- Rewrite README.md around new_docker_build_release2.sh: Release 2
scope (4 build groups, 61 targets), Ninja/RelWithDebInfo configure,
the 22-test ctest baseline, DEB production via
rpmbuild/make_rpm_ubuntu26_release2.py, options reference,
troubleshooting and known quirks (the legacy validation-scope driver
is no longer documented as the main flow).
- Remove leftover blank lines and stale explanatory comments in
Lts/DDS/Generic (dynamic_data_reader.h, generic_base.h),
Lts/DDS/QosProfile (DDS_DefaultQos.cpp/.h, CMakeLists.txt) and
Lts/Plugins/CMakeLists.txt.
Trim mostly-deletion changes (~700 lines) across the modules the Release
2 build compiles: unused telegrams and receiver/handler code in
TrainingAreaBase (telegrams, recv_from_dds_aes/fts, ism/main/pdm/position
handlers), dead members in SqlBrowser and ProjectDataManagement,
unused LtsPluginManager loader paths, LtsSession unit-test, WlsTetra and
DdsView cleanups, plus the stale include_directories removal in
Lts/CMakeLists.txt. Continues the Release 2 source prune (keep only what
the build compiles).
Excluded from this commit by design: the machine-specific .envrc
(MIGRATE_PIPELINE_ROOT path fix stays uncommitted) and the untracked
qt5-before/ reference tree.
GrantleeParserSmoke parses the codegen .grantlee templates with the KF6
TextTemplate engine, which loads the ktexttemplate_defaulttags plugin.
The plugin ships in the system multiarch plugin dir
(/usr/lib/x86_64-linux-gnu/qt6/plugins) via libkf6texttemplate-dev, which
the aqt-installed /opt/Qt does not search by default, so every template
failed with "Plugin library not found". Prepend the dir to
QT_PLUGIN_PATH in the driver ctest step (same pattern as
QT_QPA_PLATFORM=offscreen).
Also includes the pending comment-only cleanup of this file (explanatory
comment blocks removed; the only functional change is the QT_PLUGIN_PATH
export line).
get8Byte round-trips the bits correctly (the uint64_t path for the same
value passed), but QCOMPARE of the signed int64_t result against the
unsigned literal 0x8234567812345678 fails in QtTest even though the bit
patterns are identical (top bit set). Compare against the explicitly
signed form instead: QCOMPARE(i, static_cast<int64_t>(0x8234567812345678ULL)).
utils_test now passes 37/37.
Follow-up to the FileManagement cleanup: an audit of every
include_directories() in the tree resolved each path against the source
and a fully-built build dir. The pruned Lts/Viewer (SituationMap,
LtsDataView, Symbology), the LtsUtilsGui/ConnectDialog codegen dir, the
pruned Messaging/OperatorManagement services, and the plugin-level
AudioVideoManagement/GuezNasat/TopicEditorBase/OperatorManagement/LtsDataView
entries all pointed at paths that do not exist, so each one only emitted
-Wmissing-include-dirs warnings. A missing -I dir cannot provide any
header, so the removals are functionally neutral. The audit now reports
zero missing include dirs.
The root CMakeLists still listed include dirs for services removed during
the Release 2 prune (FileManagement, InfrastructureManagement, AssetManagement,
AudioVideoManagement, ExerciseEvent/Management, NotificationManagement,
ParticipantManagement, ReportManagement, SituationReplay, Evaluations) and for
the pruned GuezNasat proxy. None of the referenced source or build paths
exist anymore, so every entry only produced
"cc1plus: warning: ... No such file or directory [-Wmissing-include-dirs]"
on every translation unit. A missing -I dir cannot provide any header, so the
removal is functionally neutral.
Running the driver with only --docker-image silently ignored the flag
(USE_PREBUILT_IMAGE stayed OFF), so ubuntu:26.04 was provisioned from the
in-script apt list - which lacks libbz2-dev - and the configure failed
with "Could NOT find BZip2" in Z/src/nglcompression.
