fix arm issues with gdb, aria2 and libhdf5 (octave) and mpd

[its-pointless]

No description provided.

[tomty89]

I would strongly recommend building our own zlib (#3455) instead patching Makefiles if the former also avoid the same problem the latter addresses.

[tomty89]

Also, we should probably consistently use CPPFLAGS as it appears that we want the flags to be used for both clang and clang++?

[its-pointless]

I need feedback from testing though since some interactions are only going to show up on further testing.
CPPFLAGS ? Yeah good point.
I will look into it later

[tomty89]

Should we just ditch this condition and always use -Oz?

termux-packages/scripts/build/termux_step_setup_toolchain.sh

Lines 51 to 55 in 61785d9

	if [ $TERMUX_ARCH = arm ]; then
	CFLAGS+=" -Os"
	else
	CFLAGS+=" -Oz"
	fi

Also can we have -fexception by default, for at least arm maybe?

[tomty89]

I have to say this is no better than #2588, sorry.

[its-pointless]

Its not meant to be better its the only way iv gotten gmic to run properly on arm. Using zlib from your pr didn't work even if built with -fexceptions.

[tomty89]

"Didn't work" doesn't really tell anything. You need to at least show that gmic is and is only linked to libz.so.1 (or maybe the symlink libtermuz.so) but segfault still originates from it (the zlib). The former may not be the case if some hardcoded -lz is used at the end.

[its-pointless]

It doesn't originate there its caused because the static libunwind.a isn't linked propertly.

[tomty89]

Not even sure what you mean. Actually there is even a note seeming to hint that we should avoid linking libraries against libgcc.a and libunwind.a (and now we may only need to exclude libgcc.a as it's a linker script to both the real libgcc and libunwind).

And still you are not showing any log (on what linkage the gmic has and how things have/have not changed with a zlib built by ourselves).

[tomty89]

Oh forgot the link: https://android.googlesource.com/platform/ndk/+/master/docs/BuildSystemMaintainers.md#Unwinding

[its-pointless]

current gmic
Program received signal SIGSEGV, Segmentation fault.
0xf7189c3c in __gnu_Unwind_Resume () from /system/lib/libz.so
built on your zlib branch
Program received signal SIGSEGV, Segmentation fault.
0xf7407020 in __gnu_Unwind_Resume () from /system/lib/libm.so

its not linking properly since it looks kind of like this.
android/ndk#889

[tomty89]

Well you gmic was still linked against /system/lib/libz.so, so

[its-pointless]

Should we just ditch this condition and always use -Oz?
termux-packages/scripts/build/termux_step_setup_toolchain.sh

Lines 51 to 55 in 61785d9

if [ $TERMUX_ARCH = arm ]; then
CFLAGS+=" -Os"
else
CFLAGS+=" -Oz"
fi

Also can we have -fexception by default, for at least arm maybe?

Don't recommend -fexceptions as default as it increases install size. We need to keep things small especially for arm.
About -Oz vs -Os ... i don't know yet.

[its-pointless]

there is another fix for this. adding -Wl,--exclude-libs,libgcc_real.a to LDFLAGS
as in LDFLAGS for all arm libs...

also if you want to see and understand what is happening for yourself in an easy to do manner.
arm-linux-androideabi-clang -Wl,-u,__aeabi_memset -Wl,-u,__aeabi_idiv -shared -o test.so && nm test.so | grep Unw

[its-pointless]

well you then ask "how could this affect mpd?"
mpd uses exceptions so it needs the libunwind.a versions of these functions. Because some libs are exporting these functions using libgcc_real.a versions mpd gets libgcc_real.a version of those functions at link time instead of libunwind.a. which is why putting -lunwind before those libs that export libgcc_real.a versions fixes it.

[its-pointless]

android/ndk#379
explains this situation