Objects containing `importc` fields without `completeStruct` fail to
compile when used as const/static. The C codegen generates "aggregate
initialization" which is invalid for opaque types.
Fixes#25405.
Nim code:
```nim
type
OpaqueInt {.importc: "_Atomic int", nodecl.} = object
ContainsImportc = object
normal: int
opaque: OpaqueInt
const c = default(ContainsImportc)
```
Resulting C code:
```c
// Invalid C - cannot aggregate-init opaque type
NIM_CONST ContainsImportc c = {((NI) 0), {}};
^^ error: illegal initializer type
```
## Solution
Fix in `ccgexprs.nim`:
1. Skip opaque importc fields when building aggregate initializers
2. Use "designated initializers" (`siNamedStruct`) when opaque fields
are present to avoid positional misalignment
```c
// Valid C:
// - opaque field is omitted and implicitly zero-initialized by C
// - other fields are explitly named and initialized
NIM_CONST ContainsImportc c = {.normal = ((NI) 0)};
```
This correctly handles the case where the opaque fields might be in any
order.
A field is considered "opaque importc" if:
- Has `sfImportc` flag
- Does NOT have `tfCompleteStruct` flag
- Either has `tfIncompleteStruct` OR is an object with no visible fields
The `containsOpaqueImportcField` proc recursively checks all object
fields, including nested objects and variant branches.
Anonymous unions (from variant objects) are handled by passing an empty
field name, which skips the `.fieldname = ` prefix since C anonymous
unions have no field name.
Note that initialization for structs without opaque importc fields
remains the same as before this changeset.
## Test Coverage
`tests/ccgbugs/timportc_field_init.nim` covers:
- Simple struct with one importc field
- Nested struct containing struct with importc field
- Variant object (case object) with importc field in a branch
- Array of structs with importc fields
- Tuple containing struct with importc field
- `completeStruct` importc types (still use aggregate init)
- Sandwich case (opaque field between two non-opaque fields)
- Fields with different C names (`{.importc: "c_name".}`, `{.exportc.}`)
- `{.packed.}` structs with opaque fields
- `{.union.}` types with opaque fields
- Deep nesting (3+ levels)
- Multiple opaque fields with renamed fields between them
TODO:
- [ ] test writing of .nif files
- [x] implement loading of fields in PType/PSym that might not have been
loaded
- [ ] implement interface logic
- [ ] implement pragma "replays"
- [ ] implement special logic for `converter`
- [ ] implement special logic for `method`
- [ ] test the logic holds up for `export`
- [ ] implement logic to free the memory of PSym/PType if memory
pressure is high
- [ ] implement logic to close memory mapped files if too many are open.
---------
Co-authored-by: demotomohiro <gpuppur@gmail.com>
Co-authored-by: ringabout <43030857+ringabout@users.noreply.github.com>
Co-authored-by: Jacek Sieka <arnetheduck@gmail.com>
fixes#25265;
fixes#23453
`(addr deref (ptr object))` generated weak typedesc before, which causes
problems for old GCC versions. As a bonus, by generating a typedesc for
`deref (ptr object)`, it also fixes#23453
fixes#25007
```nim
proc setLengthSeqUninit(s: PGenericSeq, typ: PNimType, newLen: int, isTrivial: bool): PGenericSeq {.
compilerRtl.} =
```
In this added function, only the line `zeroMem(dataPointer(result,
elemAlign, elemSize, newLen), (result.len-%newLen) *% elemSize)` is
removed from `proc setLengthSeqV2` when enlarging a sequence.
JS and VM versions simply use `setLen`.
fixes#17571
Objects in the VM are represented as object constructor nodes that
contain every single field, including ones in different case branches.
This is so that every field has a unique invariant index in the object
constructor that can be written to and read from. However when
converting this node back into semantic code, fields from inactive case
branches can remain in the constructor which causes bad codegen,
generating assignments to fields from other case branches.
To fix this, fields from inactive branches are now detected in
`semmacrosanity.annotateType` (called in `fixupTypeAfterEval`) and
marked to prevent the codegen of their assignments. In #24441 these
fields were excluded from the resulting node, but this causes issues
when the node is directly supposed to go back into the VM, for example
as `const` values. I don't know if this is the only case where this
happens, so I wasn't sure about how to keep that implementation working.
Doing this early is useful so we can move the indentation logic into
`Builder` itself rather than mix it with the block logic in `ccgstmts`
(the `if` statements in #24381 have not been indented properly either).
However it also means `Builder` is now used for code that still
generates raw C code, so the diff won't be as clean when these get
updated.
follows up #24381
This is about another 30% of ccgexprs (remaining is up to around line
3000), stopped right before the point where #24391 is required.
The `Genode::log` and `Genode::Cstring` calls in `genEcho` are left as
is but could be mangled in the future.
Most of what ccgexprs uses is now ported to cbuilder, so this PR makes
around ~25% of ccgexprs use it, along with adding `if` stmts (no
`while`/`switch` and `for` which is only used as `for (tmp = a; tmp < b;
tmp++)`). The `if` builder does not add indents for blocks since we
can't make `Builder` an object yet rather than an alias to `string`,
this will likely be one of the last refactors.
Somewhat unrelated but `ccgtypes` is not ready yet because proc
signatures are not implemented.
The types of the typed operations are not necessarily correct, the mixed
types of the operands in some cases are left the same to keep the C
integer promotion working as expected. To fix any wrong types the
integer promotion can be done explicitly instead, but this is a behavior
change to codegen.
Different from `intLiteral`, we add procs that just try to generate
integer values, assuming they're not an edge case like `<= low(int32)`
or `> high(int32)`.
