[PATCH] D33996: Improve consistency and flexibility with llvm-pdbdump subcommands

[Zachary Turner via Phabricator via llvm-commits]

zturner added a comment.

In https://reviews.llvm.org/D33996#776225, @inglorion wrote:

> While we're at it - how much value is there in having both yaml and raw that is almost the same, but with slightly different syntax? I've added some code to dump things in raw format before we understood what they meant, but I think now that we can dump pretty much everything as YAML, we should consider removing raw.


I've been thinking about what to do here, and I think there is a better option.  We should change `raw` mode to be more concise.  raw was originally supposed to be for human consumption while yaml was for machine consumption.  But they are equally verbose, and there's no clear argument that raw is more readable than yaml.  It might even be worse.  So what I've been thinking about is changing raw to be very compressed.  For example, instead of this:

  D:\foo> llvm-pdbdump raw -all foo.pdb
  FileHeaders {
    BlockSize: 4096
    FreeBlockMap: 2
    NumBlocks: 1145
    NumDirectoryBytes: 5484
    Unknown1: 0
    BlockMapAddr: 1143
    NumDirectoryBlocks: 2
    DirectoryBlocks: [1141, 1142]
    NumStreams: 236
  }
  Streams [
    Stream 0: [Old MSF Directory] (40 bytes)
    Stream 1: [PDB Stream] (118 bytes)
    Stream 2: [TPI Stream] (407284 bytes)
    Stream 3: [DBI Stream] (210538 bytes)
    Stream 4: [IPI Stream] (496844 bytes)
    Stream 5: [Named Stream "/LinkInfo"] (0 bytes)
  ...
  PDB Stream {
    Version: 20000404
    Signature: 0x59303684
    Age: 2
    Guid: {6EE2E173-08F2-BF4D-AA11-4734E99F4974}
    Features: 0x1
    Named Streams {
      /names: 6
      /LinkInfo: 5
      /src/headerblock: 233
    }
  }
  Type Info Stream (TPI) {
    TPI Version: 20040203
    Record count: 8478
    Records [
      {
        ArgList (0x1000) {
          TypeLeafKind: LF_ARGLIST (0x1201)
          NumArgs: 0
          Arguments [
          ]
        }
      }
      {
        Procedure (0x1001) {
          TypeLeafKind: LF_PROCEDURE (0x1008)
          ReturnType: void (0x3)
          CallingConvention: NearC (0x0)
          FunctionOptions [ (0x0)
          ]
          NumParameters: 0
          ArgListType: () (0x1000)
        }
      }

We print this:

  D:\foo> llvm-pdbdump raw -all foo.pdb
  MSF [BlockSize=4096, FreeBlockMap=2, NumBlocks=1145, DirSize=5484, BlockMapAddr=1143, NumDirBlocks=2, Streams=236]
  
  Streams [
    Stream 0: [Old MSF Directory] (40 bytes)
    Stream 1: [PDB Stream] (118 bytes)
    Stream 2: [TPI Stream] (407284 bytes)
    Stream 3: [DBI Stream] (210538 bytes)
    Stream 4: [IPI Stream] (496844 bytes)
    Stream 5: [Named Stream "/LinkInfo"] (0 bytes)
    ...
  ]
  
  PDB Stream [Ver: 20000404, Sig: 0x59303684, Age: 2, Guid: {6EE2E173-08F2-BF4D-AA11-4734E99F4974}, Features: 0x1]
    Named Streams {
      /names: 6
      /LinkInfo: 5
      /src/headerblock: 233
    }
    
  TPI Stream [Ver: 20040203]
    Records [
      TI: 0x1000 - LF_ARGLIST (0x1201) {}
      TI: 0x1001 - LF_PROCEDURE (0x1001) { Ret: void (0x03), Conv: NearC (0x0), Options: 0x0, Args: () (0x1000)}
      ...
    ]

or some such.


https://reviews.llvm.org/D33996