Closures are not necessarily constant values. For example, Zig
code might do something like this:
fn foo(x: anytype) void {
const S = struct {field: @TypeOf(x)};
}
...in which case the closure_capture instruction has access to a
runtime value only. In such case we preserve the type and use a
dummy runtime value.
closes#11292
This also fixes a bug that I didn't see causing any problems yet in
generic function instantiation where it would read from a GetOrPutResult
too late.
Also it delays full resolution of generic function type parameters until
after the function body is finished being analyzed.
closes#11291
The code for detecting when a local const initialization expression
ended up being comptime-known gave up when it encountered dbg_stmt
instructions, but such instructions are not supposed to matter.
* std.meta: correct use of `default_value` in reification. stage1
accepted a wrong type for `null`.
* Sema: after instantiating a generic function, if the return type ends
up being a comptime-known type, then we return an error, undoing the
generic function instantiation, and making a comptime function call
instead.
- We also needed to clean up the dependency graph in this case.
* Sema: reified enums set tag_ty_inferred to false since an integer tag
type is provided. This is a limitation of the `@Type` builtin which
will be addressed with #10710.
* Sema: fix resolveInferredErrorSet incorrectly calling
ensureFuncBodyAnalyzed on generic functions.
That happens after a function body is analyzed. This prevents circular
dependency compile errors and yet a way to mark types that need to be
fully resolved before a given function is sent to the codegen backend.
The runtime behavior allowed this in both stage1 and stage2, but stage1
fails with index out of bounds during comptime. This behavior makes
sense to support, and comptime behavior should match runtime behavior. I
implement this fix only in stage2.
We were enforcing bounds on the index of an elem_ptr in pointerDeref,
but we want to support out-of-bounds accesses by reinterpreting memory.
This removes that check, so that the deref falls back to bitcasting, as
usual.
This was masked by another bug that was forcing bitcasts incorrectly,
which is why this wasn't noticed earlier.
Sema.pointerDeref() assumes that elem_ptrs have been "flattened" when
they were created, so that you an elem_ptr will never be the array_ptr
of another elem_ptr when they share the same type.
Value.elemPtr does this already, but a couple of places in Sema were
bypassing this logic.
This flag is used when building stage1 to omit the stage2 backends from
the compiler to save memory on the CI server. It regressed with the
merging of e8813b296b because Value
functions started calling into Sema functions.
The end goal for this build option is to eliminate it.
This commit adds a new optional argument to several Value methods which
provides the ability to resolve types if it comes to it. This prevents
having duplicated logic inside both Sema and Value.
With this commit, the "struct contains slice of itself" test is passing
by exploiting the new lazy_align Value Tag.
Also update std/build.zig to use stage2 function pointer semantics.
This gets us a little bit closer to `zig build` working, although it is
now hitting a new crash in the compiler.
Made most `Value` functions require a `Type`. If the provided type is a
vector, then automatically vectorize the operation and return with
another vector. The Sema side can then automatically become vectorized
with minimal changes. There are already a few manually vectorized
instructions, but we can simplify those later.
The existing `cmp_*` instructions get their result type from `lhs`, but
vector comparison will always return a vector of bools with only the
length derived from its operands. This necessitates the creation of a
new AIR instruction.
Example scenario:
test {
const a: i32 = blk: {
if (false) break :blk 24;
};
_ = a;
}
Prior to this, this would panic the compiler with a source needed error.
This provides the source as `sema.src`. This is not ideal, since the
line it points to is pretty far from the true issue. (One block out)
But, this prevents the compiler from straight up crashing and follows a
pattern used by similar ZIR which don't provide a src loc.
* make it always return a fully qualified name. stage1 is inconsistent
about this.
* AstGen: fix anon_name_strategy to correctly be `func` when anon type
creation happens in the operand of the return expression.
* Sema: implement type names for the "function" naming strategy.
* Put "enum", "union", "opaque", or "struct" in place of "anon" when
creating respective anonymous Decl names.
* std.testing: add `expectStringStartsWith`. Didn't end up using it
after all.
Also this enables the real test runner for stage2 LLVM backend (sans
wasm32) since it works now.
Instead of doing it before the switch tower, do it afterwards, so that
special handling may be done before undefined gets casted to the
destination type.
In this case the special handling we want to do is *[N]T to []T setting
the slice length based on the array length, even when the array value is
undefined.
When a generic call evaluates to a generic type, the call will be re-generated.
However, the old function was not freed before being re-generated, causing a memory leak.
So rather than only returning an error, we first free the old value.
Notably, Value.eql and Value.hash are improved to treat NaN as equal to
itself, so that Type/Value can be hash map keys. Likewise float hashing
normalizes the float value before computing the hash.
* Sema: fix `zirTypeInfo` allocating with the wrong arenas for some
stuff.
* LLVM: split `airDbgInline` into two functions, one for each AIR tag.
- remove the redundant copy to type_map_arena. This is the first
thing that lowerDebugType does so this hack was probably just
accidentally avoiding UB (which is still present prior to this
commit).
- don't store an inline fn inst into the di_map for the generic
decl.
- use a dummy function type for the debug info to avoid whatever UB
is happening.
- we are now ignoring the function type passed in with the
dbg_inline_begin and dbg_inline_end.
* behavior tests: prepare the vector tests to be enabled one at a time.
Mitigates #11199.
I think that reusing the ComptimePtrLoad infrastructure is ultimately
less logic and more robust than adding a `direct` flag to elem_ptr.
* Some code in zirTypeInfo needed to be fixed to create proper
Type/Value encodings.
* comptime elemVal works by constructing an elem_ptr Value and then
using the already existing pointerDeref function.
There are some remaining calls to Value.elemValue which should be
considered code smells at this point.
