* Remove the ability for GenZir parent Scope to be null. Now there is a
Top Scope at the top.
* Introduce Scope.Namespace to contain a table of decl names in order
to emit a compile error for name conflicts.
* Fix use of invalid memory when reporting compile errors by
duplicating decl names into a temporary heap allocated buffer.
* Fix memory leak in while and for loops, not cleaning up their
labeled_breaks and store_to_block_ptr_list arrays.
* Fix stage2 test cases because now the source location of redundant
comptime keyword compile errors is improved.
* Implement compile error for local variable shadowing declaration.
commitDeclDebugInfo: stop trying to write length 0 debug info - avoids
hitting an error where zig tries to move the debug info section
unnecessarily, gets confused, and reports `error.InputOutput`.
The 2 pieces of code that looked at the source and tried to compute
source line offsets is now improved to match link/Elf - use the
line/column data stored in the Decl object to skip the costly scanning
of source bytes. No need to load Zig source code, AST, or tokens, when
we have the ZIR!
* `-lc++` now implies `-lc`.
* `-lunwind` is now pulled out into a separate `link_libunwind` flag in
the frontend driver code. This allows a project to request zig to
provide libunwind even if the default situation that causes it to be
implicitly added, is not active.
* build.zig: ask for -lunwind when building the self-hosted compiler on
Linux. Otherwise we get linker errors with unresolved symbols to
libunwind.
* for cross testing stage2 tests, we use `darling shell` command
since the path to the tested binary is relative to cwd
* for the `zig test` command, we simply use `darling` since the
path to the binary is absolute
With this change, it is now possible to use zig-bootstrap as the
initial C/C++ compiler, and then build stage2 with the resultant
stage1 without the need to link any system libs!
In `Module.semaDecl`, the source location being used was double-relative
to the `Decl`, causing a crash when trying to compute byte offset for
the compile error.
Change the source location to node_offset = 0 since the scope being used
makes the source location relative to the Decl, which already has the
source node index populated.
5ac91794cce8bd53916a378815be01e4365d53d9 made Zig link against the
system libc when targeting the native C ABI. However this made it stop
putting libunwind.a on the linker line when it needed to sometimes,
causing undefined symbols when linking against C++ code.
5ac91794cce8bd53916a378815be01e4365d53d9 made Zig link against the
system libc when targeting the native C ABI. However this made it stop
putting libunwind.a on the linker line when it needed to sometimes,
causing undefined symbols when linking against C++ code.
Before this change, the attempt to save the most recent successful ZIR
code only worked in some cases; this reworks the code to be more robust,
thereby fixing a crash when running the stage2 "enums" CBE test cases.
Without this, and with `-Denable-qemu -Denable-foreign-glibc`,
i386-linux-gnu tests failed because the `-L` path passed to qemu
was "i386-linux-gnu" (nonexistent) rather than "i686-linux-gnu".
Fixes std lib and behavior tests when using these options to enable
test coverage with cross compiled glibcs.
Conflicts:
* lib/std/os/linux.zig
* lib/std/os/windows/bits.zig
* src/Module.zig
* src/Sema.zig
* test/stage2/test.zig
Mainly I wanted Jakub's new macOS code for respecting stack size, since
we now depend on it for debug builds able to pass one of the test cases
for recursive comptime function calls with `@setEvalBranchQuota`.
The conflicts were all trivial.
The CPU detection code is nearly at feature parity, we do support
detecting the native CPU on Sparc systems and macos, our ARM/AArch64
model list is quite comprehensive and so is our PPC one.
The only missing pieces are:
- ARM32 detection on Darwin hosts (I don't think anybody is planning on
running the compiler on a old-ass iPhone)
- s390x detection on Linux hosts, this can be easily added at a later
stage.
The current spanZ() function will not scan for a 0 terminator if the
type is not 0 terminated. This encourages using 0 terminated array
types to bind C arrays which hold 0 terminated strings. However, this is
a big footgun as nothing in the C type system guarantees there to be a
0 terminator at the end of the array and if there is none this becomes
Illegal Behavior in Zig.
To solve this, deprecate spanZ() and lenZ(), adding a new sliceTo()
function that always scans for the given terminator even if the type is
not sentinel terminated.
* File stores `root_decl: Decl` instead of `namespace: *Namespace`.
This maps more cleanly to the actual ownership, since the `File` does
own the root decl, but it does not directly own the `Namespace`.
* `semaFile` completes the creation of the `Decl` even when semantic
analysis fails. The `analysis` field of the `Decl` will contain the
results of semantic analysis. This prevents cleaning up of memory
still referenced by other Decl objects.
* `semaDecl` sets `Struct.zir_index` of the root struct decl, which
fixes use of undefined value in case the first update contained a ZIR
compile error.
* Compilation: iteration over the deletion_set only tries to delete the
first one, relying on Decl destroy to remove itself from the deletion
set.
* link: `freeDecl` now has to handle the possibility of freeing a Decl
that was never called with `allocateDeclIndexes`.
* `deleteDecl` recursively iterates over a Decl's Namespace sub-Decl
objects and calls `deleteDecl` on them.
- Prevents Decl objects from being destroyed when they are still in
`deletion_set`.
* Sema: fix cleanup of anonymous Decl objects when an error occurs
during semantic analysis.
* tests: update test cases for fully qualified names
This avoids causing false positive compile errors when, for example, a
file had ZIR errors, and then code tried to look up a public decl from
the failed file.
