Our glibc stub assembly file looked something like this:
```
.globl _Exit_2_2_5
.type _Exit_2_2_5, %function;
.symver _Exit_2_2_5, _Exit@@GLIBC_2.2.5
.hidden _Exit_2_2_5
_Exit_2_2_5:
```
With clang 12, the shared objects this produced stopped having any
exported symbols. When I removed the `.hidden` directive, it resolved
the issue, however, there are now unwanted exports:
```
$ readelf -W --dyn-syms libc.so.6 | grep sys_errlist
139: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 _sys_errlist_GLIBC_2_3
147: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 _sys_errlist_GLIBC_2_4
395: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 _sys_errlist_GLIBC_2_2_5
487: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 sys_errlist_GLIBC_2_2_5
1266: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 _sys_errlist@@GLIBC_2.12
1267: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 _sys_errlist@GLIBC_2.2.5
1268: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 _sys_errlist@GLIBC_2.3
1269: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 _sys_errlist@GLIBC_2.4
2137: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 sys_errlist@@GLIBC_2.12
2138: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 sys_errlist@GLIBC_2.2.5
2139: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 sys_errlist@GLIBC_2.3
2140: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 sys_errlist@GLIBC_2.4
2156: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 sys_errlist_GLIBC_2_3
2161: 000000000001ee08 0 FUNC GLOBAL DEFAULT 7 sys_errlist_GLIBC_2_4
```
Every line here without an `@` symbol is an unwanted export. Before, the
unwanted ones had LOCAL HIDDEN linkage.
As a mitigation, I did two things:
* Added `_GLIBC_` to the unwanted exports so that they would not
conflict with anything.
* Made the default export (the `@@` one) the bare symbol name. This
appears to reduce the unwanted exports to only symbols that have more
than one symbol (which is still quite many).
This will unblock progress on this branch, however, there is now a new
issue to solve, that the provided glibc stub .so files have too many
symbols exported. We will have to find a way to avoid this.
This completes the process. All target CPU features are now
auto-generated by the tools/update_cpu_features.zig script, which
contains all the overrides.
Invoking this tool against LLVM 12rc2 now produces an empty git diff.
With this change, added & modified cpus & features participate in the
same pruning system, and sorting takes into account the zig name, not
the pre-modified llvm name.
The modified target files in this commit are due to the improved
sorting and pruning.
The script now fully supports extra cpus & features.
The tools/update_cpu_features script is coming along, and generates
correct information for all these targets. The remaining targets are:
* arm
* aarch64
* amdgpu
* riscv
I will commit them once the issues with the updater tool are resolved.
This introduces {'} to indicate escape for a single-quoted string,
and {} to indicate escape for a double quoted string.
Without this, there would be unnecessary \' inside double quoted
strings, and unnecessary \" inside single quoted strings.
Motivated by the llvm12 branch, in the new tool I am writing for
updating target CPU features.
also avoid unnecessary escaping of single quotes inside double quoted
strings (depends on a master branch commit that will be merged into this
branch in a future commit)
This replaces the previous target cpu features tool, taking advantage of
llvm-tblgen --dump-json instead of trying to use python to parse the .td
files.
This is an initial version that has the basics working, including a
simple feature override system, as well as multi-threaded processing.
Follow-up commits will do clean ups to make the diff of the newly generated
source files against previous versions be as desired.
LLVM 12 requires sret attributes to have the struct type as a parameter,
and provides no C function for supplying it. Therefore, we must add
another C++ wrapper API for adding the sret attribute.
Fixes ability to build from source in the llvm12 branch.
Conflicts:
* src/clang.zig
* src/llvm.zig
- this file got moved to src/llvm/bindings.zig in master branch so I
had to put the new LLVM arch/os enum tags into it.
* lib/std/target.zig, src/stage1/target.cpp
- haiku had an inconsistency with its default target ABI, gnu vs
eabi. In this commit we make it gnu in both places to match the
latest changes by @hoanga.
* src/translate_c.zig
This is an accident from a merge conflict introduced in
7edb204edfa41e11776ac009da5a20fb1c907f5f.
The new pipe2 code I believe is supposed to work for all posix-like
systems. If haiku needs special handling here, it should be
re-introduced.
* no isHaiku() function since there is not more than one os tag that
this applies to.
* clean up some control flow into a switch
* add some TODO comments to investigate panics that suspiciously look
like they should be compile errors (see #363)
Add a new allocated_registers bitmap to keep track of all callee-saved
registers allocated during generation of this function.
Function(.arm).gen uses this data to generate instructions in the
function prologue and epilogue to push and pop these registers
respectively.
When two pointers are subtracted, both shall point to elements of the
same array object, or one past the last element of the array object;
the result is the difference of the subscripts of the two array elements.
The size of the result is implementation-defined, and its type
(a signed integer type) is ptrdiff_t defined in the <stddef.h> header.
If the result is not representable in an object of that type,
the behavior is undefined.
See C Standard, ยง6.5.6 [ISO/IEC 9899:2011]
Fixes#7216
Reverts bf642204b373e01314ecfb0c50a643dc4b05746f and uses a different
workaround, suggested by @LemonBoy.
There is either a compiler bug or a design flaw somewhere around here.
It does not have to block this branch, but I need to understand exactly
what's going on here and make it so that nobody ever has to run into
this problem again.
