[PATCH] D93265: [llvm-objdump-macho] print per-second-level-page encodings for option --unwind-info

Mon Dec 21 11:57:52 PST 2020

gkm added inline comments.

================
Comment at: llvm/test/tools/llvm-objdump/MachO/unwind-info-excess-x86_64.test:1
+# RUN: llvm-objdump --unwind-info %p/Inputs/unwind-info-excess.macho-x86_64 2>/dev/null | FileCheck %s
+
----------------
MaskRay wrote:
> A prebuilt binary is difficult to inspect and update. Can you rewrite the input with yaml2obj?
> 
> There are plenty of examples in the ELF/ folder (and in various `test/tools/*/ELF` directories)
> 
> The test should also include a file level comment documenting its purpose.
Very good. FYI, I was following prior practice, where there are already 71 binaries and not even one `*.yaml` in `llvm/test/tools/llvm-objdump/MachO/Inputs/`.

Yaml isn't so great either: it's text, but nearly as opaque as a binary. I favor using `llvm-mc` on assembler inputs. My compact-unwind test in `lld/test/MachO` goes further: Because the assembler input is 1 MiB, I have a python script (which also serves as a fuzzer) to generate it. The downside is significant CPU time for `llvm-mc` to process the test. It's not so bad on a release tree (approx 1s), but slow on a debug tree (approx 10s). I could solve that by preferring the installed `llvm-mc`, but all the existing tests just do `RUN: llvm-mc ...` and get the build tree's version, so I'm not sure how to do that.

Please advise.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D93265/new/

https://reviews.llvm.org/D93265