make: newlib: check if newlib nano folder was found

[kaspar030]

Contribution description

The logic for finding the newlib nano include dir can, if no folder is found, end up with an empty "-I".
This leads to weird error messages, like:

cc1: error: -I/home/kaspar/src/riot/core/include: No such file or directory [-Werror=missing-include-dirs]

followed by many headers that cannot be found.

This is caused by "-I -Icpu/include". I lost a while debugging this, and just yesterday another user on IRC popped up with the same problem.

Underlying is a fedora newlib packaging problem, which seems to not build with nanospecs.

This PR adds a simple check whether NEWLIB_NANO_INCLUDE_DIR is empty despite nanospecs enabled, and outputs an error.

Testing procedure

Move the newlib nano folder (/usr/arm-none-eabi/include/newlib-nano on my arch) out of the way. Compile on master and with this PR.

Issues/PRs references

none

[cladmi]

I had a similar change there #9384 putting a warning, so was thinking about the same thing.

By re-reading the error message mentioning the private USE_NEWLIB_NANO variable (only used in this file, not documented) I was thinking that I would prefer that it says something about newlib_nano module, not an internal variable. But maybe it just showed me the issue.

So I re-read the file to know the difference and now I question if the global handling of newlib_nano is correct.

Why do we need the USE_NEWLIB_NANO in the first place ?

In this file USE_NEWLIB_NANO is set if newlib_nano module is requested and available for the linker

RIOT/makefiles/libc/newlib.mk

Lines 2 to 4 in db6cb1a

	# Test if nano.specs is available
	ifeq ($(shell $(LINK) -specs=nano.specs -E - 2>/dev/null >/dev/null </dev/null ; echo $$?),0)
	USE_NEWLIB_NANO = 1

But why then treat not having the header as an error and not just set USE_NEWLIB_NANO = 0 in the same way.

However, in some tests I used #ifdef MODULE_NEWLIB_NANO to know if newlib nano is enabled, but it may be set without being enabled, so it is inconsistent, and now I do not know how to get the information.

So I question if there is any reason justifying having USE_NEWLIB_NANO ?

[jcarrano]

Tested. The error message is much more clear now.

[gschorcht]

@kaspar030 When upgrading the compiler version for ESP32 as a prerequisite for upgrading to the current ESP-IDF version, I encountered a problem with the variable NEWLIB_NANO_INCLUDE_DIR.

Although xtensa-esp32-elf-gcc version 8.4.0 uses the nanospecs, there is no separate include directory for the nano version of newlib.h. When the nano version of newlib is created with crosstool-NG, the newlib.h in the default include directory is just the nano version. This was not a problem with xtensa-esp32-elf-gcc version 5.2.0 because it doesn't use nanospecs.

Do you have any idea how to solve the problem other than compiling newlib-nano and newlib separately and preparing the folder structure by hand? I could not find any option in crosstool-NG to create separate directories for the nano version of newlib.

[gschorcht]

I wonder if NEWLIB_NANO_INCLUDE_DIR has to point to a different directory than NEWLIB_INCLUDE_DIR or if it can just be the same if no separate directory is found.