plc-user on PR #693: the crash is fixed, but the color/font field and
the on-diagram text no longer update until you leave the properties
list and click in the diagram -- previously it updated as soon as you
clicked OK.
That's a side effect of the crash fix itself. The old, crashing code
returned a *live* QColorDialog as the item view's editor; clicking its
OK button called accept()/hide() on it, and hiding the active editor
happens to trip the base delegate's own focus-lost commit path -- so
the value applied immediately, racily, as a side effect of the same
mechanism that crashed on Enter. The fix (commit 4bd9b6b21) replaced
that with running the dialog synchronously inside createEditor() and
returning an inert placeholder with the result stashed in a property.
Correct for the crash, but it also removed that accidental commit
trigger: the placeholder never had focus to lose, so nothing tells
the view to read the value back until some unrelated interaction
(clicking away) incidentally triggers it.
Fix: explicitly emit commitData()/closeEditor() for the resolved
editor, deferred via QTimer::singleShot(0, ...) since the view only
registers createEditor()'s return value as "the active editor" after
createEditor() itself returns -- emitting synchronously, before
returning, would target a widget the view doesn't know about yet.
Applied to both font and color, since both share the exact same
"resolve synchronously in createEditor(), return an inert
placeholder" shape and thus the exact same gap; font just hadn't been
reported.
Verified with the same standalone harness from the crash fix (real
QTreeView + DynamicTextItemDelegate + QAbstractItemView::edit()),
this time deliberately *not* sending the synthetic Enter keypress the
crash-fix verification needed: clicks the dialog's real OK button,
lets the event loop run, and confirms the picked color lands in the
model on its own. Also reconfirmed the crash fix itself still holds
(clean exit, no synthetic-Enter needed either way now) and did a full
Release build (504/504) with no new warnings.
Bugtracker #323: crash changing a label's color, but only when
confirmed via Enter -- clicking the dialog's own OK button with the
mouse doesn't crash. Reported on Windows 11 and Debian, with
"QObject::installEventFilter(): Cannot filter events for objects in
a different thread" immediately before the segfault.
Root cause: DynamicTextItemDelegate::createEditor()'s color case
constructed a QColorDialog and returned it directly as the item
view's editor widget for the color cell -- unlike every other case
in this same function, which returns a small inline widget
(QSpinBox, QComboBox, or, for the adjacent font case, a plain
placeholder). A QColorDialog is not designed to be used this way: it
is not one of the objectNames this delegate's own eventFilter()
special-cases, so Enter is handled by the base
QStyledItemDelegate::eventFilter() as an ordinary "commit and
destroy this small editor" trigger -- racing the dialog's own
internal OK-button accept/close path, which on Windows can hand off
to the native color picker. Clicking OK with the mouse doesn't go
through the same key-press path, which is why only Enter crashed.
Verified structurally: an embedded QColorDialog editor is a *child*
widget of the view's viewport rather than a proper top-level dialog
(confirmed with a standalone Qt program driving the real delegate
through QAbstractItemView::edit() -- searching QApplication's
top-level widgets never found it, only a search of the viewport's
children did), which is the same "used as something it isn't"
pattern, just observed a different way.
Fix: mirror the font case immediately above -- resolve the color via
the static, blocking QColorDialog::getColor() inside createEditor(),
and hand back a plain QWidget with the result stashed in two
properties (mirroring the font case's "ok" property) for
setModelData() to read. By the time the view processes any commit
trigger, the "editor" is an inert placeholder with no dialog state
left to race.
Verified end-to-end with the same standalone program: creates the
model item, triggers editing, finds the real (top-level, this time)
QColorDialog, clicks its actual OK button, confirms the color lands
on the placeholder's properties, sends the editor a synthetic Enter
keypress (the exact trigger from the bug report), and confirms the
final committed value in the model matches the picked color. Also
confirmed a full Release build (333/333) with no new warnings.
NewDiagramPage::applyConf() writes every other default to QSettings —
border, title block, conductors, folio reports and the guides — but the
cross-reference branch only fetched the properties into a local hash and
then dropped it on the floor. hash_xrp was never used.