- --docker-image now implies USE_PREBUILT_IMAGE=ON (the usage text already
documented it as the image to use); --no-prebuilt-image passed after it
still forces the base image.
- Add libbz2-dev (BZip2), libgl1-mesa-dev (OpenGL) and pkg-config
(FindGrantlee5/FindPopplerQt6) to the apt list of both the non-prebuilt
flow and the prebuilt-image Dockerfile.
The driver only built the per-group library/tool targets, so 15 of the 22
ctest-registered tests were never compiled and ctest reported them as
'Not Run' (32% pass rate). Add the missing test executables to the group
that owns their module (DDS tests to GROUP1, Lts/Common tests to GROUP3,
LogViewer/SqlBrowser tests to GROUP5).
NGL_ADD_SUBDIR_GUARDED(translations) after Lts (same position as
story/release-validation-scope): translations/CMakeLists.txt reads the
global NGL_TS_TARGETS property accumulated by the NGL macros, so it must
be added last to see every target.
BUILD_TS_FILES is OFF by default (cmake/NGLMacros.cmake), so normal
builds are unaffected; with -DBUILD_TS_FILES=ON the dir generates the
lupdate .pri files and the update_ts_{en_US,en_GB,de_DE} custom targets.
This commit also carries a comment-only cleanup of the file that was
already in the working tree (explanatory comment blocks removed; no
functional line changed - the RPATH $ORIGIN entries, the test
relocation block and the guarded subdirs are all intact). The removed
comments are recoverable from the previous revision if wanted.
Verified with a throwaway BUILD_TS_FILES=ON configure: 0 errors, all
three update_ts_* targets registered, 63 lupdate .pri files generated.
The TDS_INSTALL_TRANSLATION (.qm) calls stay commented as in the source.
Test binaries (naming: test_* | *_test | *_smoke | *Tests - 26 targets
incl. the ctest-registered ones and pdmTests) now build into
<build>/bin/unit-test/ instead of <build>/bin/.
- CMakeLists: post-processing pass after all add_subdirectory calls sets
RUNTIME_OUTPUT_DIRECTORY (overrides the legacy EXECUTABLE_OUTPUT_PATH)
plus a 5-level $ORIGIN BUILD_RPATH entry (from bin/unit-test the RTI
libs are one level further away than from bin/). Targets are collected
by walking the SUBDIRECTORIES tree - the global TARGETS property is
empty under CMake 4.x, which silently broke the first attempt.
- add_test() uses $<TARGET_FILE:...>, so the ctest registry follows the
new location without changes.
- rpmbuild/make_rpm_ubuntu26*.py: test discovery points at
build/bin/unit-test; the release2 runtime DEB mirrors the layout into
usr/bin/unit-test/ so the test binaries stay in the package.
- README: bin/unit-test/ documented in the output list and snippets.
Verified (fresh /workspace-mount build): all 26 test executables in
bin/unit-test/, ctest 20/22 (same 2 pre-existing out-of-scope failures),
RUNPATH carries the 5-level $ORIGIN RTI entry and ddscore_unit_test runs
from bin/unit-test with no LD_LIBRARY_PATH / RTI_LICENSE; app binaries in
bin/ keep their 4-level entry.
The build directory is now a fixed 'build' next to the tree
(qt6-after/build), not in-tree build4 / build4_<timestamp>_<pid> dirs.
- CMakeLists: the $ORIGIN RTI RUNPATH entry is now 4 levels up from a
bin/ or lib/ artifact (the build tree is one level shallower)
- new_docker_build_release2.sh: BUILD_DIR is the fixed ../build; the DEB
tar drops the ./build4* exclude (the build dir is outside the tree)
- new_docker_build.sh: Docker mounts the tree's parent so ../build lands
next to the tree on the host; in-container paths (cmake source,
SRCROOT, deb_stage, rpms, staging tar) updated accordingly
- rpmbuild/make_rpm_ubuntu26*.py: stale build4/ references now say build/
- README: build4 -> ../build in every snippet and doc line
Verified: fresh full-scope configure + ninja all in qt6-after/build is
green; ctest 20/22 (same 2 pre-existing out-of-scope failures as the
previous baseline); RUNPATH carries the 4-level $ORIGIN entry and
ddscore_unit_test runs with no LD_LIBRARY_PATH / RTI_LICENSE.
