* Logic to check whether a bin file is not emitted is more complicated
in between `Compilation.create` and `Compilation.update`. Fixed the
logic that decides whether to build compiler-rt and other support
artifacts.
* Basically, one cannot inspect the value of `comp.bin_file.emit` until
after update() is called - fixed another instance of this happening
in the CLI.
* In the CLI, `runOrTest` is updated to properly use the result value
of `comp.bin_file.options.emit` rather than guessing whether the
output binary is.
* Don't assume that the emit output has no directory components in
sub_path. In other words, don't assume that the emit directory is the
final directory; there may be sub-directories.
The two CacheMode values are `whole` and `incremental`.
`incremental` is what we had before; `whole` is new.
Whole cache mode uses everything as inputs to the cache hash;
and when a hit occurs it skips everything including linking.
This is ideal for when source files change rarely and for backends that
do not have good incremental compilation support, for example
compiler-rt or libc compiled with LLVM with optimizations on.
This is the main motivation for the additional mode, so that we can have
LLVM-optimized compiler-rt/libc builds, without waiting for the LLVM
backend every single time Zig is invoked.
Incremental cache mode hashes only the input file path and a few target
options, intentionally relying on collisions to locate already-existing
build artifacts which can then be incrementally updated.
The bespoke logic for caching stage1 backend build artifacts
is removed since we now have a global caching mechanism for
when we want to cache the entire compilation, *including* linking.
Previously we had to get "creative" with libs.txt and a special
byte in the hash id to communicate flags, so that when the cached
artifacts were re-linked, we had this information from stage1
even though we didn't actually run it. Now that `CacheMode.whole`
includes linking, this extra information does not need to be
preserved for cache hits. So although this changeset introduces
complexity, it also removes complexity.
The main trickiness here comes from the inherent differences between the
two modes: `incremental` wants a directory immediately to operate on,
while `whole` doesn't know the output directory until the compilation is
complete. This commit deals with this problem mostly inside `update()`,
where, on a cache miss, it replaces `zig_cache_artifact_directory` with a
temporary directory, and then renames it into place once the compilation is
complete.
Items remaining before this branch can be merged:
* [ ] make sure these things make it into the cache manifest:
- @import files
- @embedFile files
- we already add dep files from c but make sure the main .c files make
it in there too, not just the included files
* [ ] double check that the emit paths of other things besides the binary
are working correctly.
* [ ] test `-fno-emit-bin` + `-fstage1`
* [ ] test `-femit-bin=foo` + `-fstage1`
* [ ] implib emit directory copies bin_file_emit directory in create() and needs
to be adjusted to be overridden as well.
* [ ] make sure emit-h is handled correctly in the cache hash
* [ ] Cache: detect duplicate files added to the manifest
Some preliminary performance measurements of wall clock time and
peak RSS used:
stage1 behavior (1077 tests), llvm backend, release build:
* cold global cache: 4.6s, 1.1 GiB
* warm global cache: 3.4s, 980 MiB
stage2 master branch behavior (575 tests), llvm backend, release build:
* cold global cache: 0.62s, 191 MiB
* warm global cache: 0.40s, 128 MiB
stage2 this branch behavior (575 tests), llvm backend, release build:
* cold global cache: 0.62s, 179 MiB
* warm global cache: 0.27s, 90 MiB
The status quo for the `build.zig` build system is preserved in
the sense that, if the user does not explicitly override
`dylib.setInstallName(...);` in their build script, the default
of `@rpath/libname.dylib` applies. However, should they want to
override the default behaviour, they can either:
1) unset it with
```dylib.setIntallName(null);```
2) set it to an explicit string with
```dylib.setInstallName("somename.dylib");```
When it comes to the command line however, the default is not to
use `@rpath` for the install name when creating a dylib. The user
will now be required to explicitly specify the `@rpath` as part
of the desired install name should they choose so like so:
1) with `build-lib`
```
zig build-lib -dynamic foo.zig -install_name @rpath/libfoo.dylib
```
2) with `cc`
```
zig cc -shared foo.c -o libfoo.dylib -Wl,"-install_name=@rpath/libfoo.dylib"
```
Notating a symbol to be exported in code will only tell the linker
where to find this symbol, so other object files can find it. However, this does not mean
said symbol will also be exported to the host environment. Currently, we 'fix' this by force
exporting every single symbol that is visible. This creates bigger binaries and means host environments
have access to symbols that they perhaps shouldn't have. Now, users can tell Zig which symbols
are to be exported, meaning all other symbols that are not specified will not be exported.
Another change is we now support `-rdynamic` in the wasm linker as well, meaning all symbols will
be put in the dynamic symbol table. This is the same behavior as with ELF. This means there's a 3rd strategy
users will have to build their wasm binary.
While investigating slow build times with [a large project](https://github.com/hexops/mach/issues/124),
I found that the compiler was reading from disk nearly every C source file in my project
when rebuilding despite no changes having been made. This accounted for several seconds of
time (approx. 20-30% of running `zig build` without any changes to the sources.)
