
-stroke none -fill '#e01b24'
-draw "circle $((x + 17)),$((y + 17)) $((x + 17)),$((y + 33))"
-fill white -stroke none -pointsize 22
-draw "text $((x + 10)),$((y + 25)) '$i'")
done <<<"$boxes"
magick "$raw" -crop "${win_w}x${win_h}+0+0" +repage "${draw[@]}" "$out/$name.png"
local sd
sd=$(magick identify -format '%[standard-deviation]' "$out/$name.png")
awk -v s="$sd" 'BEGIN{exit !(s+0 > 0)}' ||
{ echo "make-help-shots: the '$name' figure is blank (stddev=$sd)" >&2; return 1; }
echo "make-help-shots: wrote $out/$name.png ($(magick identify -format '%wx%h' "$out/$name.png"))"
# Print what each number marks, so the script is self-documenting in CI.
local n=0
while IFS= read -r line; do
[ -n "$line" ] || continue
n=$((n + 1))
printf ' %s = %s\n' "$n" "${line%%:*}"
done <<<"$boxes"
}
# grab NAME LOG BOXES [PICKER] -- grab the picked window, fail if it is blank, stop the app,
# annotate it at its own size.
grab() {
local name=$1 log=$2 boxes=$3 picker=${4:-pick_window} win geo win_w win_h raw sd
win=$($picker)
[ -n "$win" ] || { echo "make-help-shots: no window for '$name'" >&2; stop_app; return 1; }
geo=$(xdotool getwindowgeometry "$win" 2>/dev/null | awk '/Geometry/{print $2}')
win_w=${geo%x*}; win_h=${geo#*x}
raw="$tmp/$name.raw.png"
import -window "$win" "$raw"
sd=$(magick identify -format '%[standard-deviation]' "$raw")
awk -v s="$sd" 'BEGIN{exit !(s+0 > 0)}' ||
{ echo "make-help-shots: '$name' grabbed a blank window (stddev=$sd)" >&2; stop_app; return 1; }
stop_app
annotate "$name" "$raw" "$log" "$boxes" "$win_w" "$win_h"
}
# --- scene 1: the live window -------------------------------------------------
echo "make-help-shots: scene 'main-window'"
APP_ARGS="--plots 3" start_app "$tmp/main-window.log" UDPSCOPE_ENABLE_ALL_MS=2500
sleep 8
grab main-window "$tmp/main-window.log" "$boxes_main"
# --- scene 2: a held capture --------------------------------------------------
echo "make-help-shots: scene 'trigger-capture'"
APP_ARGS="--plots 2" start_app "$tmp/trigger-capture.log" \
UDPSCOPE_ENABLE_ALL_MS=1200 UDPSCOPE_ARM_TRIGGER_MS=1500
sleep 9
grab trigger-capture "$tmp/trigger-capture.log" "$boxes_trigger"
# --- scene 3: the math editor -------------------------------------------------
# The editor is a window of its own, presented on the main one, and `math_probe`
# (`UDPSCOPE_PROBE_MATH_MS`) opens it -- "editor opened" in its log is the reading. No clicks
# are needed: the probe presents it and reopens it after a structural edit, so by the time the
# wait ends it is up. It is grabbed on its own rather than through the main window, which is
# the whole point of the figure.
echo "make-help-shots: scene 'math-editor'"
APP_ARGS="--plots 1" start_app "$tmp/math-editor.log" \
UDPSCOPE_ENABLE_ALL_MS=1500 UDPSCOPE_PROBE_MATH_MS=2500
sleep 10
grab math-editor "$tmp/math-editor.log" "$boxes_math" pick_other_window
# The last step, and it is not cosmetic: `import` records the root window's background, so the
# PNG leaves ImageMagick with a `bKGD` chunk -- forbidden by the PNG specification on an image
# with an alpha channel (colour types 4 and 6), which these are. ImageMagick writes it and reads
# it back happily, so a round-trip through `magick` proves nothing; a stricter decoder refuses
# the file. In Yelp that showed as a figure that never loaded, and over the `help:` URI as a page
# that spun forever, because the load event never fired. Normalise every figure to the chunks a
# decoder is guaranteed to accept, whatever the grab produced.
for f in "$out"/*.png; do
magick "$f" -strip -define png:exclude-chunk=bKGD,cHRM,tIME,tEXt,tXMP,exif \
-define png:compression-level=9 "$f"
done
echo "make-help-shots: done"