For some reason, meson does not detect Darwin and passes -Wl,-O1 to the compiler:

https://github.com/mesonbuild/meson/blob/bbcf80d7348135e0590622a7b85c30429f005c07/mesonbuild/compilers/compilers.py#L1479

Both C and ObjC compilers used are

clang version 5.0.2 (tags/RELEASE_502/final)
Target: x86_64-apple-darwin17.7.0
Thread model: posix
InstalledDir: /nix/store/frq6nffl9iszwlznlha23jyv2p1rrgc9-clang-5.0.2/bin

which causes CompilerType.CLANG_STANDARD to be selected for ObjC compiler, causing the -O1 flag being passed to the linker. C compiler works fine since it detects Darwin with mesonlib.for_darwin(want_cross, self). As a stopgap, we could add the same check for ObjC but in the long term, meson should probably check the linker version.

Opened an upstream issue with Meson: mesonbuild/meson#4784

@jtojnar it looks like ofborg Darwin build of glib has actualy failed?

Are there any other dconf uses we need to get rid of? We can also make dconf an optional propagated input of glib to reduce duplication.

In general, packages should not depend on dconf (unless they explicitly need libdconf for low-level access). Instead, they should pick up the dconf-based GSettings backend module from the environment (passed by wrapGAppsHook). We should probably add the !isDarwin condition to all manual wrapProgram invocations (or change them to wrapGAppsHook).

I am not sure what duplication do you mean but dconf depends on glib so we definitely cannot propagate it from there.

I think this broke glibmm on darwin: https://hydra.nixos.org/build/87684380

Complains about not finding a header file:

desktopappinfo.cc:27:10: fatal error: 'gio/gdesktopappinfo.h' file not found
#include <gio/gdesktopappinfo.h>
^~~~~~~~~~~~~~~~~~~~~~~
1 error generated.

There's similar failures in other darwin derivations.

Needed by inkscape: https://hydra.nixos.org/job/nixpkgs/trunk/unstable#tabs-constituents

Yeah, glib does not build them on Darwin: https://gitlab.gnome.org/GNOME/glib/blob/be10f19752223f2607ba1fe00408c75920792aba/gio/meson.build#L412-429

It previously did build them because it detects Mac by the presence of Cocoa. Maybe we need to convince glibmm glibmm it is on Darwin in a same way? Or report a bug?

Yeah, there are the same checks, we need cocoa in glibmm:

https://gitlab.gnome.org/GNOME/glibmm/blob/96a997b1f4cf902a7690de8a27e1173f1d217042/configure.ac#L58-76

https://gitlab.gnome.org/GNOME/glibmm/blob/96a997b1f4cf902a7690de8a27e1173f1d217042/gio/src/desktopappinfo.hg#L26