fix(packaging): Windows relocated LoopEditor --quick-smoke - #532
Draft
mberrys wants to merge 34 commits into
Draft
fix(packaging): Windows relocated LoopEditor --quick-smoke#532mberrys wants to merge 34 commits into
mberrys wants to merge 34 commits into
Conversation
setLibraryPaths() overrides windeployqt's qt.conf beside LoopEditor.exe. Stop setting library paths on Windows, keep Linux AppImage probing, skip Sentry on the --quick-smoke fast path, and log staged stderr markers plus install-tree diagnostics when relocated Windows smoke fails. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Relocated smoke showed qt.conf without Plugins= and no stderr markers, implicating QPA startup before Quick loads. Mirror PdfTool by copying qoffscreen into usr/bin/platforms at install time, append Plugins=plugins to qt.conf, drop attachConsole, and probe --help before --quick-smoke. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Stage markers show the relocated crash happens after QGuiApplication but before runQuickSmoke returns, while --help exits earlier from the parser. Route --quick-smoke through settings migration and translator setup, try the software Quick backend first on Windows, and add logging through EditorHost construction and QML module load. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Relocated Windows packaging smoke crashes in EditorHost construction before Quick QML loads. Stage markers stop after QGuiApplication while --help exits earlier from the parser. The offscreen platform can expose a primaryScreen() that faults when DPI is queried; use 96dpi defaults instead, matching headless smoke on Linux AppImage. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows packaging smoke still faults in EditorHost because the offscreen guard only checked platformName(), which can differ from the requested QT_QPA_PLATFORM value. Treat QT_QPA_PLATFORM=offscreen as headless and add stage logging through parser, settings, translator, and EditorHost. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows relocated packaging smoke faults during EditorHost construction. Always inject default viewport DPI on Windows instead of probing primaryScreen(), and emit stderr markers through DocumentViewSession and EditorHost construction to localize any remaining crash. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Adding usr/bin to QCoreApplication::libraryPaths() on Windows makes Qt treat product DLLs as plugins and faults during DocumentViewSession construction. Use packagedLibraryPaths on all platforms and log member-init stages to localize any remaining relocated smoke crash. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Co-authored-by: michael berry <mberrys@users.noreply.github.com>
The context logging lambda returned PDFDocumentContext by value, which does not compile because the type is non-copyable. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows relocated smoke faults after scheduler_end and before facade_begin. Initialize CommandCatalog and PDFDocumentContext (with proper QObject parents) before PDFJobScheduler worker threads start, and add submitter/catalog/context stage markers for the next CI slice. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Comma-operator member init materialized temporary CommandCatalog and PDFDocumentContext values and copy-constructed them, which MSVC rejects as C2280. Keep catalog and context before PDFJobScheduler and log stages with copy-free InitStage members. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
pwsh does not stop on a nonzero cmake --build, so a C2280 compile error continued into cmake --install and looked like a missing loop-default.json smoke failure. Check LASTEXITCODE after configure, build, install, and surface verification, matching reusable-windows.yml. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
cursor
Bot
changed the title
Windows relocated smoke faults between context_begin and context_end when PDFDocumentContext is parented during DocumentViewSession member init. Construct the empty context without a parent, setParent in body_begin, and lazily create the document session on first getSession() so session setup runs after the view session shell exists. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows relocated smoke reaches facade_end then faults before body_begin while scheduler worker threads are already running during Qt signal wiring. Defer thread creation until the first submit() so DocumentViewSession member initialization completes on the main thread without a background race. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows relocated smoke reaches body_begin and before surfaces, then faults inside PageSurfaceCoordinator construction. Prime the empty snapshot after the view session shell is wired, avoid eager PDFDocumentSession creation while applying startup cache limits, and add surface-stage markers for the next CI slice. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows relocated smoke reaches after surface_ctor then faults inside setPageCacheBudget/refreshPageCacheBudget. Store the shared budget during startup and refresh only after primeInitialSnapshot. Fix clang-format on pdfjobscheduler.h for agent-fast. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows relocated --quick-smoke faulted in setTotal when LoopEditor and LoopLibInteraction called header-inline mutex methods on a budget object owned by LoopLibCore. Export the implementations from a single translation unit like PDFResourceBudget. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Keep trim/rebuild deferred until primeInitialSnapshot while still updating coordinator limits when a shared PDFPageCacheBudget is attached. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows relocated smoke regressed when setPageCacheBudget eagerly read PDFPageCacheBudget totals from LoopLibInteraction during attach. Keep attach pointer-only until primeInitialSnapshot, project limits from static partition math during startup, and route DocumentViewSession cache wiring through setCacheLimit instead of refreshPageCacheBudget. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Drop staged stderr markers and install-tree probes now that relocated Windows_MSI smoke is green. Keep the packaging and initialization fixes. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Relocated Windows smoke still faults on native Quick before software on the packaged tree. Keep software-first ordering as a packaging requirement, not diagnostic scaffolding. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows relocated --quick-smoke still faulted in DocumentViewSession setCacheLimit when LoopEditor and LoopLibInteraction called PDFPageCacheBudget::setTotal across the LoopLibCore DLL boundary. Publish totals via PDFDocumentContext::setPageCacheTotal, defer coordinator budget writes until priming, and derive coordinator bounds from the local cacheLimit authority instead of reading the shared budget from LoopLibInteraction. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Copying std::shared_ptr<PDFPageCacheBudget> from LoopLibCore into PageSurfaceCoordinator in LoopEditor can corrupt refcounts across the DLL boundary and fault during relocated --quick-smoke startup. Attach a non-owning PDFPageCacheBudget* instead and keep temporary stage markers in quick-smoke until Windows_MSI is green. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows relocated smoke faulted inside DocumentViewSession::setCacheLimit before EditorHost finished constructing. Seed DefaultCacheLimit in PDFDocumentContext during member initialization and keep startup on projected coordinator limits until primeInitialSnapshot instead of calling setPageCacheTotal from the view session constructor. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
primeInitialSnapshot faulted in trimCacheToBudget when PageSurfaceCoordinator called PDFPageCacheBudget::pageSurfacesLimit from LoopLibInteraction. Derive surface-byte limits from projected cacheLimit and bounds instead, and track resident usage from coordinator counters. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Windows relocated smoke reached scene_graph_initialized then faulted during Qt/Loop DLL teardown. Use std::_Exit(0) after the startup probe succeeds and install Liberation TTFs under usr/lib/fonts for packaged Qt text. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
PdfTool preflight crashed with 0xC0000005 on relocated Windows installs because executableDirectory was prepended to libraryPaths(), making Qt treat product DLLs as plugins during QPA startup. Use the same packagedLibraryPaths pattern as LoopEditor. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Diagnostic fprintf scaffolding is no longer needed after Windows quick-smoke passed in CI. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Relocated Windows smoke strips developer Qt env vars before PdfTool preflight. Running PreflightEngine inside a PDFJobScheduler worker thread faulted with 0xC0000005 before JSON output was emitted. Execute preflight synchronously on the GUI thread, filter the default exe-dir plugin path on Windows, install qoffscreen beside PdfTool, and fail fast on preflight in the isolated-env smoke block. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Allocate PDFDocumentSession from LoopLibCore, leak it on Windows after preflight completes, and exit with std::_Exit once the JSON envelope is written so relocated smoke no longer faults during DLL teardown. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Preflight still AV'd on Windows after session teardown fixes because readFromFile returned a PDFDocument by value into PdfTool.exe. Add heap read/destroy helpers in LoopLibCore and route preflight through them. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Move file read, session creation, and PreflightEngine::run into inspectPreflightFile so relocated Windows smoke never owns parsed PDF state in PdfTool.exe. Add temporary stage markers for CI diagnosis and fix clang-format drift in pdfdocumentreader.h. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
Preflight constructs a PDFDocumentSession without needing font/CMS/renderer helpers because evidence collection builds its own rendering stack. Eager initializeRendering() in the constructor was crashing on Windows between the after_read and after_session preflight markers. Lazy initialization keeps session construction lightweight for CLI preflight while preserving existing behavior for editor rendering paths. Co-authored-by: michael berry <mberrys@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Fixes Windows relocated-install smoke failures (
0xC0000005) for PdfTool preflight and LoopEditor--quick-smokeduring MSI packaging.Root cause (preflight)
Windows CI stage markers showed PdfTool preflight crashing during
PDFDocumentSessionconstruction — after document read (after_read) but before session creation completed (after_session). The fault was in eagerinitializeRendering(), which creates font/CMS/renderer helpers that preflight does not need (evidence collection builds its own stack).Fix
PDFDocumentSessionno longer callsinitializeRendering()in its constructor. Rendering helpers are created on first access viagetRenderer(),getFontCache(),getCMS(),getOptionalContentActivity(), orcompilePage().inspectPreflightFile()inside LoopLibCore,_Exit(0)teardown bypass, quick-smoke fixes, etc.Verification
UnitTestsPreflightCorpus,UnitTestsPreflightEngine,UnitTestsPdfToolContract— passbleed-adequate.pdfin isolated Qt env — full stage trace throughend, exit 0f188b9d6)Follow-up
Remove temporary
loop-pdftool preflight stage=markers frompreflightclirun.cpponce Windows_MSI is fully green.