PdfFile/test: CMake wiring + fixture investigation

[juliusknorr]

Summary

Migrates the qmake PdfFile/test/test.pro GoogleTest suite to a CMake/CTest target and documents the fixture situation.

• New PdfFile/test/CMakeLists.txt: project(pdffile_test), CORE_ROOT_DIR = ../.., include(common.cmake), guard-added deps, add_core_gtest(NAME pdffile_test SOURCES test.cpp LIBS ... GTEST_MAIN GTEST_FILTER ...). test.cpp has no own main(), so GTEST_MAIN is used.
• Registered in the top-level CMakeLists.txt if(EO_BUILD_TESTS) block, after the existing four suites.
• TESTING.md updated with the full fixture catalogue and exact steps to fully enable.

Dependencies

test.pro's ADD_DEPENDENCY(UnicodeConverter, kernel, graphics, PdfFile, DjVuFile, ooxmlsignature) maps cleanly to existing CMake targets — no dependency lacks a CMake target, so no library porting is needed:

dep	CMake target source
UnicodeConverter	UnicodeConverter
kernel	Common
graphics	DesktopEditor/graphics/cmake
PdfFile	PdfFile
DjVuFile	DjVuFile
ooxmlsignature	DesktopEditor/xmlsec/src

Gating decision

Registered live, but with one case excluded via GTEST_FILTER to keep CI green.

Every test case in test.cpp calls GTEST_SKIP() at the top except CPdfFileTest.EditPdfFromBase64, whose GTEST_SKIP() is commented out. That case calls LoadFromFile() (an ASSERT_TRUE on loading the missing test.pdf) and opens base64.txt, so it would hard-fail headless. It is excluded with GTEST_FILTER "-CPdfFileTest.EditPdfFromBase64". The suite then builds, links and runs cleanly (remaining cases report as skipped), so it gates the build/link without turning CI red.

Missing fixtures (none present anywhere in the repo)

Inputs: test.pdf, base64.txt, pdf.bin, pfx.pfx (PKCS#12, password 123456), test.djvu, changes.bin, test.jpeg (used by EditPdfSign; not in the original blocker list), and an resI/ directory of reference PNGs for ImgDiff.

Intermediate: ONLYOFFICEFORM.docxf (produced by GetMetaData, consumed by SetMetaData).

Outputs (generated at runtime, not required inputs): test2.pdf, test3.pdf, test_split.pdf, pdftemp/, resO/, resD/, fonts_cache/, resPdfBinToPng.png.

To fully enable

Commit the input fixtures above, stage them next to the test binary (the suite reads from NSFile::GetProcessDirectory(), so use WORKING_DIRECTORY plus a POST_BUILD copy_directory), drop the GTEST_FILTER, and un-GTEST_SKIP() the desired cases.

Build/CI note

There is no vcpkg toolchain locally (VCPKG_ROOT unset), so this could not be fully configured/built here. The PR's CI is the gate: configure with -DVCPKG_MANIFEST_FEATURES=tests -DEO_BUILD_TESTS=ON, build, then ctest --output-on-failure.

https://claude.ai/code/session_01TJrhgZT4PwYaBKWiUCXmGg

---

Generated by Claude Code

[juliusknorr]

🏓 for review. Will rebase and resolve conflicts before merging as every of the test prs conflicts

[chrip]

Summary

Migrates the qmake PdfFile/test/test.pro GoogleTest suite to a CMake/CTest target
(pdffile_test), registers it in the top-level EO_BUILD_TESTS block, adds a portable
non-Windows definition of the *_HRESULT_* assertion macros so it compiles on Linux/macOS,
and documents the missing-fixture situation in TESTING.md. Every factual claim in the PR
body (dependency→target mapping, the single non-skipped case, the fixture catalogue, the
gating strategy) was verified against the code at the head SHA and holds up. This is a
well-scoped, honestly-labelled "wiring + investigation" change: it gates the build/link/load
path for the suite but, by design, exercises no PdfFile production logic yet. CI is fully
green and a maintainer (@DmySyz) has already approved at the current head.

