Add std.Target.Cpu.Model.generic which is even more empty than baseline.
CPU model and feature detection uses this rather than baseline.
Rename cpu_detected to cpu_detection_unimplemented and flip the logic.
It can be relied on by stage2.zig to decide whether the LLVM workaround
is needed without also checking the CrossTarget.
Move the CPU detection to after the OS detection, and use the detected
OS for the CPU detection. This is relevant because operating systems
sometimes emulate certain CPU features, so knowing the OS and version is
relevant for determining CPU features.
Prepare for #4592 by passing the CPU arch to the detection code, instead
of having it rely on Target.current.
The CPU model & feature detection logic is modified. Before:
* Detect actual features
* Use as hint when detecting CPU model
* Populate dependencies of CPU model features
* Merge that into the actual features set
After:
* Detect actual features
* Use as hint when detecting CPU model
* Add known CPU model features to actual features
* Detect actual features again, overriding known CPU model features
* Populate dependencies
* `std.Target.getStandardDynamicLinkerPath` =>
`std.Target.standardDynamicLinkerPath`
* it now takes a pointer to fixed size array rather than an allocator
* `std.zig.system.NativeTargetInfo.detect` now supports reading
PT_INTERP from /usr/bin/env
Zig now supports a more fine-grained sense of what is native and what is
not. Some examples:
This is now allowed:
-target native
Different OS but native CPU, default Windows C ABI:
-target native-windows
This could be useful for example when running in Wine.
Different CPU but native OS, native C ABI.
-target x86_64-native -mcpu=skylake
Different C ABI but otherwise native target:
-target native-native-musl
-target native-native-gnu
Lots of breaking changes to related std lib APIs.
Calls to getOs() will need to be changed to getOsTag().
Calls to getArch() will need to be changed to getCpuArch().
Usage of Target.Cross and Target.Native need to be updated to use
CrossTarget API.
`std.build.Builder.standardTargetOptions` is changed to accept its
parameters as a struct with default values. It now has the ability to
specify a whitelist of targets allowed, as well as the default target.
Rather than two different ways of collecting the target, it's now always
a string that is validated, and prints helpful diagnostics for invalid
targets. This feature should now be actually useful, and contributions
welcome to further improve the user experience.
`std.build.LibExeObjStep.setTheTarget` is removed.
`std.build.LibExeObjStep.setTarget` is updated to take a CrossTarget
parameter.
`std.build.LibExeObjStep.setTargetGLibC` is removed. glibc versions are
handled in the CrossTarget API and can be specified with the `-target`
triple.
`std.builtin.Version` gains a `format` method.
* re-introduce `std.build.Target` which is distinct from `std.Target`.
`std.build.Target` wraps `std.Target` so that it can be annotated as
"the native target" or an explicitly specified target.
* `std.Target.Os` is moved to `std.Target.Os.Tag`. The former is now a
struct which has the tag as well as version range information.
* `std.elf` gains some more ELF header constants.
* `std.Target.parse` gains the ability to parse operating system
version ranges as well as glibc version.
* Added `std.Target.isGnuLibC()`.
* self-hosted dynamic linker detection and glibc version detection.
This also adds the improved logic using `/usr/bin/env` rather than
invoking the system C compiler to find the dynamic linker when zig
is statically linked. Related: #2084
Note: this `/usr/bin/env` code is work-in-progress.
* `-target-glibc` CLI option is removed in favor of the new `-target`
syntax. Example: `-target x86_64-linux-gnu.2.27`
closes#1907
See e381a42de9c0f0c5439a926b0ac99026a0373f49 for more details.
This is set up so that if we wish to make "baseline" depend on the
OS in the future, it is possible to do that.
I think this is working correctly. Without also removing sse2 from the
feature set, sse gets added back into the set because sse2 is part of
the x86_64 baseline (which is the cpu provided) and sse2 depends on sse.
in favor of CPU features. Also rearrange the `std.Target`
data structure.
* note: `@import("builtin")` was already deprecated in favor of
`@import("std").builtin`.
* `std.builtin.arch` is now deprecated in favor of
`std.builtin.cpu.arch`.
* `std.Target.CpuFeatures.Cpu` is now `std.Target.Cpu.Model`.
* `std.Target.CpuFeatures` is now `std.Target.Cpu`.
