Next up is reworking the seam between the LazySrcLoc emitted by Sema
and the byte offsets currently expected by codegen.
And then the big one: updating astgen.zig to use the new memory layout.
The memory layout for ZIR instructions is completely reworked. See
zir.zig for those changes. Some new types:
* `zir.Code`: a "finished" set of ZIR instructions. Instead of allocating
each instruction independently, there is now a Tag and 8 bytes of
data available for all ZIR instructions. Small instructions fit
within these 8 bytes; larger ones use 4 bytes for an index into
`extra`. There is also `string_bytes` so that we can have 4 byte
references to strings. `zir.Inst.Tag` describes how to interpret
those 8 bytes of data.
- This is shared by all `Block` scopes.
* `Module.WipZirCode`: represents an in-progress `zir.Code`. In this
structure, the arrays are mutable, and get resized as we add/delete
things. There is extra state to keep track of things. This struct is
stored on the stack. Once it is finished, it produces an immutable
`zir.Code`, which will remain on the heap for the duration of a
function's existence.
- This is shared by all `GenZir` scopes.
* `Sema`: represents in-progress semantic analysis of a `zir.Code`.
This data is stored on the stack and is shared among all `Block`
scopes. It is now the main "self" argument to everything in the file
that was previously named `zir_sema.zig`.
Additionally, I moved some logic that was in `Module` into here.
`Module.Fn` now stores its parameter names inside the `zir.Code`,
instead of inside ZIR instructions. When the TZIR memory layout
reworking time comes, codegen will be able to reference this data
directly instead of duplicating it.
astgen.zig is (so far) almost entirely untouched, but nearly all of it
will need to be reworked to adhere to this new memory layout structure.
I have no benchmarks to report yet, as I am still working through
compile errors and fixing various things that I broke in this branch.
Overhaul of Source Locations:
Previously we used `usize` everywhere to mean byte offset, but sometimes
also mean other stuff. This was error prone and also made us do
unnecessary work, and store unnecessary bytes in memory.
Now there are more types involved into source locations, and more ways
to describe a source location.
* AllErrors.Message: embrace the assumption that files always have less
than 2 << 32 bytes.
* SrcLoc gets more complicated, to model more complicated source
locations.
* Introduce LazySrcLoc, which can model interesting source locations
with very little stored state. Useful for avoiding doing unnecessary
work when no compile errors occur.
Also, previously, we had `src: usize` on every ZIR instruction. This is
no longer the case. Each instruction now determines whether it even cares
about source location, and if so, how that source location is stored.
This requires more careful work inside `Sema`, but it results in fewer
bytes stored on the heap, without compromising accuracy and power of
compile error messages.
Miscellaneous:
* std.zig: string literals have more helpful result values for
reporting errors. There is now a lower level API and a higher level
API.
- side note: I noticed that the string literal logic needs some love.
There is some unnecessarily hacky code there.
* cut & pasted some TZIR logic that was in zir.zig to ir.zig. This
probably broke stuff and needs to get fixed.
* Removed type/Enum.zig, type/Union.zig, and type/Struct.zig. I don't
think this quite how this code will be organized. Need some more
careful planning about how to implement structs, unions, enums. They
need to be independent Decls, just like a top level function.
on the break instruction operands. This involves a new TZIR instruction,
br_block_flat, which represents a break instruction where the operand is
the result of a flat block. See the doc comments on the instructions for
more details.
How it works: when adding break instructions in semantic analysis, the
underlying allocation is slightly padded so that it is the size of a
br_block_flat instruction, which allows the break instruction to later
be converted without removing instructions inside the parent body. The
extra type coercion instructions go into the body of the br_block_flat,
and backends are responsible for dispatching the instruction correctly
(it should map to the same function calls for related instructions).
Motivating test case:
```zig
export fn _start() noreturn {
var x: u64 = 1;
var y: u32 = 2;
var thing: u32 = 1;
const result = if (thing == 1) x else y;
exit();
}
```
The main idea here is for astgen to output ideal ZIR depending on
whether or not the sub-expressions of a block consume the result
location. Here, neither `x` nor `y` consume the result location of the
conditional expression block, and so the ZIR should communicate the
result of the condbr using break instructions, not with the result
location pointer.
With this commit, this is accomplished:
```
%22 = alloc_inferred()
%23 = block({
%24 = const(TypedValue{ .ty = type, .val = bool})
%25 = deref(%18)
%26 = const(TypedValue{ .ty = comptime_int, .val = 1})
%27 = cmp_eq(%25, %26)
%28 = as(%24, %27)
%29 = condbr(%28, {
%30 = deref(%4)
< there is no longer a store instruction here >
%31 = break("label_23", %30)
}, {
%32 = deref(%11)
< there is no longer a store instruction here >
%33 = break("label_23", %32)
})
})
%34 = store_to_inferred_ptr(%22, %23) <-- the store is only here
%35 = resolve_inferred_alloc(%22)
```
However if the result location gets consumed, the break instructions
change to break_void, and the result value is communicated only by the
stores, not by the break instructions.
