Fail build 99.0.4818.0 (dev channel) with GTK3 3.24.31

[sL1pKn07 SpinFlo]

hi. i get this fail when build chromium 99.0.4818.0 with GTK3 (GTK4 is not installed in the system)

[392/3721] /usr/bin/python "../../build/toolchain/gcc_solink_wrapper.py" --readelf="readelf" --nm="nm"  --sofile="./libgtk.so" --tocfile="./libgtk.so.TOC" --output="./libgtk.so" -- /tmp/makepkg/chromium-dev/src/chromium-99.0.4818.0/third_party/llvm-build/Release+Asserts/
bin/clang++ -shared -Wl,-soname="libgtk.so" -fuse-ld=/tmp/makepkg/chromium-dev/src/chromium-99.0.4818.0/third_party/llvm-build/Release+Asserts/bin/ld.lld -Wl,--fatal-warnings -Wl,--build-id -fPIC -Wl,-z,noexecstack -Wl,-z,relro -Wl,--icf=all -Wl,--color-diagnostics -Wl,-
-no-call-graph-profile-sort -m64 -no-canonical-prefixes -Wl,-O2 -Wl,--gc-sections -rdynamic -nostdlib++ -Wl,-z,defs -Wl,--as-needed -Wl,-rpath=\$ORIGIN -Wl,-O1,--sort-common,--as-needed,-z,relro,-z,now -o "./libgtk.so" @"./libgtk.so.rsp"  
FAILED: libgtk.so libgtk.so.TOC  
/usr/bin/python "../../build/toolchain/gcc_solink_wrapper.py" --readelf="readelf" --nm="nm"  --sofile="./libgtk.so" --tocfile="./libgtk.so.TOC" --output="./libgtk.so" -- /tmp/makepkg/chromium-dev/src/chromium-99.0.4818.0/third_party/llvm-build/Release+Asserts/bin/clang++
-shared -Wl,-soname="libgtk.so" -fuse-ld=/tmp/makepkg/chromium-dev/src/chromium-99.0.4818.0/third_party/llvm-build/Release+Asserts/bin/ld.lld -Wl,--fatal-warnings -Wl,--build-id -fPIC -Wl,-z,noexecstack -Wl,-z,relro -Wl,--icf=all -Wl,--color-diagnostics -Wl,--no-call-gr
aph-profile-sort -m64 -no-canonical-prefixes -Wl,-O2 -Wl,--gc-sections -rdynamic -nostdlib++ -Wl,-z,defs -Wl,--as-needed -Wl,-rpath=\$ORIGIN -Wl,-O1,--sort-common,--as-needed,-z,relro,-z,now -o "./libgtk.so" @"./libgtk.so.rsp"  
ld.lld: error: undefined symbol: g_file_new_for_path
>>> referenced by gtk_compat.cc
>>>               obj/ui/gtk/gtk/gtk_compat.o:(gtk::GtkFileChooserSetCurrentFolder(_GtkFileChooser*, base::FilePath const&))
>>> referenced by select_file_dialog_impl_gtk.cc
>>>               obj/ui/gtk/gtk/select_file_dialog_impl_gtk.o:(gtk::SelectFileDialogImplGTK::CreateFileOpenHelper(std::__Cr::basic_string<char, std::__Cr::char_traits<char>, std::__Cr::allocator<char> > const&, base::FilePath const&, aura::Window*))

ld.lld: error: undefined symbol: g_content_type_get_icon
>>> referenced by gtk_ui.cc
>>>               obj/ui/gtk/gtk/gtk_ui.o:(gtk::GtkUi::GetIconForContentType(std::__Cr::basic_string<char, std::__Cr::char_traits<char>, std::__Cr::allocator<char> > const&, int) const)

ld.lld: error: undefined symbol: g_file_get_type
>>> referenced by select_file_dialog_impl_gtk.cc
>>>               obj/ui/gtk/gtk/select_file_dialog_impl_gtk.o:(gtk::SelectFileDialogImplGTK::OnSelectMultiFileDialogResponse(_GtkWidget*, int))
clang++: error: linker command failed with exit code 1 (use -v to see invocation)

i see in gentoo build ok, so im not sure what happen here (have the same gtk3 version than me)

the revelant code

https://chromium.googlesource.com/chromium/src/+blame/refs/heads/main/ui/gtk/gtk_compat.cc#274

https://chromium.googlesource.com/chromium/src/+blame/refs/heads/main/ui/gtk/select_file_dialog_impl_gtk.cc#86

https://chromium.googlesource.com/chromium/src/+blame/refs/heads/main/ui/gtk/gtk_ui.cc#540

pushed in march/may in the last year. and i have 98.0.4750.0 builded without this fail

any hint? is a bug?

greetings

[Kubo Da Costa, Raphael]

+thomasanderson

These function calls come from gio (and therefore glib), so the exact GTK version is not relevant here, I think. I'd start by checking that your system's gio is exporting those symbols and that gio is being passed to the linker when linking libgtk.so below (something like "ninja -C <your-build-dir> -d keeprsp ui/gtk:gtk" will generate a <your-build-dir>/libgtk.so.rsp with the part of the linker invocation that's not shown below).

[Kubo Da Costa, Raphael]

+list again

I guess that explains it :-) Please file a bug so that it's easier to track; I'm not sure if use_gio=false is even supposed to work or is tested at this point, but I'll leave that to Thomas.

On Mon, 2022-01-17 at 16:03 +0100, sL1pKn07 SpinFlo wrote:

Hi

gio is disabled by explicit 'use_gio=false' in my build rules (setted long time ago), so seems a bug/regression

open a CRbug for this?

greetings