The cause of this was that comparisons of file mtimes would _always_ fail (the mtime of the file on
disk was always newer than that stored in the cache manifest), and so the cache logic would always
fall back to byte-for-byte file content comparisons with what is on disk vs. in the cache-reading every
C source file in my project from disk during each rebuild. Because file contents were the same, a cache
hit occurred, and _despite the mtime being different the cache manifest would not be updated._
One can reproduce this by building a Zig project so the cache is populated, and then changing mtimes
of their C source files to be newer than what is in the cache (without altering file contents.)
The fix is rather simple: we should always write the updated cache manifest regardless of
whether or not a cache hit occurred (a cache hit doesn't indicate if a manifest is dirty) Luckily,
`writeManifest` already contains logic to determine if a manifest is dirty and becomes no-op if no
change to the manifest file is necessary-so we merely need to ensure it is invoked.
Signed-off-by: Stephen Gutekanst <stephen@hexops.com>
All Zig code is eligible to `@import("builtin")` which is mapped to a
generated file, build.zig, based on the target and other settings.
Zig invocations which share the same target settings will generate the
same builtin.zig file and thus the path to builtin.zig is in a shared
cache folder, and different projects can sometimes use the same file.
Before this commit, this led to race conditions where multiple
invocations of `zig` would race to write this file. If one process
wanted to *read* the file while the other process *wrote* the file, the
reading process could observe a truncated or partially written
builtin.zig file.
This commit makes the following improvements:
- limitations:
- avoid clobbering the inode, mtime in the hot path
- avoid creating a partially written file
- builtin.zig needs to be on disk for debug info / stack trace purposes
- don't mark the task as complete until the file is finished being populated
(possibly by an external process)
- strategy:
- create the `@import("builtin")` `Module.File` during the AstGen
work, based on generating the contents in memory rather than
loading from disk.
- write builtin.zig in a separate task that doesn't have
to complete until the end of the AstGen work queue so that it
can be done in parallel with everything else.
- when writing the file, first stat the file path. If it exists, we are done.
- otherwise, write the file to a temp file in the same directory and atomically
rename it into place (clobbering the inode, mtime in the cold path).
- summary:
- all limitations respected
- hot path: one stat() syscall that happens in a worker thread
This required adding a missing function to the standard library:
`std.fs.Dir.statFile`. In this commit, it does open() and then fstat()
which is two syscalls. It should be improved in a future commit to only
make one.
Fixes#9439.
This branch introduced std.Target.TargetAbi when we already had
std.Target.Abi which was, unsurprisingly, already suited for this task.
Also pull out the -mabi= cc flag addition to the common area instead of
duplicating it for assembly and c files.
The target abi can also be set in build.zig via LibExeObjStep.target_abi
The value passed in is checked that it is a valid value in
std.Target.TargetAbi
The target abi is also validated against the target cpu
Previously there was only `--single-threaded`.
This flag now matches other boolean flags, instead of only being able to
opt in to single-threaded builds, you can now force multi-threaded
builds. Currently this only has the possibility to emit an error
message, but it is a better user experience to understand why one cannot
choose to enable threads in some cases.
This is breaking change to the CLI.
Related: #10143
* Add missing Linux headers. Closes#9837
* Update existing headers to latest Linux.
* Consolidate headers that are the same for multiple Zig target CPU
architectures. For example, Linux has only an x86 directory for both
x86_64 and x86 CPU architectures. Now Zig only ships an x86 directory
for Linux headers, and will emit the proper corresponding -isystem
flags.
* tools/update-linux-headers.zig is now available for upgrading to
newer Linux headers, and the update process is now documented on the
wiki.
Since we are already detecting the path to the native SDK,
if available, also fetch SDK's version and route that to the linker.
The linker can then use it to correctly populate LC_BUILD_VERSION
load command.
In 7e23b3245a9bf6e002009e6c18c10a9995671afa I made -O flags to the
linker emit a warning that the argument does nothing. That was not
correct however; LLD does have some logic that does different things
depending on -O0, -O1, and -O2. It defaults to -O1, and it does less
optimizations with -O0 and more with -O2.
With this commit, e.g. `-Wl,-O1` is supported by the `zig cc` frontend,
and by default we pass `-O0` to LLD in debug mode, and `-O3` in release
modes.
I also fixed a bug in the LLD ELF linker line which was incorrectly
passing `-O` flags instead of `--lto-O` flags for LTO.
* Improve the logic for determining whether emitting an import lib is
eligible, and improve the error message when the user provides
contradictory arguments.
* Integrate with the EmitLoc / Emit system that already exists, and use
the `-femit-implib[=path]`/`-fno-emit-implib` convention that already
exists.
* Proper integration with the caching system.
* CLI: fix bug in error reporting for resolving EmitLoc values for
other parameters.
This mechanism for sending arbitrary linker args to LLD has no place in
the Zig frontend, because our goal is for the frontend to understand all
the arguments and not treat linker args like a black box.
For example we have self-hosted linking in addition to LLD, so we want to
have the options make sense to both linking codepaths, not just the LLD one.
Passing -O linker args will now result in a warning that the arg does
nothing.
* add support for compiling Objective-C++ code
Prior to this change, calling `step.addCSourceFiles` with Obj-C++ file extensions
(`.mm`) would result in an error due to Zig not being aware of that extension.
