| Issue |
166993
|
| Summary |
yaml2obj fails to roundtrip Mach-O dSYM files
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
rjmansfield
|
`yaml2obj` fails when converting Mach-O dSYM YAML (generated by `obj2yaml`) back to a binary with section offset alignment errors.
This seems to happen with basically any dSYM file. For example:
```
#include <stdio.h>
int add(int a, int b) { return a + b; }
int multiply(int x, int y) { return x * y; }
struct Point { int x, y; const char *label; };
struct Point make_point(int x, int y, const char *label) {
struct Point p = {x, y, label};
return p;
}
int main(void) {
int a = add(5, 3);
struct Point p = make_point(10, 20, "test");
printf("%d, %d\n", a, p.x);
return 0;
}
```
```
$ clang repro.c -o repro -g
$ ~/llvm/llvm-project/build/bin/obj2yaml repro.dSYM/Contents/Resources/DWARF/repro -o repro.yaml
$ ~/llvm/llvm-project/build/bin/yaml2obj repro.yaml -o repro_roundtrip.out
yaml2obj: error: wrote too much data somewhere, section offsets in section __debug_aranges for segment __DWARF don't line up: [cursor=0x210c], [fileStart=0x0], [sectionOffset=0x20b7]
$ ~/llvm/llvm-project/build/bin/obj2yaml --version
LLVM (http://llvm.org/):
LLVM version 22.0.0git
Optimized build.
```
using HEAD as of this report which is 917d815d4ee4.
It looks like the problem is that `obj2yaml` copies the DWARF compilation unit `Length` field e.g 347 bytes from the binary
and then when `yaml2obj` re-emits the DIEs, ULEB128 encoding and other things it can cause the size to differ e.g. 363 bytes and then this size difference causes the next section to be misaligned. Maybe it would make sense to not to emit the `Length` field from obj2yaml and let `yaml2obj` calculate the `Length` based on what it actually emits.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
