Makefile.include: resolve dependencies before Makefile.include

[fjmolinas]

Contribution description

This PR moves BOARD and CPU Makefile includes after dependency resolution.

Testing procedure

• green murdock

• run time ./dist/tools/buildsystem_sanity_check/save_all_dependencies_resolution_variables.sh on this PR and master. There should be no diff. I ran it on a version of this branch rebased on sys/arduino: include arduino_sketches in Makefile.dep #14350. But I think that shouldn't matter since that module was included after dependency resolution already so result should not be affected by this PR.

I'm ticking it as major since it affects all BOARD CPU, but IMO the change if fine if no dependencies have changed :)

Issues/PRs references

Ticks an item in #9913

[aabadie]

There are a couple of other build system tricks that could (or should ?) be removed as well:

RIOT/boards/native/Makefile.include

Line 81 in a665fcc

EEPROM_FILE_FLAGS = $(if $(filter periph_eeprom,$(FEATURES_USED)),--eeprom $(EEPROM_FILE),)

RIOT/makefiles/arch/cortexm.inc.mk

Line 74 in a665fcc

CFLAGS_FPU ?= $(if $(filter cortexm_fpu,$(USEMODULE)),$(_CORTEX_HW_FPU_CFLAGS),-mfloat-abi=soft)

I don't know if they should be part of this PR or if there are others.

[leandrolanzieri]

Some others:

RIOT/cpu/esp_common/Makefile.include

Lines 54 to 56 in a665fcc

	# NOTE: These checks can be turned into normal conditionals when #9913 is fixed
	# use 32 priority levels if any WiFi interface or the ETH interface is used
	CFLAGS += $(if $(filter esp_wifi_any esp_eth,$(USEMODULE)),-DSCHED_PRIO_LEVELS=32)

RIOT/cpu/esp_common/Makefile.include

Lines 44 to 45 in a665fcc

	# override default CFLAGS_OPT in case module esp_gdb is enabled
	CFLAGS_OPT ?= $(if $(filter esp_gdb,$(USEMODULE)),-Og,-Os)

RIOT/cpu/esp_common/Makefile.include

Lines 135 to 137 in a665fcc

	# NOTE: These checks can be turned into normal conditionals when #9913 is fixed
	# increase the test timeout for file system tests that use the SPI flash drive
	RIOT_TEST_TIMEOUT = $(if $(filter spiffs littlefs,$(USEMODULE)),120)

RIOT/cpu/esp8266/Makefile.include

Lines 51 to 55 in a665fcc

	# NOTE: These can be turned into normal conditionals after #9913 is fixed
	GDBSTUB_DIR ?= $(if $(filter esp_gdbstub,$(USEMODULE)),$(RIOTCPU)/$(CPU)/vendor/esp-gdbstub)
	CFLAGS += $(if $(filter esp_gdbstub,$(USEMODULE)),-DGDBSTUB_BREAK_ON_INIT=1)
	INCLUDES += $(if $(filter esp_gdbstub,$(USEMODULE)),-I$(GDBSTUB_DIR))
	BASELIBS += $(if $(filter esp_now, $(USEMODULE)), -lespnow)

RIOT/cpu/esp32/Makefile.include

Lines 46 to 47 in a665fcc

	INCLUDES += $(if $(filter esp_eth,$(USEMODULE)),-I$(RIOTCPU)/$(CPU)/vendor/esp-idf/include/ethernet)
	INCLUDES += $(if $(filter esp_eth,$(USEMODULE)),-I$(ESP32_SDK_DIR)/components/ethernet/include)

RIOT/cpu/esp32/Makefile.include

Lines 67 to 69 in a665fcc

	# NOTE: These checks can be turned into a normal conditional when #9913 is fixed
	BASELIBS += $(if $(filter esp_wifi_any,$(USEMODULE)),$(_ESP_WIFI_ANY_BASELIBS))
	BASELIBS += $(if $(filter esp_now,$(USEMODULE)),$(_ESP_NOW_BASELIBS))

[fjmolinas]

@leandrolanzieri should I do those changes in this PR? IMO we could do it in a followup, and can come after the feature freeze.

[leandrolanzieri]

@leandrolanzieri should I do those changes in this PR? IMO we could do it in a followup, and can come after the feature freeze.

I'm OK with having that on a followup

[fjmolinas]

@leandrolanzieri should I do those changes in this PR? IMO we could do it in a followup, and can come after the feature freeze.

I'm OK with having that on a followup

I'll open right away.

[leandrolanzieri]

I'm running the scripts now...

[maribu]

Nice one :-)

[fjmolinas]

@leandrolanzieri found a diff where test_utils_interactive_sync was not included for a couple of applications.

[aabadie]

@leandrolanzieri found a diff where test_utils_interactive_sync was not included for a couple of applications.

Is this problematic from your point of view ?

[fjmolinas]

@leandrolanzieri I rand it again and I only get a diff in FEATURES_REQUIRED which comes from 9dea6a1 being removed and therefore there not being feature duplication. But FEATURES_REQUIRED_SORTED is all good.

[leandrolanzieri]

@leandrolanzieri found a diff where test_utils_interactive_sync was not included for a couple of applications.

This is not really an issue. It was happening because of my local setup. Disregard.

[leandrolanzieri]

@leandrolanzieri I rand it again and I only get a diff in FEATURES_REQUIRED which comes from 9dea6a1 being removed and therefore there not being feature duplication. But FEATURES_REQUIRED_SORTED is all good.

It would be good to post the diff for the record

[fjmolinas]

Here is the diff I had: diff.txt

[leandrolanzieri]

As explained in #9913 this change makes sense as it helps resolving the problem of setting certain configurations depending on the used modules.

@fjmolinas provided the diff that shows that there are no impacts in dependency resolution. From my side it's an ACK.

Maybe we can have a second one? @aabadie?

[aabadie]

Maybe we can have a second one?

I'm all for this change and I trust your results ! ACK

[leandrolanzieri]

And GO!

[fjmolinas]

Awesome to get this finally done, finishing what @cladmi started! Thanks for all the help along the way @leandrolanzieri @aabadie!

[cladmi]

Congrats