Turns out the tests for 'trace' have been broken in native mode since the
original commit (4674). Dangers of running my tests on Darwin, where I
can't run them natively.
The test failures didn't get flagged on CI because I'd forgotten to update
the exit code of the factorial app in commit 4664. At least that's fixed
in this commit.
I'm still trying to figure out what the defaults should be. At the moment
you have to explicitly pass in every file you want loaded into the output
binary. Maybe that control is a good thing. The examples need no libraries
so far.
Start requiring a '-o' flag to designate the output binary when translating.
Things currently get funky if you pass in multiple inputs, but that's ok.
This is the first step to supporting multiple input files for a single
output binary.
check_ints_equal now prints a newline after the failure message on failure.
We still don't know how to print a final newline after all the tests have
run, for the common case when all tests pass.
Ideally I could just emit a few instructions to `run_tests`. But I'd also
need to add a variable for the newline to the data segment. Or I need some
literal syntax for newlines in strings. We don't have support for backslash-escapes
yet.
Extract a helper from the factorial unit test: check_ints_equal.
Start of a vocabulary for unit tests.
I *could* also start thinking of supporting multi-file programs, but I'm
going to resist the temptation for now. Copy helpers as necessary, and
allow them to mutate and diverge for a while before we pummel them into
a Procrustean "standard library". Extracting a body of shared code immediately
starts to discourage innovation in the shared code.
Doesn't de-duplicate in the data segment, though. If you use the literal
"foo" a hundred times in your code segment you're gonna spend a hundred
times the space you need to.
We can now simplify our test harness a bit in the factorial app, but we
still have to put in commandline args to compare with manually. We only
support length-prefixed strings, not null-terminated ones.
Even more cuddling. We want to keep lines short where the opcode and operands
are self-explanatory.
If there are any implicit registers, etc., we'll continue to do the table
layout.
The first two columns look messy now; let's see how this goes.
Maybe I'll give up on the tabular layout altogether, just string args with
a single space.
All it takes is to code-generate a simple function called 'run_tests' that
calls all functions starting with 'test_' one by one.
I've temporarily switched the factorial app to run as a test. But that's
temporary, because all the code to print '.' vs 'F' needs to get extracted
out into a helper.
Streamline the factorial function; we don't need to save a stack variable
into a register before operating on it. All instructions can take a stack
variable directly.
In the process we found two bugs:
a) Opcode f7 was not implemented correctly. It was internally consistent
but I'd never validated it against a natively running program. Turns out
it encodes multiple instructions, not just 'not'.
b) The way we look up imm32 operands was sometimes reading them before
disp8/disp32 operands.
Kartik Agaram