The result: changing the cross-reference defaults under Settings > New
project has no effect. Nothing is written, no defaultxref* key ever
appears in the configuration file, and XRefProperties::defaultProperties()
keeps handing out the hardcoded fallbacks for every new project.
Write each of the four types (coil, protection, commutator, plc) with the
"diagrameditor/defaultxref" + key prefix that defaultProperties() already
reads back.
Building QET is dominated by re-parsing Qt's headers. A 214-line source
file expands to roughly 198,000 preprocessed lines, and compiling one
translation unit costs ~4.1 s, of which only ~0.35 s is optimisation --
switching -O3 to -O0 saves just 8%, so the usual "build Debug for faster
compiles" advice does not help here. A precompiled header caches the
parsed header state, which is the part that actually costs.
Measured on a 24-thread Xeon E5-2650 v4 with Qt 5.15.18 and GCC 15.2,
same build tree, only the option differing:
compile one translation unit 4.12 s -> 1.21 s
edit one .cpp -> linked binary 5.22 s -> 1.65 s
Deliberately OFF by default. A PCH satisfies includes that a source file
neglected to make for itself, so code written with it enabled can fail to
compile for everyone else. Leaving the default off keeps CI and
contributors on the strict behaviour; only developers who opt in trade
that away for the speed.
Two details in the implementation are load-bearing:
- The generator expressions are not decoration. This target also compiles
the 18 C files of the bundled LZMA decoder, and an unguarded header list
applies to every language in the target, so the Qt headers would be fed
to the C compiler and fail with "unknown type name 'namespace'".
$<ANGLE-R> is needed because a literal '>' would end the generator
expression.
- target_precompile_headers() requires CMake 3.16 while the project still
declares a 3.5 minimum, so the block warns and skips rather than raising
the project-wide requirement for an opt-in developer feature.
Verified both ways: with the option off no PCH artefacts are generated and
the build is byte-for-byte the previous behaviour; with it on, all 18 C
files still compile, the generated PCH is C++-only (cmake_pch.hxx, with no
cmake_pch.h), the C compile commands carry no PCH, and the resulting
binary runs.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Both actions already exist and work; they were simply only reachable from
a toolbar, and those toolbars are user-hideable via Configuration >
Afficher, so hiding one made the feature unreachable entirely.
- "Afficher les guides" (m_draw_guides) goes into the Affichage menu next
to "Afficher la grille". The two are adjacent lines in the view toolbar
and do the same kind of thing, but only the grid had a menu entry.
- "Creation automatique de conducteur(s)" (m_auto_conductor) goes into the
Projet menu. It writes a project setting via
QETProject::setAutoConductor(), so the Projet menu is where a user would
look for it; it is placed with the project properties, above a separator
that keeps the folio operations grouped as before.
Also removes conductor_default and m_project_folio_list from the header.
Both are declared but never allocated and never referenced anywhere in the
tree -- that the build still links is the proof they were dead.
No new strings: both actions already carry translated text.
Found while auditing every QAction against every menu, discussion #677.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
QetLogger (discussion #644, steps 1-3) captures whatever an explicit
qDebug()/qInfo()/qWarning() call already decided to report. Most of a
session -- painting, dragging, a slow synchronous operation -- produces
no log output at all, so a silent multi-second gap in the log is
indistinguishable from the user simply not doing anything. That gap
came up directly: investigating a user-reported "the program lagged"
required inferring stalls from timestamp gaps between unrelated log
lines, which can't tell a real freeze apart from normal idle time.
EventLoopWatchdog closes that gap directly instead of inferring it. A
QTimer::PreciseTimer repeating tick (every 50ms) measures the *actual*
elapsed time since the previous tick via QElapsedTimer (monotonic,
unaffected by system clock/NTP adjustments). Qt does not queue up
missed fires for a normal repeating timer, so if the main thread is
blocked for 600ms, the timer fires once as soon as the loop frees up,
with ~600ms measured since the last tick -- that gap is the stall,
measured at its source. Only logs (via the existing qWarning() path,
so it reuses QetLogger's file/ring/rotation with no new plumbing) when
a tick is late by more than 200ms, so a healthy session produces zero
output from this class, in keeping with QetLogger's bounded-log design.
