This takes the place of `zig builtin`. This is an improvement over the
command because now the generated source will correctly show LinkMode
and OutputMode, whereas before it was always stuck as Static and Obj,
respectively.
Deleted 16,000+ lines of c++ code, including:
* an implementation of blake hashing
* the cache hash system
* compiler.cpp
* all the linking code, and everything having to do with building
glibc, musl, and mingw-w64
* much of the stage1 compiler internals got slimmed down since it
now assumes it is always outputting an object file.
More stuff:
* stage1 is now built with a different strategy: we have a tiny
zig0.cpp which is a slimmed down version of what stage1 main.cpp used
to be. Its only purpose is to build stage2 zig code into an object
file, which is then linked by the host build system (cmake) into
stage1. zig0.cpp uses the same C API that stage2 now has access to,
so that stage2 zig code can call into stage1 c++ code.
- stage1.h is
- stage2.h is
- stage1.zig is the main entry point for the Zig/C++
hybrid compiler. It has the functions exported from Zig, called
in C++, and bindings for the functions exported from C++, called
from Zig.
* removed the memory profiling instrumentation from stage1.
Abandon ship!
* Re-added the sections to the README about how to build stage2 and
stage3.
* stage2 now knows as a comptime boolean whether it is being compiled
as part of stage1 or as stage2.
- TODO use this flag to call into stage1 for compiling zig code.
* introduce -fdll-export-fns and -fno-dll-export-fns and clarify
its relationship to link_mode (static/dynamic)
* implement depending on LLVM to detect native target cpu features when
LLVM extensions are enabled and zig lacks CPU feature detection for
that target architecture.
* C importing is broken, will need some stage2 support to function
again.
Here's the doc comment from the commit:
For one example of why this is handy, consider the case of building musl libc.
We keep a lock open for each of the object files in the form of a file descriptor
until they are finally put into an archive file. This is to allow a zig-cache
garbage collector to run concurrently to zig processes, and to allow multiple
zig processes to run concurrently with each other, without clobbering each other.
This code is disabled until #6361 is implemented (getrlimit/setrlimit
are not yet added to the standard library).
* caching system: use 16 bytes siphash final(), there was a bug in the
std lib that wasn't catching undefined values for 18 bytes. fixed in
master branch.
* fix caching system unit test logic to not cause error.TextBusy on windows
* port the logic from stage1 for building glibc shared objects
* add is_native_os to the base cache hash
* fix incorrectly freeing crt_files key (which is always a reference to
global static constant data)
* fix 2 use-after-free in loading glibc metadata
* fix memory leak in buildCRTFile (errdefer instead of defer on arena)
This reverts commit 40cb712d13aff4bfe83256858ad6b18d82e70211.
Thanks to Ava & Luna of Lavatech, we don't need to resort to this, they
have graciously given zig a SourceHut instance to use that gives us 8GB
RAM.
Before merging, do this for every item in the file:
* solve the issue, or
* convert the task to a github issue and update the comment
to link to the issue (and remove "TODO" text from the comment).
Then delete the file.
Related: #363
Drew won't give us enough RAM for stage1 to build stage2. We'll still
have freebsd builds available on releases but we're going to lose
freebsd CI testing for master branch builds until we fully switch over
to stage2 (and have lower memory usage).
Let me know if anyone wants to run a SourceHut instance and give zig
access to run on slightly more powerful machines. We need about 8 GiB
RAM to run the CI test suite for now.
After we're fully self hosted I expect to re-enable this.
The API is pretty specific to the implementationt details of the
self-hosted compiler. I don't want to have to independently support
and maintain this as part of the standard library, and be obligated
to not make breaking changes to it with changes to the implementation of
stage2.
This is not strictly necessary but it increases the likelihood of cache
hits because foo.c and bar.c now will have different cache directories
and can be updated independently without clobbering each other's cache
data.