Verification

Claim / Item	Reality (verified at head b994641)	Status
add_core_gtest supports NAME/SOURCES/LIBS/GTEST_MAIN/GTEST_FILTER	Defined common.cmake:380-436; exactly those options/args exist	✓
GTEST_MAIN needed because test.cpp has no main()	No int main in test.cpp; macro writes a bundled entry point	✓
GTEST_FILTER "-CPdfFileTest.EditPdfFromBase64"	Valid GoogleTest negative filter; passed as --gtest_filter= (common.cmake:422-423)	✓
All 6 deps map to existing CMake targets	UnicodeConverter→UnicodeConverter, Common→project(kernel), DesktopEditor/graphics/cmake→project(graphics), PdfFile→PdfFile, DjVuFile→DjVuFile, DesktopEditor/xmlsec/src→project(ooxmlsignature); all add_library(... SHARED) present	✓
Every case self-skips except EditPdfFromBase64	25 TEST_F cases; 24 call GTEST_SKIP() first; EditPdfFromBase64 (test.cpp:531-533) has //GTEST_SKIP(); commented out	✓
That case would hard-fail headless	It runs LoadFromFile() (ASSERT_TRUE) + opens base64.txt; excluding it via filter is the reason	✓
HRESULT macros were the Linux compile blocker	EXPECT_HRESULT_SUCCEEDED used at test.cpp:285,293,331,739,747…; guarded by #ifndef so Windows (where gtest defines them) is unaffected	✓
Registered after the existing four suites in EO_BUILD_TESTS	Added at CMakeLists.txt:47, inside if(EO_BUILD_TESTS) after test_odf	✓
New CMakeLists.txt follows repo convention	Matches sibling test dirs (OOXML/test, OdfFile/Test/test_odf): cmake_minimum_required(3.10) → project() → CORE_ROOT_DIR → include(common.cmake) → if(NOT TARGET) guards → add_core_gtest()	✓
CI status	build (Windows x64), build (Linux), Build WASM, build-wasm, DCO — all SUCCESS	✓

Issues & Suggestions

⚠️ Major

(none)

ℹ️ Minor / 💡 Suggestions

• Zero production-code coverage (by design, disclosed). After the filter, the suite is
  24 self-skips + 1 excluded case, so it exercises no PdfFile logic — it is purely a
  build/link/loader gate until fixtures land. The PR body and TESTING.md state this plainly,
  so it is not a defect, but the coverage ceiling is worth keeping in mind: this PR's value is
  "the suite compiles, links and runs green in CI", not "PdfFile is tested".
• EditPdfFromBase64 self-consistency (test.cpp:531-533). The one hard-failing case is
  suppressed only at the CTest layer via GTEST_FILTER, while its in-source GTEST_SKIP()
  stays commented out. Running the binary directly (outside ctest, or ctest with an overridden
  filter) would still hit the ASSERT_TRUE on the missing test.pdf. Cleaner would be to
  restore GTEST_SKIP() in source like the other 24 cases (making the filter belt-and-braces),
  though the commented marker doubles as a useful "enable this one first" pointer — discussable.
• Commit 1 message not Conventional Commits. PdfFile/test: CMake wiring + fixture investigation uses a path prefix, not a type; the repo convention is fix:/feat:/chore:/test:
  (e.g. recent chore: cleanup code). Commit 2 (test: define HRESULT assertion macros…) is
  fine. Consider test(PdfFile): … on a rebase. Non-blocking.
• Rebase pending. The author already noted (2026-06-26) they will rebase to resolve
  conflicts with the other test PRs before merge; PR currently reports MERGEABLE. No action now.

Verdict

Approve — every claim in the PR verifies against the code at head, the CMake/CTest wiring
and dependency mapping are correct and idiomatic for this repo, CI is green, and a maintainer
has already approved; the only open points are minor/discussable (net-zero coverage is
intentional and disclosed, the EditPdfFromBase64 filter-vs-source nit, and commit-1 message
style). Merge after the planned rebase.

Assisted-by: ClaudeCode:claude-opus-4-8