cache.h before system includes to avoid duplicate include warnings
- except for stddef.h, which should come before cache.h to avoid warnings on
  Linux/gcc
some reordering for the same reason
remove some unused includes
misc cleanup

Still incomplete

Info 793: ANSI/ISO limit of 31 'significant characters in an external
    identifier', '__pthread_unregister_cancel_restore(struct
    __pthread_unwind_buf_t *)', exceeded -- processing is unaffected
Info 793: ANSI/ISO limit of 31 'significant characters in an external
    identifier', 'pthread_mutexattr_getprioceiling(const union
    pthread_mutexattr_t *, int *)', exceeded -- processing is unaffected
Info 793: ANSI/ISO limit of 31 'significant characters in an external
    identifier', 'pthread_mutexattr_setprioceiling(union pthread_mutexattr_t *,
    int)', exceeded -- processing is unaffected

duplicate includes via cache.h and common.h

export missing VSB symbols from libvarnishapi

Don't:

	MACRO(
		a,
	#if FOO
		b,
	#else
		c,
	#endif
		d
	)

A function like macro was invoked whose arguments extended for
multiple lines, which included preprocessor statements. This is
almost certainly an error brought about by a missing right parenthesis.

By the rules of Standard C the preprocessing directive is absorbed
into the macro argument but then will not subsequently get executed.
For this reason some compilers treat the apparent preprocessor
directive as a directive. This is logical but not portable.

It is therefore best to avoid this construct.

wouldn't make sense.

Fixes: #2099

NB: This _may_ expose certain corner cases of unwise symbol overloading
    in VCL.

no revision 1.7

thanks @dridi

am__check_pre in the automake-generated Makefile basically runs

	test -d tests || mkdir tests || exit $?

When running out-of-tree with parallel make (as vtest does), this
becomes racy. Avoid the race by a mkdir ignoring the error

The previous commit did not solve the issue for all cases
- we also change OFLAGS in configure

In particular with --enable-developer-warnings, we select strict compiler
options for CFLAGS, which are too strict for the configure built-in conftest.c
code.

Make distcheck calls configure again for an out-of-tree build, so it would
see the CFLAGS the "outer" configure built. To avoid this situation, we
restore the original CFLAGS for the inner configure.

I wonder if there is a better, more autoconf-ish, solution to this problem.

Tested with SolarisStudio 12.4 on amd64/x86

and define it as NOP for Sun CC

It took me way too long to understand that empty declarations
found by suncc were caused by the extra ;

To avoid an undefined symbol __stack_chk_fail_local due to compiling
and linking gcc not seeing the same flags, we need to trick libtool
into passing a flag it would otherwise remove.

Please hold your whitespace ocds and exclude the non-standard indent
chosen to avoid excessive line length.

I am not entirely happy about this and we should try to find a better solution
to this chicken-egg-ish situation:

- part of our documentation is hand-written (in git/srcdir), part is auto-
  generated

- because we add generated rsts to the dist, they are not in srcdir when
  building from git, but they are when building from a dist (tarball)

- sphinx wants all files in one doctree - there seems to be no include path

I tend to lean towards removing generated rsts from the dist to unify the
src tree for git vs. dist builds, but I'd like to avoid too much change at
this point due to other priorities.

... which revealed that they actually fail atm, so another fix will be coming up