`make arch-test` fails unless `ENABLE_ARCH_TEST` is explicitly set

[vestata]

I encountered an issue on an x86_64 machine(Ubuntu 24.04) when running make arch-test directly. The error output is as follows:

$ make arch-test 
  AR    build/softfloat/softfloat.a
  CC    build/map.o
  CC    build/utils.o
  CC    build/decode.o
  CC    build/io.o
  CC    build/syscall.o
  CC    build/emulate.o
  CC    build/riscv.o
  CC    build/log.o
  CC    build/elf.o
  CC    build/cache.o
  CC    build/mpool.o
  CC    build/syscall_sdl.o
  CC    build/main.o
  LD    build/rv32emu
Fetching SHA-1 of prebuilt binaries ... [OK]
Checking SHA-1 of prebuilt binaries ... [OK]
git submodule update --init tests/
cp: cannot stat 'build/rv32emu-prebuilt-sail-x86': No such file or directory
make: *** [mk/riscv-arch-test.mk:23: arch-test] Error 1

PR #547 introduced the ARCH_TEST feature.
In CI environment (.ci/riscv-tests.sh), this problem doesn't occur because ENABLE_ARCH_TEST is explicitly set to 1 before running make arch-test. However, the default ENABLE_ARCH_TEST is set to 0. As a result, if the repository is cloned or make distclean is executed and the rv32emu-prebuilt-sail-$(HOST_PLATFORM) file has not been downloaded previously, users who manually run make arch-test will encounter the error shown above.

When ENABLE_ARCH_TEST is set to 0, the condition for fetching rv32emu-prebuilt-sail-$(HOST_PLATFORM) is not met, and therefore the necessary file is not fetched.
The relevant snippet in mk/artifact.mk:

ifeq ($(call has, PREBUILT), 1)
    ifeq ($(call has, SYSTEM), 1)
        $(call fetch-releases-tag,Linux-Image,rv32emu-linux-image-prebuilt.tar.gz,Linux image)
    else ifeq ($(call has, ARCH_TEST), 1)
        $(call fetch-releases-tag,sail,rv32emu-prebuilt-sail-$(HOST_PLATFORM),Sail model)
    else
        $(call fetch-releases-tag,ELF,rv32emu-prebuilt.tar.gz,Prebuilt blob)
    endif

    PREBUILT_BLOB_URL = https://github.com/sysprog21/rv32emu-prebuilt/releases/download/$(LATEST_RELEASE)

Maybe we could:

1. Update README.md to mention that users should set ENABLE_ARCH_TEST=1 when running make arch-test. For example:

make arch-test ENABLE_ARCH_TEST=1

2. Modify Makefile to automatically set ENABLE_ARCH_TEST=1 when attempts to run make arch-test.

Related to #506

[vacantron]

When ENABLE_ARCH_TEST is set to 0, the condition for fetching rv32emu-prebuilt-sail-$(HOST_PLATFORM) is not met, and therefore the necessary file is not fetched.
The relevant snippet in mk/artifact.mk:

rv32emu/mk/artifact.mk

Lines 62 to 68 in 59fb73f

	ifeq ($(call has, SYSTEM), 1)
	$(call fetch-releases-tag,Linux-Image,rv32emu-linux-image-prebuilt.tar.gz,Linux image)
	else ifeq ($(call has, ARCH_TEST), 1)
	$(call fetch-releases-tag,sail,rv32emu-prebuilt-sail-$(HOST_PLATFORM),Sail model)
	else
	$(call fetch-releases-tag,ELF,rv32emu-prebuilt.tar.gz,Prebuilt benchmark)
	endif

If we build it with make ENABLE_SYSTEM=1 first and then make arch-test ENABLE_ARCH_TEST=1, we will fetch the Kernel image instead of the Sail model. We need to run make distclean or delete build/.config to clear the old configuration that we could be able to fetch the correct binary.

Do we explicitly add make distclean in the guide?

[vestata]

Adding make distclean before make arch-test ENABLE_ARCH_TEST=1 to the README guidelines is an effective temporary solution.

Additionally, would it be more robust to replace the current if-else structure with separate ifeq statements for SYSTEM and ARCH_TEST in artifact.mk? This could help prevent unintended dependencies when switching configurations.

For instance, we could refactor the condition checks along with other relevant checks in artifact.mk:

rv32emu/mk/artifact.mk

Lines 61 to 68 in 59fb73f

	ifeq ($(LATEST_RELEASE),)
	ifeq ($(call has, SYSTEM), 1)
	$(call fetch-releases-tag,Linux-Image,rv32emu-linux-image-prebuilt.tar.gz,Linux image)
	else ifeq ($(call has, ARCH_TEST), 1)
	$(call fetch-releases-tag,sail,rv32emu-prebuilt-sail-$(HOST_PLATFORM),Sail model)
	else
	$(call fetch-releases-tag,ELF,rv32emu-prebuilt.tar.gz,Prebuilt benchmark)
	endif

[vacantron]

Additionally, would it be more robust to replace the current if-else structure with separate ifeq statements for SYSTEM and ARCH_TEST in artifact.mk? This could help prevent unintended dependencies when switching configurations.

These options could not be used at the same time, but our build system won't capture this problem. Maybe #473 can solve it.

I think we just always runing make distclean after updating the configuration as the temporary solution.

[vestata]

These options could not be used at the same time, but our build system won't capture this problem. Maybe #473 can solve it.

Thanks for the clarification. I will submit a PR to update the README with the make distclean step as a temporary solution. Let me know if you have any suggestions.