The RTI Connext libraries live outside the tree (pipeline-root
external/rti_connext_dds-7.7.0). CMake bakes their absolute path into
the RUNPATH, which only resolves on the build host's mount - running
the built binaries under another mount (e.g. /workspace) failed with
'libnddsc.so: cannot open shared object file'.
Add a $ORIGIN-relative RUNPATH entry (5 levels up from any bin/ or
lib/ artifact is the pipeline root) so the RTI libraries resolve
regardless of the mount point. No LD_LIBRARY_PATH needed anymore.
Verified: ddscore_unit_test runs with no env vars set.
- LogView: take the complete XmlPatterns (d327c66c4) file set - the
auto-merged mix (XmlPatterns .cpp + validation .hpp) left duplicated
m_logger members and constructor signature mismatches, because the
two lineages reworked the module differently.
- LogView CMakeLists: the cleanup version (separate LogViewLogger
static library, required by tabPluginTetraJson and present in the
Release 2 target list).
- LtsUtilsGuiDds: link ddscore (BUILD_RTI) for the Core::FunctionReply
symbols, as in story/release-2-cleanup.
Full-scope ninja all is green; ctest 20/22 (the two failures,
GrantleeParserSmoke and utils_test, are out of the driver's 46-target
scope and 'Not Run' in the official baseline).
- Lts/CMakeLists.txt: restore the cleanup scope (Tools, DDS, Common,
Services, Proxies, Plugins) - the validation version configured only
Lts/Common and TrainingAreaBase, silently dropping the GROUP1 DDS
targets, the GROUP5 tools and the service libraries.
- 11 further scope CMakeLists from cleanup: the validation versions had
removed the per-module unit-test subdirectories (ctest baseline
targets), LtsUtilsGui, and used a stale GeographicLib link style.
- PrtCnv: keep the validation CMAKE_CURRENT_BINARY_DIR include
(generated prtcnvlib_export.h).
- sPDM executable: guard on Lts/Common/LtsWebSocket (pruned, link
orphan) like the sTAB executable.
The merge from story/release-validation-scope only carried files that
existed at the common base or on that branch. In-scope files added on
story/release-2-cleanup after the base (per-module unit-test dirs, the
Release 2 driver and rpmbuild script) were missing, breaking
ADD_SUBDIRECTORY(unit-test) references.
- Restore the 33 missing in-scope files from story/release-2-cleanup.
- Lts/Common/CMakeLists.txt and LtsUtilsGui/CMakeLists.txt: take the
story/release-2-cleanup versions (the validation versions still
reference out-of-scope subdirectories: httq, LtsRestClient,
LtsWebSocket, ConnectDialog, StyleManager).
Bring the validation-scope line (in-tree Jenkins CI copy, lcov coverage
wiring, step-by-step release-scope README, TetraIF unit tests, the
LogView runtime icon/resource fix, Dockerfile, rpmbuild and the
new_docker_build.sh driver with the DEB output path) on top of the
XmlPatterns branch pruned to the Release 2 scope.
Conflict policy:
- Out-of-scope targets/modules: keep the prune (deletion wins).
- In-scope files that validation had deleted: restored from
story/release-2-cleanup (the Release 2 closure, with the
pruned-module guards).
- In-scope sources carrying migration work: XmlPatterns versions
(Qt XmlPatterns rework on QDomDocument, DDS member and logger
fixes, first-build fixes).
- Scope-aware CMakeLists (top level, PluginInterfaces, LtsWidgets)
and .envrc: story/release-2-cleanup / ours versions.
Delete everything that is not part of the Release 2 build closure
defined by the story/release-2-cleanup README (46 targets,
GROUP1/3/5/6): out-of-scope Lts modules (Viewer, OSG/BuildingViewer
floor, Schusstafel, MSDL/SAAB import tools, ...), the include/ stub
surplus, tests/, specs/, translations/, the migration report files,
cmake/FindOpenSceneGraph.cmake, external/Schusstafel and the legacy
new_docker_build.sh driver.
