54 lines
1.6 KiB
Plaintext
54 lines
1.6 KiB
Plaintext
== Goal
|
|
|
|
A memory-safe language with a simple translator to x86 that can be feasibly written in x86.
|
|
|
|
== Definitions of terms
|
|
|
|
Memory-safe: it should be impossible to:
|
|
a) create a pointer out of arbitrary data, or
|
|
b) to access heap memory after it's been freed.
|
|
|
|
Simple: do all the work in a 2-pass translator:
|
|
Pass 1: check each instruction's types in isolation.
|
|
Pass 2: emit code for each instruction in isolation.
|
|
|
|
== types
|
|
|
|
int
|
|
char
|
|
(address _)
|
|
(array _ n)
|
|
(ref _)
|
|
|
|
== implications
|
|
|
|
addresses can't be saved to stack or global,
|
|
or included in compound types
|
|
or used across a call (to eliminate possibility of free)
|
|
|
|
<reg x> : (address T) <- advance <reg/mem> : (array T), <reg offset> : (index T)
|
|
|
|
arrays require a size
|
|
(ref array _) may not include a size
|
|
|
|
argv has type (array (ref array char))
|
|
|
|
variables on stack, heap and global are references. The name points at the
|
|
address. Use '*' to get at the value.
|
|
|
|
instructions performing lookups write to register, so that we can reuse the register for temporaries
|
|
instructions performing lookups can't read from the register they write to.
|
|
But most instructions read from the register they write to?! (in-out params)
|
|
|
|
== open questions
|
|
|
|
If bounds checks can take multiple instructions, why not perform array
|
|
indexing in a single statement in the language?
|
|
|
|
But we want addresses as intermediate points to combine instructions with.
|
|
|
|
Maybe disallow addresses to function calls, but allow addresses to non-heap
|
|
structures to be used spanning function calls and labels.
|
|
|
|
That's just for ergonomics. Doesn't add new capability.
|