The current MJIT relies on SIGCHLD and fork(2) to be performant, and
it's something mswin can't offer. You could run Linux MJIT on WSL
instead.
[Misc #18968]
As commented in include/ruby/internal/abi.h, since teeny versions of
Ruby should guarantee ABI compatibility, `RUBY_ABI_VERSION` has no role
in released versions of Ruby.
On other platforms, RUBY_SO_NAME is defined from RUBY_API_VERSION.
ruby_version contains the ABI version, which is not needed.
RUBY_API_VERSION is defined as MAJOR.MINOR.
* Rename --jit to --mjit
[Feature #18349]
* Fix a few more --jit references
* Fix MJIT Actions
* More s/jit/mjit/ and re-introduce --disable-jit
* Update NEWS.md
* Fix test_bug_reporter_add
With /Z7, no .pdb file is generated, so trying to link it during build
fails on my machine even though it's okay on CI.
By the way, in my local testing, no .pdb is generated in cwd at runtime
even without the /Fd option. I guess we can pass it just in case.
With MSVC, MJIT uses the /Fd option on an installed PDB file when
compiling. Combined with the /Zi option, this causes the PDB file to be
modified every time MJIT compiles. Concurrent modifications to the same
PDB file is known to cause problems. MSVC even has an option, /FS to
deal with it. When running MJIT tests in parallel, sometimes this leads
to corrupting the PDB file, breaking subsequent compilations. On CI,
we get messages like these:
rb_mjit_header-3.1.0.pdb is not the pdb file that was used when this precompiled header was created, recreate the precompiled header.
To avoid this race, use the /Z7 option when building precompiled header,
which asks the compiler to put debug info into the .obj file,
eliminating the need for pointing the compiler to the PDB file for the
precompiled header.
The /Fd option is changed to use a unique path based on the name of the
dll output. Because of the /debug linker flag, we generate a PDB file
at runtime even though we use /Z7.
There are a couple things missing from this change:
- Because MJIT uses the interpreter's CFLAGS build option and that
contains /Zi, putting /Z7 at the end leads to a build warning
- With /Z7 no PDB file is built anymore, so the code for installing
the PDB file can be removed
There might also be other problems with this change I haven't noticed
while developing this change using Github Actions. I don't have a
Windows dev environment with Visual Studio so I can't finish this
change easily. Please feel free to complete this change if it makes
sense.
Note:
- On master, you can see the PDB file changing with llvm-pdbutil or a
simple checksum. There is an age field in the file that is bumped
- I'm not sure if users can specify compile flags on MSVC. If they
couldn't, maybe it's easier to change MJIT's compile options to
use /Z7 when building the precompile header.
- MJIT could pass different options at runtime to generate fewer
files. Right now it inherits the /DEBUG linker flag which causes
a PDB file to be generated at runtime even though /Z7 is used.
Relevant MSVC docs:
- [/Zi,/Z7](https://docs.microsoft.com/en-us/cpp/build/reference/z7-zi-zi-debug-information-format?view=msvc-160)
- [/DEBUG](https://docs.microsoft.com/en-us/cpp/build/reference/debug-generate-debug-info?view=msvc-160)
- [/FS](https://docs.microsoft.com/en-us/cpp/build/reference/fs-force-synchronous-pdb-writes?view=msvc-160)
Now we require C99, these features available of course.
* prototypes
* stdarg prototypes
* token pasting
* stringization
* string literal concatenation
Recent (since 2012 maybe?) MSVC ships a header named xkeycheck.h, which
(kindly!) aborts compilation on redefinition of C++ keywords. Let's not
define this in case of C++.
C++ keyword `nullptr` represents a null pointer (note also that NULL is
an integer in C++ due to its design flaw). Its type is `std::nullptr_t`,
defined in <cstddef> standard header. Why not support it when the
backend implementation can take a null pointer as an argument.