Same QET_WATCHDOG_DISABLE=1 escape-hatch convention as QetLogger's own
QET_LOG_DISABLE=1.
Deliberately not included: attributing a stall to what caused it. This
tells you a stall happened and how long -- pairing that timestamp with
gdb attached to a running session (as used for the CLI hang, PR #661)
is still how you get from "it stalled" to a root cause.
Stacked on #647 (feature-diagnostic-logging-crash) for QetLogger/
qWarning() plumbing this depends on -- diff includes its commits until
that merges.
Verified against the compiled binary, not just read: temporarily
injected a QThread::msleep(600) via a one-shot QTimer 2s after
startup, confirmed the exact expected warning
("EventLoopWatchdog: main thread stalled for 620 ms") at the right
severity through the real qWarning()/QetLogger path, then removed the
test hook and reconfirmed a normal run produces no output from this
class at all.
Stacked on the steps 1-3 branch (feature-diagnostic-logging, PR #646).
Kept as its own PR rather than folded into that one, matching the
discussion's own framing: step 4 is explicitly "the highest-risk piece
... lands last, behind its own switch."
## Step 4 -- crash-time ring flush (CrashHandler)
Installs a handler for SIGSEGV/SIGABRT/SIGBUS/SIGFPE/SIGILL (POSIX) /
SetUnhandledExceptionFilter (Windows) that flushes the in-memory ring to
a fixed crash_dump.log before the process dies.
This required reworking LogRing (step 3) to be genuinely lock-free, not
just mutex-protected: a signal handler that blocks on a lock the
crashing thread (or another thread) already holds turns a clean crash
into a hang -- no ring dump *and* no core dump, worse than doing
nothing. append() now claims a slot with a single atomic fetch-add;
dumpToFd() reads the preallocated entries directly and writes them with
write(2) only, looping on EINTR/short writes. Accepted tradeoff: at most
one entry can be read torn if a crash lands mid-append into that exact
slot -- documented in logring.h, and the alternative (a seqlock to
detect and retry) wasn't judged worth the complexity for that window.
Other invariants implemented per the discussion:
- sigaltstack with a static 64 KiB buffer, SA_ONSTACK -- a stack-
overflow SIGSEGV has no usable stack for a handler without one.
- Nothing under the actual handler touches Qt, QString or the
allocator: the dump path and a small header (version/git/OS/Qt) are
precomputed into fixed char buffers by install(), which runs once at
startup in normal context.
- Atomic test-and-set so only the first crash writes a dump; a second
concurrent/nested fault goes straight to restore-and-re-raise.
- After writing, the handler restores SIG_DFL and re-raises (POSIX) /
returns EXCEPTION_CONTINUE_SEARCH (Windows) so the OS's own crash
path -- core dump, Windows Error Reporting -- still runs. A handler
that "fixed" the crash by swallowing the signal would destroy exactly
the post-mortem evidence this whole design exists to preserve.
Tested in this environment: POSIX/Linux only, all five signals. Sent
each directly to a running process and confirmed (a) crash_dump.log is
written with the correct header and ring contents, mode 0600, and (b)
the process still terminates via the signal with the kernel's own
"core dumped" flag set (exit code 128+signal, confirmed for all five).
The Windows path is implemented per the discussion's guidance but is
untested -- no Windows build available in this sandbox.
## Step 5 -- getting the data back out
- QETApp::checkCrashDump(), called from checkBackupFiles() only when
there's no stale project file to recover this run (so the two
prompts never both show, per the discussion), offers an unretrieved
crash dump via DiagnosticsReportDialog and then deletes it regardless
of the user's choice -- offered exactly once.
- A new "Aide > Enregistrer un rapport de diagnostic..." action
(QETMainWindow) builds the same kind of report from the *current*
session (QetLogger::buildDiagnosticsReport(): header + this session's
log file) for a manual "attach this to a bug report" flow, not tied
to a crash.
