The toolchain variables were overwritten anyway in
`makefiles/toolchain/gnu|llvm.inc.mk` as they use `=` affectation,
except for `OBJCOPY`.
On my Linux ubuntu bionic machine, the only difference for `info-build`
with both gnu/llvm and also with LTO=1
is that it now uses an absolute path for `objcopy`
-OBJCOPY: objcopy
+OBJCOPY: /usr/bin/objcopy
On the OSX 17.7.0 I tested, there was no difference.
This variable is only used for the term recipe (and maybe for flashing). They
should not be evaluated if they are not needed and the user should not see a
warning that the port is not set if he does not use port (for example in make
all.)
DEBUGGER/DEBUGGER_FLAGS/DEBUGSERVER/DEBUGSERVER_FLAGS are evaluated by the
main Makefile.include or by file included by it.
Their value does not need to be exported.
Testing
-------
`git diff --word-diff` only reports `export` being removed.
`git show --stat` reports `55 insertions(+), 55 deletions(-)`
Which is the same amount as lines that where matching
`export[[:blank::]]\+VARIABLE`.
FLASHER and FFLAGS are evaluated by the main Makefile.include or by file
included by it. Their value does not need to be exported.
This will also prevent evaluating 'PORT' for FFLAGS when not needed.
Testing
-------
`git diff --word-diff` only reports `export` being removed.
`git show --stat` reports `84 insertions(+), 84 deletions(-)`
Which is the same amount as lines that where matching
`export[[:blank::]]\+VARIABLE`.
Some modules used a 'NATIVEINCLUDES' with different include path and no other
included directories.
It was defining basic 'include' in a different order and not using other things
defined in INCLUDES.
After doing some checks with the given include path and possible conflicting
files, there should be no conflict when using the default one.
* No common headers between all the NATIVEINCLUDES directories
* No common headers files between board/native/include, cpu/native/include and
other files in the repository (except other boards/cpus of course).
`PORT` is unset if `netdev2_tap` is not in `USEMODULES`.
But unfortunately, `netdev2_tap` is set by boards/native/Makefile.dep
*after* `PORT` is set, so that `PORT` remains empty.
This commit breaks this circular dependency by checking for
`netdev_default` and `gnrc_netdev_default` instead.
Without this commit, `make term` throws an error for examples that
require a tap interface, when `PORT` is not explicitely given.
It can be useful to know if the current compilation builds the native
board module, because e.g. then we must not include `sys/types.h`.
This diff adds `-DNATIVE_INCLUDES` to the make variable
`NATIVEINCLUDES`.
Sometimes boards/*/Makefile.include (e. g. in case of the msba2) gets included
twice somehow, leading the TERMFLAG to be set twice and faulty. This
fixes that.
Creating all object files in one directory is bound to produce name
clashes. RIOT developers may take care to use unique file names, but
external packages surely don't.
With this change all the objects of a module (e.g. `shell`) will be
created in `bin/$(BOARD)/$(MODULE)`.
I compared the final linker command before and after the change. The
`.o` files (e.g. `startup.o`, `syscall.o` ...) are included in the same
order. Neglecting the changed path name where the `.o` files reside, the
linker command stays exactly the same.
A major problem could be third party boards, because the location of the
`startup.o` needs to the specified now in
`boards/$(BOARD)/Makefile.include`, e.g.
```Makefile
export UNDEF += $(BINDIR)msp430_common/startup.o
```
The new target names are easier to remember / decipher:
term
term-cachegrind
term-gprof
term-valgrind
all
all-cachegrind
all-gprof
all-valgrind
all-debug
eval-gprof
eval-cachegrind
native modules will never need the dynamic INCLUDES, so we define our
own NATIVEINCLUDES. Due to the current make structure, the only way to
not use INCLUDES is to redefine the build rules.
An application might want to use C11 features. The user would assume
that setting `CFLAGS=-std=gnu11` in the Makefile would work. It does not
since the board's Makefile.include shadows the `-std` flag.
This patch removes the `-std=gnu99` from the various Makefile.includes,
and sets the flag in the common Makefile.include of RIOT instead.
If an `-std` flag was provided by an earlier Makefile (the application,
the board, or the CPU [whilst only the former one should]), then no
additional flag is set. It is first tested if the supplied compiler
understands `-std=gnu99`, then `-std=c99`.
* add missing $(AD) to osx $(LINK) invocation
* move osx build determination to native makefile
* move old libc test to native makefile
* set objcopy to "true" - it is superfluous for native no matter what the system is
* add some documentation to natives makefile
wrap some libc functions that do system calls (terminal output)
wrap read/write with syscall guard
define real_read/write (next dynamic linker find for read/write)
guard system calls in remaining code
introduce native_internhal.h
throw out some debug statements that break things
clean up includes a bit
declare board_init in native_internhal.h
add -ldl to LINKFLAGS for cpu/syscalls