In-scope files keep the XmlPatterns-branch versions (Qt XmlPatterns
rework on QDomDocument, BZip2 nglcompression, first-build fixes).
Tracked file count: 16140 -> 2166.
- QT5_ADD_RESOURCES was a no-op stub under Qt6, silently dropping .qrc
files: LogView's :/image/* icons were never compiled into the binary.
Map the stub onto qt6_add_resources() (outfiles-variable form, which
appends the generated qrc_*.cpp to the source list). LogView is the
only caller of the stub.
- nglwidgets' Qt6 branch only processed rme_light_style.qrc;
nglresources.qrc (which provides :icons/16x16/* toolbar/window icons)
was left as a raw file in the source list and never compiled
(CMAKE_AUTORCC is off). Process both qrcs.
- LogView never loads libnglwidgets.so at runtime (--as-needed drops it:
the code uses no nglwidgets symbols), so its :icons/16x16 resources
were never registered for the app. Embed the 20 icons LogView actually
uses directly into LogViewResources.qrc with explicit aliases (rcc
rejects XML comments in .qrc and, without an alias, registers the full
relative path as the resource name).
- new_docker_build.sh: new --run-coverage mode (mutually exclusive with
--run-jenkins-builder). Instruments the release-scope build with
-O0 -g --coverage -fPIC -DNDEBUG (RELEASE flag override keeps -DNDEBUG
and -fPIC so behavior and PIC match the green baseline), runs ctest,
then collects the lcov report: capture --external -> extract on the
four release dirs -> remove test sources -> summary -> genhtml, all
with --ignore-errors mismatch,inconsistent,unused (GCC 15 + lcov 2.4).
Report lands in --coverage-output-dir (default ./coverage-out):
html/index.html, coverage.info, coverage_summary.txt.
- Dockerfile image gains lcov via APT_PACKAGES (genhtml ships in the
same package).
- Jenkinsfile: RunCoverage parameter (default on) + per-OS Coverage
stage running the driver on the agent and archiving
coverage/<project>/<os>/**.
- release-16-10-2026.sh: same coverage step, gated by DRYRUN_COVERAGE
(default ON, mirrors the Jenkins default).
- README: Step 6 (code coverage) and options/env reference entries.
- .gitignore: coverage-out/
Scope is now described only as what the release build contains
(TetraIF, Z/src, Lts/Common, TrainingAreaBase). All steps (image,
compile, tests, DEB packaging, Jenkins) kept and refreshed; the
license note is reduced to 'no license file is required'.
- Drop the 'Stub file ... BUILD_RTI=OFF / Auto-generated' headers from the
tab_* and pdm_* stub files (RTI is gone on this branch; the names say it)
- Drop stub comments from MockCoordTrans.h and the coordTransTests placeholders
- Drop the DDS exclusion list comment from Lts/Common/CMakeLists.txt
- Shorten the staging and packages-out comments in new_docker_build.sh
- Merge build_single_deb and build_single_deb_nobuild into one build_deb()
- Add build_nobuild_package() factory for the 4 no-build packages
- Extract run(), prune_dirs() and write_deb_file() helpers
- Drop the unreachable release_scope copy branch (all call sites used the
full-tree copy; the --release_scope CLI flag is kept)
- Trim pipeline_files to the 8 files that actually exist and are packaged
- Remove unused imports (configparser, getpass, re, datetime, Path, stat),
unused vars (test_patterns, script_extensions) and a duplicated write block
Audited the release-scope compilation (TetraIF, Z/src, Lts/Common +
TrainingAreaBase, BUILD_RTI=OFF) for every file under include/:
Removed (zero references from release-scope sources or CMake):
- service-stub header dirs (AlarmManagement, AreaEffectSimulation,
AssetManagement, AudioVideoManagement, Core, ExerciseEventManagement,
ExerciseManagement, GuezNasat, InfrastructureManagement, Lts,
LtsProjectionWorkstation, LtsScriptTimeline, LtsSession,
MarkerManagement, NotificationManagement, OperatorManagement,
ParticipantManagement, ProjectDataManagement, ReportManagement,
Services, SituationReplay, TopicEditorBase, Viewer)
- include/grantlee (codegen modules excluded from the release build)
- include/nglapplication (all headers duplicate Z/src/nglapplication)
- include/poppler (Qt5 header; the build includes poppler/qt6 from the
system package)
- include/rti (142 Connext DDS headers; only consumers are behind
BUILD_RTI, which is OFF for this branch)
- TetraIF/LogView zsksxd_tetraForLogView{,_guez}.cpp and
zsksxd_tetraForLogView_guez.x (the build compiles the .x; the guez
variant is referenced nowhere; LogView loads the .x from the app dir)
Kept (actively included): include/GeographicLib, include/LtsUtilsGuiDds,
include/NGLLogger.
