test(types_test.cpp.o): Build error in Fedora 35

[tim77]

In Fedora 34 compiles fine, but build error in f34. GCC version basically identical - 11.0.1.

FAILED: test/types_test.p/types_test.cpp.o 
g++ -Itest/types_test.p -Itest -I../test -Iinclude -I../include -I/usr/include -fdiagnostics-color=always -pipe -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -Wnon-virtual-dtor -Wextra -std=c++17 -Wno-deprecated-declarations -O2 -flto=auto -ffat-lto-objects -fexceptions -g -grecord-gcc-switches -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -fstack-protector-strong -specs=/usr/lib/rpm/redhat/redhat-annobin-cc1 -m64 -mtune=generic -fasynchronous-unwind-tables -fstack-clash-protection -fcf-protection -MD -MQ test/types_test.p/types_test.cpp.o -MF test/types_test.p/types_test.cpp.o.d -o test/types_test.p/types_test.cpp.o -c ../test/types_test.cpp
In file included from /usr/include/signal.h:315,
                 from /usr/include/c++/11/csignal:42,
                 from ../test/doctest.h:3241,
                 from ../test/types_test.cpp:2:
../test/doctest.h:4403:45: error: size of array ‘altStackMem’ is not an integral constant-expression
 4403 |         static char             altStackMem[SIGSTKSZ];
      |                                             ^~~~~~~~
../test/doctest.h:4453:48: error: size of array ‘altStackMem’ is not an integral constant-expression
 4453 |     char    FatalConditionHandler::altStackMem[SIGSTKSZ] = {};
      |                                                ^~~~~~~~

• Full build log.
• GCC: 11.0.1
• Meson f34: 0.56.2
• Meson f35: 0.57.2

[alebastr]

Already fixed by using system doctest in #43.

wf-config.spec diff

--- a/wf-config.spec
+++ b/wf-config.spec
@@ -6,14 +6,17 @@ Summary:        Library for managing configuration files, written for wayfire
 License:        MIT
 URL:            https://github.com/WayfireWM/wf-config
 Source0:        %{url}/archive/v%{version}/%{name}-%{version}.tar.gz
+# Fix build caused by glibc 2.34 SIGSTKSZ changes
+# Fedora doctest package is already patched, but wf-config was using bundled copy
+Patch0:         %{url}/pull/43.patch#/wf-config-0.7.0-use-system-doctest.patch

 BuildRequires:  gcc-c++
 BuildRequires:  meson
 BuildRequires:  cmake
+BuildRequires:  cmake(doctest)
 BuildRequires:  cmake(glm)
 BuildRequires:  pkgconfig(libevdev)
 BuildRequires:  pkgconfig(libxml-2.0)
-BuildRequires:  pkgconfig(wlroots) >= 0.12.0

 %description
 %{summary}.
@@ -40,6 +43,10 @@ Development files for %{name}.
 %meson_install


+%check
+%meson_test
+
+
 %files
 %license LICENSE
 %{_libdir}/lib%{name}.so.0*

Similar issue with wayfire build should be fixed by updating bundled wf-touch to WayfireWM/wf-touch@8974eb0

[tim77]

@alebastr thank you! Updated package.

[tim77]

If you don't mind i'd reopen issue since we still have a problem with tests, but now on 32-bit arches like i686 and armv7hl.

+ /usr/bin/meson test -C i686-redhat-linux-gnu --num-processes 48 --print-errorlogs
ninja: Entering directory `/builddir/build/BUILD/wf-config-0.7.0/i686-redhat-linux-gnu'
ninja: no work to do.
 1/10 Types test          OK              0.03s
 2/10 OptionBase test     OK              0.03s
 3/10 Option test         OK              0.02s
 4/10 Option wrapper test OK              0.02s
 5/10 Section test        OK              0.02s
 6/10 XML test            OK              0.02s
 7/10 ConfigManager test  OK              0.02s
 8/10 Log test            OK              0.01s
 9/10 File parsing test   OK              0.31s
10/10 Duration test       FAIL            0.37s   exit status 1
>>> MALLOC_PERTURB_=147 /builddir/build/BUILD/wf-config-0.7.0/i686-redhat-linux-gnu/test/duration_test
――――――――――――――――――――――――――――――――――――― ✀  ―――――――――――――――――――――――――――――――――――――
[doctest] doctest version is "2.4.5"
[doctest] run with "--help" for options
===============================================================================
../test/duration_test.cpp:12:
TEST CASE:  wf::animation::duration_t
../test/duration_test.cpp:19: ERROR: CHECK( duration.running() == false ) is NOT correct!
  values: CHECK( true == false )
../test/duration_test.cpp:20: ERROR: CHECK( duration.progress() == doctest::Approx{1.0} ) is NOT correct!
  values: CHECK( -9392801.5 == Approx( 1.0 ) )
===============================================================================
[doctest] test cases:  3 |  2 passed | 1 failed | 0 skipped
[doctest] assertions: 56 | 54 passed | 2 failed |
[doctest] Status: FAILURE!
――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――
Summary of Failures:
10/10 Duration test       FAIL            0.37s   exit status 1
Ok:                 9   
Expected Fail:      0   
Fail:               1   
Unexpected Pass:    0   
Skipped:            0   
Timeout:            0   

Full log.

[alebastr]

Ah, that's a funny one: duration_t::impl::is_ready assumes that get_elapsed on freshly initialized object will return sufficiently large positive number. That works on 64-bit architectures, as now() - 0 fits into 64-bit long. That also fails on all 32-bit architectures as the value overflows 32-bit long type and becomes negative.

wf-config/src/duration.cpp

Lines 33 to 38 in 4a69935

	long get_elapsed() const
	{
	using namespace std::chrono;
	auto now = system_clock::now();
	return duration_cast<milliseconds>(now - start_point).count();
	}

I guess get_elapsed should be returning std::chrono::milliseconds::rep.

[ammen99]

Ah, that's a funny one: duration_t::impl::is_ready assumes that get_elapsed on freshly initialized object will return sufficiently large positive number. That works on 64-bit architectures, as now() - 0 fits into 64-bit long. That also fails on all 32-bit architectures as the value overflows 32-bit long type and becomes negative.

wf-config/src/duration.cpp

Lines 33 to 38 in 4a69935

	long get_elapsed() const
	{
	using namespace std::chrono;
	auto now = system_clock::now();
	return duration_cast<milliseconds>(now - start_point).count();
	}

I guess get_elapsed should be returning std::chrono::milliseconds::rep.

Nice find, we could also just return int64_t, I am not a fan of such long types :)

[tim77]

Thank you. Built for f35 and all tests passed.