Clang supports an `-ObjC++` compilation mode flag, but it was only possible to use
if you violated standards and renamed your `.mm` Obj-C++ files to `.m` (Obj-C) to
workaround Zig being unaware of the extension.
This change makes Zig aware of `.mm` files so they can be compiled, enabling compilation
of projects such as [Google's Dawn WebGPU](https://dawn.googlesource.com/dawn/) using
a `build.zig` file only.
Helps hexops/mach#21
Signed-off-by: Stephen Gutekanst <stephen@hexops.com>
* test/standalone: add ObjC++ compilation/linking test
Based on the existing objc example, just tweaked for ObjC++.
Signed-off-by: Stephen Gutekanst <stephen@hexops.com>
After this change, the default for dynamic libraries (`-l` or
`--library`) is to only link them if they end up being actually used.
With the Zig CLI, the new options `-needed-l` or `--needed-library` can
be used to force link against a dynamic library.
With `zig cc`, this behavior can be overridden with `-Wl,--no-as-needed`
(and restored with `-Wl,--as-needed`).
Closes#10164
This is a breaking change. Before, usage looked like this:
```zig
const held = mutex.acquire();
defer held.release();
```
Now it looks like this:
```zig
mutex.lock();
defer mutex.unlock();
```
The `Held` type was an idea to make mutexes slightly safer by making it
more difficult to forget to release an aquired lock. However, this
ultimately caused more problems than it solved, when any data structures
needed to store a held mutex. Simplify everything by reducing the API
down to the primitives: lock() and unlock().
Closes#8051Closes#8246Closes#10105
--import-memory import memory from the environment
--initial-memory=[bytes] initial size of the linear memory
--max-memory=[bytes] maximum size of the linear memory
--global-base=[addr] where to start to place global data
See #8633
Because ArrayList.initCapacity uses 'precise' capacity allocation, this should save memory on average, and definitely will save memory in cases where ArrayList is used where a regular allocated slice could have also be used.
Add an option to allow the '-z notext' option to be passed to the linker
via. the compiler frontend, which is a flag that tells the linker that
relocations in read-only sections are permitted. Certain targets such as
Solana BPF rely on this flag.
Expose all linker options i.e. '-z nodelete', '-z now', '-z relro' in
the compiler frontend. Usage documentation has been updated accordingly.
Expose the '-z notext' flag in the standard library build runner.
There is a table of `misc_failures` which previously did not allow
multiple errors for the same enum tag. Now it allows this by freeing the
previous compile error and replacing it with the new one. Typically this
will happen because multiple sub-Compilation objects fail with the same
problem, such as not being able to build glibc because of not having
LLVM extensions enabled.
The main problem that motivated these changes is that global constants
which are referenced by pointer would not be emitted into the binary.
This happened because `semaDecl` did not add `codegen_decl` tasks for
global constants, instead relying on the constant values being copied as
necessary. However when the global constants are referenced by pointer,
they need to be sent to the linker to be emitted.
After making global const arrays, structs, and unions get emitted, this
uncovered a latent issue: the anonymous decls that they referenced would
get garbage collected (via `deleteUnusedDecl`) even though they would
later be referenced by the global const.
In order to solve this problem, I introduced `anon_work_queue` which is
the same as `work_queue` except a lower priority. The `codegen_decl`
task for anon decls goes into the `anon_work_queue` ensuring that the
owner decl gets a chance to mark its anon decls as alive before they are
possibly deleted.
This caused a few regressions, which I made the judgement call to add
workarounds for. Two steps forward, one step back, is still progress.
The regressions were:
* Two behavior tests having to do with unions. These tests were
intentionally exercising the LLVM constant value lowering, however,
due to the bug with garbage collection that was fixed in this commit,
the LLVM code was not getting exercised, and union types/values were
not implemented correctly, due to me forgetting that LLVM does not
allow bitcasting aggregate values.
- This is worked around by allowing those 2 test cases to regress,
moving them to the "passing for stage1 only" section.
* The test-stage2 test cases (in test/cases/*) for non-LLVM backends
previously did not have any calls to lower struct values, but now
they do. The code that was there was just `@panic("TODO")`. I
replaced that code with a stub that generates the wrong value. This
is an intentional miscompilation that will obviously need to get
fixed before any struct behavior tests pass. None of the current
tests we have exercise loading any values from these global const
structs, so there is not a problem until we try to improve these
backends.
Each element of the output JSON has the VM address of the generated
binary nondecreasing (some elements might occupy the same VM address
for example the atom and the relocation might coincide in the address
space).
The generated JSON can be inspected manually or via a preview tool
`zig-snapshots` that I am currently working on and will allow the user
to inspect interactively the state of the linker together with the
positioning of sections, symbols, atoms and relocations within each
snapshot state, and in the future, between snapshots too. This should
allow for quicker debugging of the linker which is nontrivial when
run in the incremental mode.
Note that the state will only be dumped if the compiler is built with
`-Dlink-snapshot` flag on, and then the compiler is passed `--debug-link-snapshot`
flag upon compiling a source/project.
