Consider a (legal according to the `@bitCast` rules) conversion from u16
to [2]u8: since the former is a scalar and the latter is a pointer
(arrays are represented at pointers in the codegen phase) we have to
allocate a temporary slot on the stack and then bitcast the resulting
pointer to the desired destination type.
Beware that this means the lifetime of the resulting value is the same
of the function it's contained in and for all intents and purposes
should be regarded as a local (eg. it should not escape).
Closes#4395Closes#5121
The codegen would sometimes change the LLVM type for some constants to
an unnamed structure in order to accomodate extra padding. This is fine
as long as the alignment of each field is still respected and it was not
the case for structure types, leading to ill-formed constants being
generated.
Optional types suffer from this to a lower extent as their layout is
quite lucky, the only missing piece was the tail padding.
Closes#4530Closes#4594Closes#4295
Closes my will to live
This is a follow-up to d27ef1aaef8ac12801ba4e6c2ed748c7151096a7
and avoids compiler warnings:
warning: argument unused during compilation: '-nostdinc++'
Before, this would cause a link failure when mixing Zig and C code for
RISC-V targets.
Now, the ABIs match and Zig and C code can be mixed successfully.
I will file a follow-up issue for the ability to deal more explicitly
with ABIs.
closes#4863
Make fallthrough an error when compiler supports it. This requires a new
macro that is defined with such compilers to be used as a statement, at
all fallthrough sites:
switch (...) {
case 0:
...
ZIG_FALLTHROUGH;
case 1:
...
break;
default:
...
break;
}
If we ever move to C++17 as minimal requirement, then the macro can be
replaced with `[[fallthrough]];` at statement sites.
This caused link errors in c++ code because it was not correct to pass
these flags to child codegens. And that was the only reason to detect
these flags. Otherwise we can safely rely on non-explicitly-detected
flag forwarding.
* `-isystem` instead of `-I` for system include directories
fixes a problem with native system directories interfering with zig's
bundled libc.
* separate Stage2Target.is_native into Stage2Target.is_native_os and
Stage2Target.is_native_cpu.
* Don't add an extra slot for the sentinel. Most of the code keeps using
the constant value from the type descriptor, let's harmonize all the
code dealing with sentinels.
* Properly write out sentinel values when reinterpreting pointers at
comptime.
* Allow the reading of the 0th element in a `[0:S]T` type.
When building an object file from only one source file, instead of
having a two-stage cache system, we special case it and use the cache
directory that the .o file is output to as the final cache directory for
all the build artifacts.
When there are more than 1 source file, the linker has to merge objects
into one, and so the two stage approach makes sens. But in the case of
only one source file, this prevents needlessly copying the object file.
This commit fixes an issue with the previous one, where zig with cache
enabled would print a directory that actually did not have any build
artifacts in it.
