[PATCH] D70658: [LLD][ELF] - Make compression level be dependent on -On.

[Shoaib Meenai via Phabricator via llvm-commits]

smeenai added inline comments.


================
Comment at: lld/test/ELF/compressed-debug-level.test:23
+# HEADER: [Nr] Name        Type     Address  Off    Size
+# LEVEL1: [ 1] .debug_info PROGBITS 00000000 000094 00001c
+# LEVEL6: [ 1] .debug_info PROGBITS 00000000 000094 00001b
----------------
grimar wrote:
> smeenai wrote:
> > MaskRay wrote:
> > > smeenai wrote:
> > > > The size appears to depend on the zlib version. For example, I'm running with zlib 1.2.8, which manages to compress these contents to the same size with both level 1 and 6. Is there any way to make this agnostic to zlib version?
> > > I run zlib 1.2.11 and it works for me. Can you alter the input a bit to make the length different for level 1 and level 6, for zlib 1.2.8?
> > I fixed it with 75fd939bb917e8f843395684a2970d86bc0199c0.
> > I fixed it with 75fd939bb917e8f843395684a2970d86bc0199c0.
> 
> Thanks! FTR, I used zlib 1.2.3 under windows when wrote this. Now I wonder how much brittle to check the size produced.
> Seems it is not safe enough?
> I think what we can do instead is to print llvm-readelf output to files for pairs [-O0, -O2], [-O1, -O2] etc and then do something
> like
> 
> ```
> # CHECK: [ 1] .debug_info PROGBITS 00000000 000094 [[SIZE:*]]
> # CHECK-NOT: [ 1] .debug_info PROGBITS 00000000 000094 [[SIZE]]
> ```
> 
> i.e. this should check that sized are just different. Should we?
The issue I ran into with the original test case was that the compressed data had the same size with levels 1 and 6, so just verifying the sizes were different wouldn't help.

I agree that checking sizes isn't super robust, but it seems to work okay for now with the updated test case? 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D70658/new/

https://reviews.llvm.org/D70658