This PR is somewhat large, worst case it can be split into one with just
assignments and one with just fields/derefs etc.
Assignments with calls as values have not been touched so they can be
done when calls are implemented. Similarly codegen with complex logic
i.e. `genEnumInfo`, `genTypeInfoV2`, `unaryExpr` is not completely
ported yet so they can be done in standalone PRs.
closes https://github.com/nim-lang/RFCs/issues/554
Adds a symmetric difference operation to the language bitset type. This
maps to a simple `xor` operation on the backend and thus is likely
faster than the current alternatives, namely `(a - b) + (b - a)` or `a +
b - a * b`. The compiler VM implementation of bitsets already
implemented this via `symdiffSets` but it was never used.
The standalone binary operation is added to `setutils`, named
`symmetricDifference` in line with [hash
sets](https://nim-lang.org/docs/sets.html#symmetricDifference%2CHashSet%5BA%5D%2CHashSet%5BA%5D).
An operator version `-+-` and an in-place version like `toggle` as
described in the RFC are also added, implemented as trivial sugar.
The remaining followup from #24259. A body for building the type doesn't
seem necessary here since the types with array fields are generally
atomic/already built from `getTypeDescAux`.
fixes#24319
`byRefLoc` (`mapType`) requires the Loc `a` to have the right type.
Without `lfEnforceDeref`, it produces the wrong type for `deref (var
array)`, which may come from `mitems`.
follows up #24259
This is the remaining missing use of `StructInitializer` in
`getDefaultValue` after #24259 and #24302. The only remaining direct C
code in getDefaultValue is [this
line](922f7dfd71/compiler/ccgexprs.nim (L3525))
which creates a global array variable, which isn't implemented yet. Next
steps would be all remaining variable and `typedef` declarations, then
hopefully we can move on to general statements and expressions.
`StructInitializer` is now used for most braced initializers in the C
generation, mostly in `genBracedInit`, `getNullValueAux`,
`getDefaultValue`. The exceptions are:
* the default case branch initializer for objects uses C99 designated
initializers with field names, which are not implemented for
`StructInitializer` yet (`siNamedStruct`)
* the uses in `ccgliterals` are untouched so all of ccgliterals can be
done separately and in 1 go
There is one case where `genBracedInit` does not use cbuilder, which is
the global literal variable for openarrays. The reason for this is
simply that variables with C array type are not implemented, which I
thought would be best to leave out of this PR.
For the simplicity of the implementation, code in `getNullValueAuxT`
that reset the initializer back to its initial state if the `Sup` field
did not have any fields itself, is now disabled. This was so the
compiler does not generate `{}` for the Sup field, i.e. `{{}}`, but
every call to `getNullValueAuxT` still generates `{}` if the struct
doesn't have any fields, so I don't know if it really breaks anything.
The case where the Sup field doesn't have any fields but the struct does
also would have generated `{{}, field}`.
Worst case, we can implement either the "resetting" or just disable the
generation of the `Sup` field if there are no fields total. But a better
fix might be to always generate `{0}` if the struct has no fields, in
line with the `char dummy` field that gets added for all objects with no
fields. This doesn't seem necessary for now but might be for the NIFC
output, in which case we can probably keep the logic contained inside
cbuilder (if no fields generated for `siOrderedStruct`/`siNamedStruct`,
we add a `0` for the `dummy` field). This would stipulate that all uses
of struct initializers are exhaustive for every field in structs.
fixes#24274
The code in the `if` branch replaces the current destination `d` with a
new one. But the location `d` can be an assignment location, in which
case the provided expression isn't generated. To fix this, don't trigger
this code for when the location already exists. An alternative would be
to call `putIntoDest` in this case as is done below.
refs #24032, split from #24036
Conversion from variables of range types or base types of range types to
the other are now considered mutable for `var` params, similar to how
distinct types are mutable when converted to their base type or vice
versa. There are 2 main differences:
1. Conversions from base types to range types need to emit
`nkChckRange`, which is not generated for things like tuple/object
fields.
2. Range types can still correspond to different types in the backend
when nested in other types, such as `set[range[3..5]]` vs
`set[range[0..5]]`.
Since the convertibility check for `var` params and a check whether to
emit a no-op for `nkConv` (and now also `nkChckRange`) so that the
output is still addressable both use `sameType`, we accomplish this by
adding a new flag to `sameType` that ignores range types, but only when
they're not nested in other types. The implementation for this might be
flawed, I didn't include children of some metatypes as "nested in other
types", but stuff like `tyGenericInst` params are respected.
ref #20653
```nim
Error* = object
case kind*: ErrorType
of ErrorA:
discard
of ErrorB:
discard
```
For an object variants without fields, it shouldn't generate empty
brackets for default values since there are no fields at all in case
branches.
fixes#23627
```nim
type
TestObj = object of RootObj
TestTestObj = object of RootObj
testo: TestObj
proc `=destroy`(x: TestTestObj) =
echo "Destructor for TestTestObj"
proc testCaseT() =
echo "\nTest Case T"
let tt1 {.used.} = TestTestObj(testo: TestObj())
```
When generating const object fields, it's likely that
we need to generate type infos for the object, which may be an object
with
custom hooks. We need to generate potential consts in the hooks first.
https://github.com/nim-lang/Nim/pull/20433 changed the semantics of
initialization. It should evaluate`BracedInit` first.
`reset`, `wasMoved` and `move` doesn't support primitive types, which
generate `null` for these types. It is now produce `x = default(...)` in
the backend. Ideally it should be done by ast2ir in the future