Also drop the now-dangling include_directories for the removed dirs in
the top-level CMakeLists (include/Viewer/SituationMap) and
Lts/CMakeLists.txt (six stub dirs) - otherwise GCC warns
-Wmissing-include-dirs for every compile.
Verified with a full pipeline rebuild (scope-rebuild15): 0 warnings,
ctest baseline unchanged, all 5 DEBs produced.
The release validation scope (TetraIF, Z/src, Lts/Common +
TrainingAreaBase) never consumes an RTI license: no CMake file or
build step reads RTI_LICENSE_FILE, and the only RTI Connext DDS
consumers are all behind BUILD_RTI (OFF for this branch, stubs
compiled instead). The license was pure plumbing:
- new_docker_build.sh: remove the require_file validation, the
bind mount and the in-container RTI_LICENSE_FILE export
- jenkins-ci-rme: remove the RTI_LICENSE_FILE export and the
Jenkinsfile workspace-layout comment
- README: document that no license file is needed
Verified with a full pipeline rebuild (scope-rebuild12) run while
rti_license.dat was absent: 0 warnings, ctest baseline unchanged,
all 5 DEBs produced, none of them containing the license.
Document the full flow: prebuilt Docker image build, release-scope
compilation (mandatory CMake flags), running the unit tests (ctest
baseline plus direct execution of the test binaries), DEB package
production with the in-tree rpmbuild script, the Jenkins/CI entry
points, the options reference and troubleshooting notes.
New TetraIF/unit-test/ directory with one QtTest executable per
TetraIF release-scope module, wired into ctest and the DEB pipeline:
- coprtcnv_unit_test: central and radio telegram header fields,
payload access, length checks, fragmentation/reassembly roundtrip,
telegram-id string helpers
- logview_unit_test: IntegerCalculator arithmetic, operator priority,
parentheses, error codes and variables (compiles the LogView class
source directly; the class otherwise lives in the LogView
executable)
- prtcnv_unit_test: PrtCnvMember state and fragmented-telegram
handling (compiles the class source directly because PrtCnvLib is
built with hidden visibility)
TetraIF/CMakeLists.txt adds the unit-test subdirectory and
new_docker_build.sh builds and ctest-runs the three new targets.