Implementation:
* The GenZIR scope that conditional branches uses now has an optional
result location pointer field and a count of how many times the
result location ended up being an rvalue (not consumed).
* When rvalue() is called on a result location for a block, it
increments this counter. After generating the branches of a block,
astgen for the conditional branch checks this count and if it is 2
then the store_to_block_ptr instructions are elided and it calls
rvalue() using the block result (which will account for peer type
resolution on the break operands).
astgen has many functions disabled until they can be reworked with these
new semantics. That will be done before merging the branch.
There are some new rules for astgen to follow regarding result locations
and what you are allowed/required to do depending on which one is passed
to expr(). See the updated doc comments of ResultLoc for details.
I also changed naming conventions of stuff in this commit, sorry about
that.
* fix wrong pointer const-ness when unwrapping optionals
* allow grouped expressions and orelse as lvalues
* ZIR for unwrapping optionals: no redundant deref
- add notes to please don't use rlWrapPtr, this function should be
deleted
* catch and orelse: better ZIR for non-lvalue: no redundant deref;
operate entirely on values. lvalue case still works properly.
- properly propagate the result location into the target expression
* Test harness: better output when tests fail due to compile errors.
* TZIR: add instruction variants. These allow fewer TZIR instructions to
be emitted from zir_sema. See the commit diff for per-instruction
documentation.
- is_null
- is_non_null
- is_null_ptr
- is_non_null_ptr
- is_err
- is_err_ptr
- optional_payload
- optional_payload_ptr
* TZIR: removed old naming convention instructions:
- isnonnull
- isnull
- iserr
- unwrap_optional
* ZIR: add instruction variants. These allow fewer ZIR instructions to
be emitted from astgen. See the commit diff for per-instruction
documentation.
- is_non_null
- is_null
- is_non_null_ptr
- is_null_ptr
- is_err
- is_err_ptr
- optional_payload_safe
- optional_payload_unsafe
- optional_payload_safe_ptr
- optional_payload_unsafe_ptr
- err_union_payload_safe
- err_union_payload_unsafe
- err_union_payload_safe_ptr
- err_union_payload_unsafe_ptr
- err_union_code
- err_union_code_ptr
* ZIR: removed old naming convention instructions:
- isnonnull
- isnull
- iserr
- unwrap_optional_safe
- unwrap_optional_unsafe
- unwrap_err_safe
- unwrap_err_unsafe
- unwrap_err_code
This patch introduces the following new things:
Types:
- inferred_alloc
- This is a special value that tracks a set of types that have been stored
to an inferred allocation. It does not support most of the normal type queries.
However it does respond to `isConstPtr`, `ptrSize`, `zigTypeTag`, etc.
- The payload for this type simply points to the corresponding Value
payload.
Values:
- inferred_alloc
- This is a special value that tracks a set of types that have been stored
to an inferred allocation. It does not support any of the normal value queries.
ZIR instructions:
- store_to_inferred_ptr,
- Same as `store` but the type of the value being stored will be used to infer
the pointer type.
- resolve_inferred_alloc
- Each `store_to_inferred_ptr` puts the type of the stored value into a set,
and then `resolve_inferred_alloc` triggers peer type resolution on the set.
The operand is a `alloc_inferred` or `alloc_inferred_mut` instruction, which
is the allocation that needs to have its type inferred.
Changes to the C backend:
* Implements the bitcast instruction. If the source and dest types
are both pointers, uses a cast, otherwise uses memcpy.
* Tests are run with -Wno-declaration-after-statement. Someday we can
conform to this but not today.
In ZIR form it looks like this:
```zir
fn_body main { // unanalyzed
%0 = dbg_stmt()
=>%1 = alloc_inferred()
%2 = declval_in_module(Decl(add))
%3 = deref(%2)
%4 = param_type(%3, 0)
%5 = const(TypedValue{ .ty = comptime_int, .val = 1})
%6 = as(%4, %5)
%7 = param_type(%3, 1)
%8 = const(TypedValue{ .ty = comptime_int, .val = 2})
%9 = as(%7, %8)
%10 = call(%3, [%6, %9], modifier=auto)
=>%11 = store_to_inferred_ptr(%1, %10)
=>%12 = resolve_inferred_alloc(%1)
%13 = dbg_stmt()
%14 = ret_type()
%15 = const(TypedValue{ .ty = comptime_int, .val = 3})
%16 = sub(%10, %15)
%17 = as(%14, %16)
%18 = return(%17)
} // fn_body main
```
I have not played around with very many test cases yet. Some interesting
ones that I want to look at before merging:
```zig
var x = blk: {
var y = foo();
y.a = 1;
break :blk y;
};
```
In the above test case, x and y are supposed to alias.
```zig
var x = if (bar()) blk: {
var y = foo();
y.a = 1;
break :blk y;
} else blk: {
var z = baz();
z.b = 1;
break :blk z;
};
```
In the above test case, x, y, and z are supposed to alias.
I also haven't tested with `var` instead of `const` yet.
