cmake: Fix checking compiler flags like -Wno-some-warning

[hebasto]

Some compilers (GCC) produce no diagnostic for -Wno-some-warning unless other diagnostics are being produced:

$ git diff
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 240557f..f976824 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -220,6 +220,7 @@ if(MSVC)
   add_compile_definitions(_CRT_SECURE_NO_WARNINGS)
 else()
   # Keep the following commands ordered lexicographically.
+  try_append_c_flags(-Wno-some-warning)
   try_append_c_flags(-pedantic)
   try_append_c_flags(-Wall) # GCC >= 2.95 and probably many other compilers.
   try_append_c_flags(-Wcast-align) # GCC >= 2.95.
$ cmake -B build -DCMAKE_C_COMPILER=gcc 2>&1 | grep -i unknown
-- Performing Test C_SUPPORTS__WNO_SOME_WARNING
-- Performing Test C_SUPPORTS__WNO_SOME_WARNING - Success
Compile options ....................... -Wno-some-warning -pedantic -Wall -Wcast-align -Wcast-align=strict -Wextra -Wnested-externs -Wno-long-long -Wno-overlength-strings -Wno-unused-function -Wshadow -Wstrict-prototypes -Wundef

This PR fixes compiler flag check in CMake-based build system.

Let me know, if the same fix is desirable in Autotools-based build system.

[fanquake]

Is this currently an issue for anything? If the check is broken, then it should be fixed before we start using it.

[real-or-random]

Is this currently an issue for anything? If the check is broken, then it should be fixed before we start using it.

I think noone has run into this so far, but yeah, we should fix it.

Concept ACK

[real-or-random]

Suggested change

	# Some compilers (GCC) produce no diagnostic for -Wno-unknown-warning
	# unless other diagnostics are being produced. Therefore, test the
	# -Wsome-warning case instead of the -Wno-some-warning one.
	# Some compilers (GCC) produce no diagnostic for -Wno-some-warning,
	# if "some-warning" is unknown to the compiler and no other diagnostics are being produced.
	# Therefore, test the
	# -Wsome-warning case instead of the -Wno-some-warning one.

Modulo paragraph formatting.

This confused me a lot because I thought -Wno-unknown-warning is a real flag that says "Don't warn on unknown warnings flags". (Clang has a real flag -Wno-unknown-warning-option for this!)

[hebasto]

Thanks! Reworked.

[hebasto]

Is this currently an issue for anything?

The newest current option---Wno-overlength-strings--is supported for GCC >=4.2. Only someone using an older compiler would be affected.

If the check is broken, then it should be fixed before we start using it.

The PR description mentions two checks, one per build system. The check in the Autotools-based build system:

secp256k1/build-aux/m4/bitcoin_secp.m4

Lines 64 to 78 in a526937

	dnl SECP_TRY_APPEND_CFLAGS(flags, VAR)
	dnl Append flags to VAR if CC accepts them.
	AC_DEFUN([SECP_TRY_APPEND_CFLAGS], [
	AC_MSG_CHECKING([if ${CC} supports $1])
	SECP_TRY_APPEND_CFLAGS_saved_CFLAGS="$CFLAGS"
	CFLAGS="$1 $CFLAGS"
	AC_COMPILE_IFELSE([AC_LANG_SOURCE([[char foo;]])], [flag_works=yes], [flag_works=no])
	AC_MSG_RESULT($flag_works)
	CFLAGS="$SECP_TRY_APPEND_CFLAGS_saved_CFLAGS"
	if test x"$flag_works" = x"yes"; then
	$2="$$2 $1"
	fi
	unset flag_works
	AC_SUBST($2)
	])

is broken.

The check in the CMake-based build system mirrored the behaviour of its Autotools' counterpart.

Therefore, both checks need to be fixed.