gbenson/i8c

infinity note compiler

Infinity Note Compiler

I8C is now hosted at https://gitlab.com/gbenson/i8c.

i8c's Issues

i8c has no README.rst

i8c has no README.rst, and it should have one.

Once it's written set long_description in setup.py as per this example.

Files need a header with licence and stuff

SSIA

There are no tests for undefined/duplicate typedefs

SSIA

"call" instruction is untested

There are no tests for the call instruction.

There are no tests for parser errors

SSIA

i8c should honour $CC etc

i8c does not honour $CC etc, and it should. Presently it just uses "gcc" and there's no way to change this short of editing the code.

Stack walker does not support loops

Excepting very basic cases the stack walker does not support entering a BasicBlock that has already been entered. This means it is impossible to compile a note that contains a loop.

Add an xderef instruction

xderef could easily be implemented, it just isn't.

It could piggy-back on the code and tests for deref (only the stack walker would be different).

i8x is largely untested

SSIA

There are no tests for stack underflow detection

SSIA

"load" instruction with named slot argument is untested

load instructions with a constant value argument is pretty well tested, but there are no tests for load instructions that reference a named slot.

There are no tests for lexer errors

SSIA

Inconsistency with GCC

GCC outputs to "file.s" if invoked as gcc -S file.c . I8C should output to file.S if invoked as i8c -S file.i8 for consistency.

I8C can't compile straight to executable

i8c note.i8 main.c -o main doesn't do anything.

i8x stack errors are just assertions

i8x stack errors are just assertions (and whatever "pop from empty list" throws, for underflow). They should be proper checks that raise a subclass of ExecutionError. stack.op is always set for things to throw with (and set cleverly enough to do the right thing across calls), it just needs the checks writing.

There are no tests for optimizer misses

Optimizer hits are adequately tested, but various optimizer misses have no coverage.

i8c does not work with Python 3

i8c currently requires Python 2.6 or 2.7. This should be fixed... sometime :)

Source language is undocumented

The source language that i8c consumes is undocumented.

GPL not suitable for i8x?

i8x loads user code (testcases) written in Python into itself. If i8x is GPL then testcases must also be GPL. Maybe i8x should be LGPL (with i8c remaining GPL)

Most optimizations are untested

So far only the cmp-bra-{const,const} optimizer has a test. At least all block optimizers should have tests before the initial 0.0.1 release.

i8x header inclusion is very strict

i8x testcases can import constants from C-style header files, but the format required is very strict. Each line must be exactly #define NAME VALUE, and VALUE must be numeric. Not even comments are allowed.

deref sign extension bytecode is untested

deref s8, deref s16 and deref s32 emit a DW_OP_deref_size followed by a sequence of DWARF operations to sign-extend the dereferenced value.

This sign-extension sequence is untested in that it has never been executed by a DWARF interpreter.

Note format is undocumented

The format of the notes i8c emits is undocumented.

There are no tests for undefined/duplicate names (i.e. in load slot)

Can't really be fixed until issue #22 is done.

Clean up --debug=stack and add a user option for it

The hidden* option --debug=stack is pretty useful for figuring out stack walker errors. It's output should be cleaned up** and a user*** option added for it.

By "hidden option" I mean "it's not shown in i8c --help".

By "cleaned up" I mean, e.g., CompareOp("bne not_done_yet") is confusing. It should probably have the filename and line number and whatever the user typed

By "user option" I mean something like --trace. Something not in the --debug= namespace.

Input from stdin and output to stdout is untested

SSIA. Put two tests in test_compiler_driver.py

There are no tests for undefined/duplicate labels

SSIA

Need a --help command-line option

SSIA.

Assembly files on some platforms require extra directives

On some platforms, even assembler files with no actual code need to contain certain assembler directives to be ABI-compatible with those generated by the compiler. In particular, this applies to ARM; see the .eabi_attribute directives in sysdeps/arm/sysdep.h.

— Joseph Myers, https://sourceware.org/ml/libc-alpha/2015-10/msg00736.html

I8C needs to handle this somehow.

Type checking bytecode not fully possible?

Should cast actually emit something so bytecode can be type checked? If so deref needs it too (or some special deref_int with a type (signed byte, 0 means regular int, -8 means s8, etc) That would get rid of that sign-extension sequence too

Docstrings need work

Two issues:

1. Not everything that should have a docstring has one.
2. Existing docstrings mostly don't comply with PEP 257.

Need to do a docstring audit, perhaps using the pep357 static analysis tool.

Need to make sure arguments/externs have unique names

Something (the entry stack builder?) should check that arguments and externs have unique names. Currently it's not possible to shadow a name with name, but it is possible to start your function with shadowed names.

Lots of comments are just "XXX"

SSIA. Needs fixing for 1.0.0 release.