The LLVM bitcode requires all type references in
structs to be to earlier defined types.
We make sure types are ordered in the builder
itself in order to avoid having to iterate the
types multiple times and changing the values
of type indicies.
implement builtin.zig file population for all modules rather than
assuming there is only one global builtin.zig module.
move some fields from link.File to Compilation
move some fields from Module to Compilation
compute debug_format in global Compilation config resolution
wire up C compilation to the concept of owner modules
make whole cache mode call link.File.createEmpty() instead of
link.File.open()
These options are only supposed to be provided to the initialization
functions, resolved, and then computed values stored in the appropriate
place (base struct or the object-format-specific structs).
Many more to go...
Much of the logic from Compilation.create() is extracted into
Compilation.Config.resolve() which accepts many optional settings and
produces concrete settings. This separate step is needed by API users of
Compilation so that they can pass the resolved global settings to the
Module creation function, which itself needs to resolve per-Module
settings.
Since the target and other things are no longer global settings, I did
not want them stored in link.File (in the `options` field). That options
field was already a kludge; those options should be resolved into
concrete settings. This commit also starts to work on that, deleting
link.Options, moving the fields into Compilation and
ObjectFormat-specific structs instead. Some fields were ephemeral and
should not have been stored at all, such as symbol_size_hint.
The link.File object of Compilation is now a `?*link.File` and `null`
when -fno-emit-bin is passed. It is now arena-allocated along with
Compilation itself, avoiding some messy cleanup code that was there
before.
On the command line, it is now possible to configure the standard
library itself by using `--mod std` just like any other module. This
meant that the CLI needed to create the standard library module rather
than having Compilation create it.
There are a lot of changes in this commit and it's still not done. I
didn't realize how quickly this changeset was going to balloon out of
control, and there are still many lines that need to be changed before
it even compiles successfully.
* introduce std.Build.Cache.HashHelper.oneShot
* add error_tracing to std.Build.Module
* extract build.zig file generation into src/Builtin.zig
* each CSourceFile and RcSourceFile now has a Module owner, which
determines some of the C compiler flags.
The motivating problem here was a memory leak in the hash maps of
Module.Namespace.
The commit deletes more of the legacy incremental compilation
implementation. It had things like use of orderedRemove and trying to do
too much OOP-style creation and deletion of objects.
Instead, this commit iterates over all the namespaces on Module deinit
and calls deinit on the hash map fields. This logic is much simpler to
reason about.
Similarly, change global inline assembly to an array hash map since
iterating over the values is a primary use of it, and clean up the
remaining values on Module deinit, solving another memory leak.
After this there are no more memory leaks remaining when using the
x86 backend in a libc-less compiler.
The main motivating change here is to prevent the creation of a fake
Decl object by the frontend in order to `@export()` a value.
Instead, `link.updateDeclExports` is renamed to `link.updateExports` and
accepts a tagged union which can be either a Decl.Index or a
InternPool.Index.
The main goal of this commit is to remove the `runtime_value` field from
`InternPool.Key` (and its associated representation), but there are a
few dominos. Specifically, this mostly eliminates the "maybe runtime"
concept from value resolution in Sema: so some resolution functions like
`resolveMaybeUndefValAllowVariablesMaybeRuntime` are gone. This required
a small change to struct/union/array initializers, to no longer
use `runtime_value` if a field was a `variable` - I'm not convinced this
case was even reachable, as `variable` should only ever exist as the
trivial value of a global runtime `var` decl.
Now, the only case in which a `Sema.resolveMaybeUndefVal`-esque function
can return the `variable` key is `resolveMaybeUndefValAllowVariables`,
which is directly called from `Sema.resolveInstValueAllowVariables`
(previously `Sema.resolveInstValue`), which is only used for resolving
the value of a Decl from `Module.semaDecl`.
While changing these functions, I also slightly reordered and
restructured some of them, and updated their doc comments.
Commit 5393e56500d499753dbc39704c0161b47d1e4d5c has a flaw pointed out
by @mlugg: the `ty` field of pointer values changes when comptime values
are pointer-casted. This commit introduces a new encoding which
additionally stores the "original pointer type" which is used to store
the alignment of the anonymous decl, and potentially other information
in the future such as section and pointer address space. However, this
new encoding is only used when the original pointer type differs from
the casted pointer type in a meaningful way.
I was able to make the LLVM backend and the C backend lower anonymous
decls with the appropriate alignment, however I will need some help
figuring out how to do this for the backends that lower anonymous decls
via src/codegen.zig and the wasm backend.
Instead of creating Module.Decl objects, directly create InternPool
pointer values using the anon_decl Addr encoding.
The LLVM backend needed code to notice the alignment of the pointer and
lower accordingly. The other backends likely need a similar change.
The previous code incorrectly added `sub_path` twice.
Also for the compilation unit, it was passing empty string to realpath,
resulting in the error handling codepath being used.
closes#17482
