Support pointer data types for sizeof operator

[ChAoSUnItY]

Currently sizeof is only capable of parsing basic data types like int, char, or user-defined data types like struct, enum etc. But not able to parsing pointer type, thus, in this PR, by introducing a temporary workaround parsing 0 or more asterisk tokens, we now partially support pointer type for sizeof operator (not including function pointer type yet).

Additionally, since the result given by HOST_PTR_SIZE in stage 0 and stage 1/2 are different, so I make another test driver named driver-stage2.sh for testing stage 2 shecc and adding corresponding sizeof test to verify the implementation.

[ChAoSUnItY]

I'm unable to reproduce broken pipe failure on my machine, and the snapshots I checked which are generated by tests/update-snapshots.sh are same as the original one.

[DrXiao]

Because this pull request involves size of pointers, I would like to ask a question.

In the header file (def.h), we can notice that there are two macros called PTR_SIZE (Line 47) and HOST_PTR_SIZE (Line 59 ~ 64). Obviously, both of them mean the size of a pointer, but I cannot understand their use and difference all the time, especially the later one (HOST_PTR_SIZE).

Thus, I hope to obtain an explanation about them through this opportunity.

[ChAoSUnItY]

In the header file (def.h), we can notice that there are two macros called PTR_SIZE (Line 47) and HOST_PTR_SIZE (Line 59 ~ 64). Obviously, both of them mean the size of a pointer, but I cannot understand their use and difference all the time, especially the later one (HOST_PTR_SIZE).

The HOST_PTR_SIZE is determined by the running host, while PTR_SIZE is supposed to be 4, which is capable of handling the arch we're targeting to. The only usages of HOST_PTR_SIZE currently are memcpy in src/ssa.c.

Which I later realized value of sizeof(type*) should not be HOST_PTR_SIZE, instead, it should be PTR_SIZE.

[DrXiao]

@ChAoSUnItY , after fetching your latest changes, I try to build shecc and check the snapshots:

$ make distclean && make ARCH=arm && tests/check-snapshots.sh 
...
  SHECC	out/shecc-stage1.elf
  SHECC	out/shecc-stage2.elf

$ make distclean && make ARCH=riscv && tests/check-snapshots.sh 
...
  SHECC	out/shecc-stage1.elf
  SHECC	out/shecc-stage2.elf
Files /dev/fd/63 and /dev/fd/62 differ
Files /dev/fd/63 and /dev/fd/62 differ

It seems that the snapshots differs when shecc uses RISC-V backend. Hope this information helps you.

[ChAoSUnItY]

Ok, I found the cause, apparently previously shecc's CI won't check snapshot on rv32 but only on arm32:

shecc/Makefile

Lines 55 to 59 in 18e005d

	check: $(TESTBINS) tests/driver.sh
	tests/driver.sh
	check-snapshots: $(OUT)/$(STAGE0) $(SNAPSHOTS) tests/check-snapshots.sh
	tests/check-snapshots.sh

shecc/.github/workflows/main.yml

Lines 6 to 27 in 18e005d

	host-x86:
	runs-on: ubuntu-22.04
	strategy:
	matrix:
	compiler: [gcc-12, clang]
	steps:
	- name: checkout code
	uses: actions/checkout@v4
	- name: build artifact
	env:
	CC: ${{ matrix.compiler }}
	run: |
	sudo apt-get update -q -y
	sudo apt-get install -q -y graphviz jq
	sudo apt-get install -q -y qemu-user
	sudo apt-get install -q -y build-essential
	make clean config
	make check-snapshots || exit 1
	make distclean config ARCH=arm
	make check || exit 1
	make distclean config ARCH=riscv
	make check || exit 1

So currently the workaround is just not to run check-snapshot on rv32.

Should we later address this issue in another PR? @jserv

[jserv]

Ok, I found the cause, apparently previously shecc's CI won't check snapshot on rv32 but only on arm32:

That was a fault. Let's fix it in another pull request.

[ChAoSUnItY]

I will wrap up some additional changes introduced besides sizeof operator implementation enhancement:

Previously, while using stage 2 compiler to compile division-related test suites in tests/driver.sh, the arm32 suffers bus error, and the rv32 suffers infinite loop.

The former one was caused by improper stack alignment, while the later one was caused by missing edge case equality check, which is INT_MIN. Thus, I have added stack_align function in reg-alloc.c to ensure every stack increment is properly aligned, and I have added a branching instruction for the edge case.

The change of total stack size of shecc itself has increased from 9225 to 9228.

[jserv]

I defer to @vacantron for confirmation.

[vacantron]

LGTM

[jserv]

Thank @ChAoSUnItY for contributing!