- Jenkinsfile: guez-only release pipeline; the prebuilt build image is
built from the in-tree Dockerfile (via new_docker_build.sh
--build-prebuilt-image) instead of a pre-pulled harbor image; the
builder runs inside the container with the workspace mounted at
/workspace; the post-publish cleanup stage was dropped so the DEBs
stay under ${WORKSPACE}/rpms/... (the documented output location)
- jenkins_builder_ubuntu26.sh: uses the in-tree rpmbuild/
make_rpm_ubuntu26.py instead of the outer-repo copy; same-file guard
on the packages-out copy (host output dir == make_rpm output dir)
- jenkins_publisher_ubuntu26.sh: uploads rpms/${PROJECT}/${OS_VERSION}/
directly - make_rpm_ubuntu26.py writes the DEBs there without a TDS
subfolder
- release-16-10-2026.sh: local driver adapted to the current
new_docker_build.sh interface (drops the no-longer-existing
--use-prebuilt-image option, builds the image on demand, outputs to
${WORKSPACE}/rpms/${PROJECT}/${OS_VERSION}/)
- make_rpm_ubuntu26.py: stage the new jenkins-ci-rme/ files (Jenkinsfile,
builder, publisher) into the build-pipeline DEB (release-16-10-2026.sh
was already listed)
Not copied (superseded): openSUSE builder/publisher, the old Dockerfile
recipes (replaced by the in-tree Dockerfile), Bitbucket notification
(jenkins_notifier.sh is not part of this project)
- Document --package-output-dir "${WORKSPACE}/rpms/${PROJECT}/${OS_VERSION}/"
(Jenkins convention) in the usage examples
- The in-container copy step now copies to the real mount point
/workspace/packages-out instead of ${PACKAGE_OUTPUT_DIR}, which is a host
path that does not exist inside the container (the copy was a silent
no-op; packages reached the host directly via make_rpm's
--output_dir /workspace/rpms/... on the workspace mount)
- Skip the copy when the host output directory is the same directory as
make_rpm's output dir (stat device:inode guard) - which is exactly the
case for the Jenkins-style path above, where make_rpm already writes the
packages in place
Resolution order (highest precedence first):
1. exported MIGRATE_PIPELINE_ROOT (e.g. from a Jenkinsfile)
2. MIGRATE_PIPELINE_ROOT set by .envrc in this directory
3. derived from the script location (three levels up)
The script always lives at <root>/qt-migration-legatus/qt6-after/
qt6-migrated-legatus, so the pipeline root is three levels up. This
makes the script work in a fresh Jenkins checkout with no environment
setup at all, and .envrc is no longer required.
- LogViewModels: replace deprecated invalidateFilter() with
begin/endFilterChange() (11 sites)
- LogViewTelegramEdit: move m_logger to match member declaration order
(-Wreorder) and consume QFile::open results
- PrtCnvBroadcastMgr: move m_logger to match member declaration order
(-Wreorder)
- PrtCnvCentralThread: add remaining AreaEffectSimulationType values to
the silent-ignore switch group (-Wswitch)
- PrtCnvMain: consume freopen results in daemonize (-Wunused-result)
The container script now runs the packaging flow itself instead of
calling the outer-repo jenkins_builder_ubuntu26.sh:
- Stage the workspace to deb_stage (same tar/excludes as the verified
jenkins flow; the workspace itself is never deleted)
- Run the in-tree rpmbuild/make_rpm_ubuntu26.py from the staged source
tree with the same cmake flags and arguments as before
- Copy the produced DEBs to the package output mount
- A packaging failure now fails the build instead of being logged and
swallowed
- Drop the now-unneeded MIGRATE_PIPELINE_ROOT/WORKSPACE/PROJECT_BASE_DIR
container env plumbing
The branch is now fully self-contained: new_docker_build.sh + Dockerfile
+ rpmbuild/. The build-pipeline DEB additionally ships
rpmbuild/make_rpm_ubuntu26.py (the staging list already named it).
Verified: zero-warning build, 5 DEBs with identical contents.
Keep make_rpm_ubuntu26.py (byte-identical to the validated working tree,
including the package-content validator fixes). It is self-contained
(standard library only, no helpers.py import) and its staging artifacts
land in the workspace root, so nothing else from the outer rpmbuild/
directory is needed:
- helpers.py: zero references from the Ubuntu 26 script (belongs to the
other platform variants)
- make_rpm.py / make_rpm7.py / make_rpm15.1.py / make_rpm15.2.py:
other platform packaging (base, RHEL7, CentOS 15.x)
- make_rpmModule.py / make_rpmModule7.py: module packaging
- make_rpmSVOnly.py: SV-only packaging
- make_rpm_extfw.py: external framework packaging
- make_tag.py: release tagging tool
- make_license_rpm.sh + ctc_TDS_license.spec.in + ctc_WMS_license.spec.in:
CTC license RPM flow
- .git (submodule pointer) and .gitignore (no artifacts are created in
this directory)