- Both go through QetLogger::redact() before ever reaching the user:
the one redaction implemented is a literal replace of the home
directory with "~", since an absolute path under it leaks the
account name. The discussion's fancier "optionally redact project
filenames too" isn't attempted -- reliably telling a project path
apart from arbitrary log text is a much fuzzier problem than a
literal prefix match.
- DiagnosticsReportDialog shows the full (already-redacted) content
before saving, per the discussion: "the user is about to attach this
to a public tracker."
Verified in a real GUI session (Xvfb): triggered a SIGSEGV, relaunched,
confirmed the crash-report dialog appears with the right header/content,
confirmed it does not reappear on a second relaunch, and confirmed the
manual "Save report" action produces a correctly-formatted report and
saves it to a chosen path.
Built clean, no new warnings.
## Build systems
Registered in both: cmake/qet_compilation_vars.cmake, and
qelectrotech.pro. The .pro needed explicit globs for the new
sources/logging/ui/ subfolder -- sources/logging/*.{h,cpp} was already
globbed, but unlike the other ui/ subfolders that one had no entry of
its own, so diagnosticsreportdialog.{h,cpp} would not have been built
under qmake.
The job trigger was narrowed to push:tags a while back, but the
job-level 'if: github.ref == refs/heads/master' was left in place.
A tag push never has github.ref == refs/heads/master, so the two
conditions are mutually exclusive: the job trigger fires only on
tag pushes, while the guard only allows master-branch refs, meaning
the job has been silently skipped on every run since the trigger
was narrowed.
This is also the likely source of the recent Git LFS bandwidth/
storage overage: while the trigger was still push-to-master (pre-
narrowing), this job ran on nearly every commit and committed a
new QElectroTech.qch (LFS-tracked) each time, via the auto-generated
update-qch PR -- accumulating one LFS object version per run."
ProjectView::initWidgets() called insertSpacing(1, 10) on a QHBoxLayout
that was still empty, inserting past the end of the item list. The
corrupted layout crashed later in QWidget::setLayout() via
QLayoutPrivate::reparentChildWidgets() and QBoxLayout::itemAt().
Use addSpacing(10) instead, which is equivalent for an empty layout.
Never leave a collection folder without a name (replaces #622)
Mark elements-folder with an exclamation-mark, when "qet_directory" is missing or faulty.
The unconditional early return was narrowed to non-directories only, so
that the just-added warning badge could be picked up once setUpData()
resolved m_qet_directory_unreadable asynchronously. But every directory
then called setIcon() on every single data(Qt::DecorationRole) query --
not just once -- and QStandardItem::setIcon() -> setData() emits
dataChanged() unconditionally (QIcon has no equality check to suppress
it). QTreeView handles dataChanged() by recomputing the row's size hint,
which re-enters data() for the same index, calling setIcon() again:
unbounded mutual recursion, confirmed by an isolated reproduction to
overflow the stack in a single frame (100k+ frames) well before the
first paint completes. Matches plc-user's report of a segfault right as
the elements tree begins drawing.
The race the guard was widened for doesn't actually occur:
ElementsCollectionModel only attaches itself to the tree view (the only
way data() becomes reachable) from loadingFinished(), which fires after
the QtConcurrent::map over every item -- this one included -- has
already finished. m_qet_directory_unreadable is therefore always final
before setUpIcon() can run for the first time, so the plain, always-only-
once guard is sufficient and the badge still works correctly.
"Add folio" always appends to the end of the project, ignoring
whatever folio is currently selected in the left panel -- even though
the panel already tracks the selected diagram's position for its
existing move up/down/top actions, and QETProject::addNewDiagram(pos)
already accepts an arbitrary insertion index, pushed as an undoable
AddDiagramCommand (QetGraphicsTableFactory::create() already relies on
this exact mechanism to insert a folio right after a specific one).
Add two new context-menu actions that compute the target position from
the selected diagram's folioIndex() and pass it straight through the
existing machinery -- no changes needed to QETProject or
AddDiagramCommand. New requestForNewDiagramAt/addDiagramToProjectAt
signal/slot pair added alongside the existing
requestForNewDiagram/addDiagramToProject rather than changing it, so
the plain "Add folio" action's append-at-end behavior is untouched.
ispyisail