[PATCH] D71932: [DWARF] Better detect errors in Address Range Tables.

[James Henderson via Phabricator via llvm-commits]

jhenderson added inline comments.


================
Comment at: llvm/lib/DebugInfo/DWARF/DWARFDebugArangeSet.cpp:167-171
+      return createStringError(
+          errc::invalid_argument,
+          "address range table at offset 0x%" PRIx64
+          " has an invalid tuple (length = 0) at offset 0x%" PRIx64,
+          Offset, *offset_ptr - tuple_size);
----------------
@ikudrin, I know this was committed a while back, but I only just ran into this.

Why was this check added? There's nothing in the DWARF spec that I know of that requires a non-zero length for all non-terminating entries. In our Sony linker, prior to the new tombstoning functionality that has recently been implemented, we would patch a reference in the .debug_aranges to have a zero length, and -1 for the address. This used to be fine, but recent versions of llvm-dwarfdump now can't dump this. Even without the Sony changes, there are occasionally functions in the DWARF with zero length (see also my post earlier today on the llvm-dev list for .debug_loc and zero length entry descriptions).


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D71932/new/

https://reviews.llvm.org/D71932