Driving caf::ProgressInfo from worker threads inside the OpenMP loop in
loadFileSummaryCaseData() marshaled setProgress() to the GUI thread via a
queued connection. These updates could be delivered after the ProgressInfo
scope had ended and the progress stacks were popped, accessing an empty stack
and resurrecting the just-closed dialog stuck at 0% with no way to close it.
Only update the progress dialog from the master thread, using an atomic
counter for a monotonic progress value, and guard the ProgressInfo mutators
against stale updates delivered after the stacks are empty.
The vcpkg binary cache was restored on every Windows run, but vcpkg rejected all binaries and rebuilt from source. vcpkg keys each binary by an ABI hash that includes the MSVC toolset version, which was not part of the cache key (only OS, compiler name, vcpkg submodule SHA and the vcpkg-configuration.json hash). When the runner image bumps MSVC, the restored binaries become ABI-incompatible and vcpkg rebuilds everything. Because the cache key stays unchanged, the action skips saving the freshly built binaries, so the stale entry can never refresh and the rebuild loop persists across runs.
Add the MSVC toolset version (VCToolsVersion, exported by the MSVC setup step) to the cache key on Windows. A toolset bump now yields a new key, producing one cold rebuild followed by a successful save under the new key, after which subsequent runs restore from cache.
reportCrash pumps a nested Qt event loop from within the signal handler to
let the async crash-report HTTP POST finish. Pumping all events also
dispatches buffered mouse/keyboard events, which re-enter arbitrary UI code
mid-crash and can re-trigger the faulting path, causing a double-fault
inside the handler. Pump with QEventLoop::ExcludeUserInputEvents so socket
and timer events still flow (the reply completes) while user input is held
back.
Fixes#14312
The polygon field was added to the UI ordering by value rather than by
address, so an empty (null) polygon reference resolved to the
PdmObjectHandle overload of PdmUiOrdering::add and aborted on the null
assertion while building the property panel. Pass the field by address so
the field-handle overload is used, matching the other add calls.
Fixes#14311
A SMSPEC file without a UNSMRY data file is re-read on every request for a summary vector, since RifReaderEclipseSummary::open returns false and no reader is cached. For an ensemble with multiple realizations missing UNSMRY, this is a severe performance problem.
Guard RifReaderEclipseSummary::open so that when no ESMRY, h5 or resdata data file is present, it returns true and creates a valid but empty reader. Skip ESMRY conversion when no UNSMRY is present, and add RifEclipseSummaryTools::hasSummaryDataFiles to detect the missing data file.
When a summary vector is selected by manipulation of a curve, recompute the
source stepping dimension from the current curve content and refresh the
stepping toolbar. This keeps the toolbar consistent with the displayed vectors,
which is required for 3D view well selection to step the plot correctly.
ecl_file_iget_named_kw() performs an unchecked vector access in ERT and
crashes (SIGSEGV) when the requested keyword does not exist in the active
file view. findAvailablePhases, readUnitsType and isExportedFromIntersect
called it for INTEHEAD without first checking the keyword is present. Guard
each call with ecl_file_get_num_named_kw(), matching the existing pattern in
this file.
QwtPlot::axisWidget() returns NULL for an axis id that is not valid for a
given plot. RiuMultiPlotPage::alignAxis dereferenced the returned
QwtScaleWidget without a null check, causing a segmentation fault when a
plot in the grid does not have the queried axis. Guard the pointer in both
alignment loops, matching the existing null-check style in the function.
Add a std::set_terminate handler (manageTerminate) that runs before the
stack unwinds when a C++ exception escapes the Qt event loop, so the
captured stack trace points at the original throw site instead of the
crash-logging plumbing reached via the SIGABRT signal handler. The
handler also recovers the exception type and message via
std::current_exception and reports them as crash attributes.
Remove the catch-and-rethrow around the event loop in RiaMain: that
handler unwound the stack to main before anything captured it, which
defeated the new terminate handler. The SIGABRT signal handler remains
as a fallback for direct abort()/assert() failures that do not go
through std::terminate.
The grid statistics plot and show commands were only added to the context menu when the cursor was inside the overlay info box, so right-clicking on the grid geometry no longer offered them. Always add the create and show commands, and keep only the copy-to-clipboard command gated by the info box position.
The file browser opened by Import Summary Case shared the general Eclipse
import dialog, which hardcoded the title to "Import Grid Model". Use
"Import Summary Case" when importing a summary file.
On a tag push, wait for the Windows/Linux build (ResInsightWithCache.yml)
and the macOS build (ResInsightMac.yml) to finish, download their
application artifacts, zip them per platform, and create a draft GitHub
Release with the archives attached for maintainer review.
The GUI thread deadlocked on a non-recursive QMutex during OSDU well
import. The download-completion slots and DataLoadController::onTaskFinished
held a mutex while triggering progress updates, which call
QApplication::processEvents(). That re-enters the event loop and dispatches
another queued parquet-download-complete event for a different well, whose
handler then blocks trying to lock the same mutex.
Release the mutex before emitting taskDone in the OSDU well path and well
log data loaders, and before calling incrementProgress() in onTaskFinished.
The map access that the mutex protects stays inside the locked scope; the
captured ProgressInfo outlives the call (kept alive by blockUntilDone) and
incrementProgress() is thread-safe via self-marshaling to the GUI thread.
Fixes#14230.
Guard against a null case pointer in findSummaryCaseFromEclipseResultCase. The reload path passes dynamic_cast<RimEclipseResultCase*> of the case, which is null for Roff and GRDECL input cases, causing a null dereference in RimCase::gridFileName().
Reject null buffers and negative buffer sizes up front, returning false instead of risking a null dereference or hitting util_abort. The buffer-overflow protection from the upstream fix is already present via #14144; these two input-validation guards are the remaining delta.
Human-gated workflow that takes an unresolved crash signature from the
resinsight-system-doc registry, locates the real crash site in source,
proposes and build-verifies a minimal fix, then on approval files the OPM
issue, pushes a fix branch to the magnesj fork, opens the PR, and records
the outcome back on the signature. Uncertain cases get an explanatory note
instead of an issue or PR. This is step 3 of the weekly crash-triage flow.
Magne Sjaastad
Kristian Bendiksen