tests/libc_newlib: add test for newlib-nano inclusion

[cladmi]

Contribution description

This test verifies that when used, newlib-nano is correctly included, both
header and also linked.

This is a standalone test and is less invasive than doing a global test like what I tried in #9512

I want to use it to debug non handled developers computers.

Testing

I tested the following setups:

• native (no newlib)
• pic32-wifire (newlib) using the flasher PR
• samr21-xpro (newlib-nano)

Issues/PRs references

#9512

This should not be merged before #9398 as currently llvm is not handled properly but it should not block adding tests in murdock.
I now think it should be included before #9398, I thought it would be all fixed faster.
I already have a fix for the llvm/clang issue #9513

[miri64]

Do we really need to run tests on this PR everytime ;-)?

[cladmi]

I was still getting a redis error from the build output.

[jcarrano]

Good. I can pass the test, but I can't fail it (Arch linux).

[jcarrano]

Can you include instructions on how to force a failure of this test? Then people can test the test.

[cladmi]

The link time test (currently done at runtime) can be tested by commenting this line:

RIOT/makefiles/libc/newlib.mk

Line 17 in db6cb1a

export LINKFLAGS += -specs=nano.specs

And the compile time test can be tested by commenting this one:

RIOT/makefiles/libc/newlib.mk

Line 80 in db6cb1a

INCLUDES := -isystem $(NEWLIB_NANO_INCLUDE_DIR) $(INCLUDES)

[jcarrano]

Tested both tests and it failed as expected.

[jcarrano]

This can be tested by binutils, so there's no need to run the application.

[cladmi]

I will try something for this.

[jcarrano]

My test results (BOARD=stm32f4discovery, Arch Linux):

2018-07-23 15:39:23,382 - INFO # main(): This is RIOT! (Version: 2018.10-devel-27-ga2307-batman-test_newlibnano)
2018-07-23 15:39:23,383 - INFO # Newlib/nano test
2018-07-23 15:39:23,383 - INFO # .
2018-07-23 15:39:23,383 - INFO # OK (1 tests)

System

Installed compiler toolchains 
-----------------------------
             native gcc: gcc (GCC) 8.1.1 20180531
      arm-none-eabi-gcc: arm-none-eabi-gcc (Arch Repository) 8.1.0
                avr-gcc: missing
       mips-mti-elf-gcc: missing
             msp430-gcc: missing
   riscv-none-embed-gcc: missing
                  clang: clang version 6.0.1 (tags/RELEASE_601/final)

Installed compiler libs
-----------------------
   arm-none-eabi-newlib: "3.0.0"
    mips-mti-elf-newlib: missing
riscv-none-embed-newlib: missing
               avr-libc: missing (missing)

Installed development tools
---------------------------
                  cmake: cmake version 3.11.4
               cppcheck: Cppcheck 1.84
                doxygen: 1.8.14
                 flake8: 3.5.0 (mccabe: 0.6.1, pycodestyle: 2.4.0, pyflakes: 2.0.0) CPython 3.6.6 on Linux
                    git: git version 2.18.0
             coccinelle: missing

[jcarrano]

Open question: How can we make it so that we force the user to run the test?

[cladmi]

I added a test in the Makefile. Currently not documented but I should move the documentation out of main.c to the makefile.

[cladmi]

I also tested wih TOOLCHAIN=llvm it currently fails during compilation because this is not fixed #9513.
I commented some parts to be able to run the test with llvm-nm at it also worked.

[cladmi]

I split the NM change in a separate PR.

@jcarrano Tell me if you like the new tests. I will update the documentation after

@miri64 I now think it should be included before #9398, I thought it would be all fixed faster.
I already have a fix for the llvm/clang issue #9513. What do you think ?
I would really like to go forward for fixing newlib for the release.

[jcarrano]

A dumber approach without sed (checks the number of different addresses)

# make N_PRINTFS = 2 if non-nano, 1 if nano

test-newlib: $(ELFFILE)
   PRINTFS=$$($(NM) -v $^ | grep '\( i\?printf\)'); \
   PRINTF_ADDRS=$$(echo "$${PRINTFS}" | cut -d' ' -f 1 | uniq | wc -l); \
   # ensure both symbols are there
   if test $$(echo "$${PRINTFS}" | wc -l) = 2 ; then \
       echo "[FAILED] Expected printf and iprintf to be present, found $${PRINTFS}" ; \
       exit 1; \
   elif test $${PRINTF_ADDRS} != $(N_PRINTFS) ; then \ 
       echo "[FAILED] Expected $(N_PRINTFS) different printfs"; \
       exit 1; \
   fi

BTW, could we define a standard set of shell functions to be included in all makefiles. I would like to have a 'die' function:

die() {
    echo "${1}" >&2
    exit 1
}
# one-line "do stuff and exit with error if it fails"
test -e fileXX.txt || die "file not found"

[cladmi]

Using more sub commands, and not simpler for me to understand as it uses two concepts, number of printfs + addresses instead of just comparing addresses.

sed is standard enough and is working properly.

Adding common functions makefile would be a different PR with its own problem.

Also for me echo; exit can be understood in one blink and not complex enough that I would like it replaced by another function that I would need to check. Also it can be done with oneliner as

test -e fileXX.txt || { echo "file not found"; exit 1};

[jcarrano]

@cladmi The nm test is good.

I now think it should be included before #9398.

I think this PR should have a high priority, the newlib thing is a bug waiting to happen.

[jcarrano]

Good. We need another PR to make this test compulsory.

[cladmi]

Reminder, for backporting I would also need to add the NM changes.

[cladmi]

Backport provided in #9633