* `std.Target` no longer has an `arch` field. Instead it has a
`cpu` field, which has `arch`, `model`, and `features`.
* `std.Target` no longer has a `cpu_features` field.
* `std.Target.Arch` is moved to `std.Target.Cpu.Arch` and
it is an enum instead of a tagged union.
* `std.Target.parseOs` is moved to `std.Target.Os.parse`.
* `std.Target.parseAbi` is moved to `std.Target.Abi.parse`.
* `std.Target.parseArchSub` is only for arch now and moved
to `std.Target.Cpu.Arch.parse`.
* `std.Target.parse` is improved to accept CPU name and features.
* `std.Target.Arch.getBaselineCpuFeatures` is moved to
`std.Target.Cpu.baseline`.
* `std.Target.allCpus` is renamed to `std.Target.allCpuModels`.
* `std.Target.defaultAbi` is moved to `std.Target.Abi.default`.
* Significant cleanup of aarch64 and arm CPU features, resulting in
the needed bit count for cpu feature set going from 174 to 138.
* Add `std.Target.Cpu.Feature.Set.addFeatureSet` for merging
feature sets together.
`-target-feature` and `-target-cpu` are removed in favor of
`-mcpu`, to conform to established conventions, and it gains
additional power to support cpu features. The syntax is:
-mcpu=name+on1+on2-off1-off2
closes#4261
* libc_installation.cpp is deleted.
src-self-hosted/libc_installation.zig is now used for both stage1 and
stage2 compilers.
* (breaking) move `std.fs.File.access` to `std.fs.Dir.access`. The API
now encourages use with an open directory handle.
* Add `std.os.faccessat` and related functions.
* Deprecate the "C" suffix naming convention for null-terminated
parameters. "C" should be used when it is related to libc. However
null-terminated parameters often have to do with the native system
ABI rather than libc. "Z" suffix is the new convention. For example,
`std.os.openC` is deprecated in favor of `std.os.openZ`.
* Add `std.mem.dupeZ` for using an allocator to copy memory and add a
null terminator.
* Remove dead struct field `std.ChildProcess.llnode`.
* Introduce `std.event.Batch`. This API allows expressing concurrency
without forcing code to be async. It requires no Allocator and does
not introduce any failure conditions. However it is not thread-safe.
* There is now an ongoing experiment to transition away from
`std.event.Group` in favor of `std.event.Batch`.
* `std.os.execvpeC` calls `getenvZ` rather than `getenv`. This is
slightly more efficient on most systems, and works around a
limitation of `getenv` lack of integration with libc.
* (breaking) `std.os.AccessError` gains `FileBusy`, `SymLinkLoop`, and
`ReadOnlyFileSystem`. Previously these error codes were all reported
as `PermissionDenied`.
* Add `std.Target.isDragonFlyBSD`.
* stage2: access to the windows_sdk functions is done with a manually
maintained .zig binding file instead of `@cImport`.
* Update src-self-hosted/libc_installation.zig with all the
improvements that stage1 has seen to src/libc_installation.cpp until
now. In addition, it now takes advantage of Batch so that evented I/O
mode takes advantage of concurrency, but it still works in blocking
I/O mode, which is how it is used in stage1.
Clean up the code a bit and introduce a few checks meant to avoid
overshooting the end of the frame chain.
The code is now stable enough not to cause panics during the call frame
walking.
Previously it was a tagged union which was one of:
* baseline
* a specific CPU
* a set of features
Now, it's possible to have a CPU but also modify the CPU's feature set
on top of that. This is closer to what LLVM does.
This is more correct because Zig's notion of CPUs (and LLVM's) is not
exact CPU models. For example "skylake" is not one very specific model;
there are several different pieces of hardware that match "skylake" that
have different feature sets enabled.
comment from this commit reproduced here:
I have observed the CPU name reported by LLVM being incorrect. On
the SourceHut build services, LLVM 9.0 reports the CPU as "athlon-xp",
which is a 32-bit CPU, even though the system is 64-bit and the reported
CPU features include, among other things, +64bit.
So the strategy taken here is that we observe both reported CPU, and the
reported CPU features. The features are trusted more; but if the features
match exactly the features of the reported CPU, then we trust the reported CPU.
