llvm dev - Oct 2014 - [LLVMdev] hexdump in test cases?

For lld to mimic the darwin linker it needs to support an option that dumps a
binary file which contains a list of the input and output files. This file is
used by Xcode to dynamically update its dependency graph.

I’m a bit stumped on how to test this. Some ideas I have:

* Use REQUIRES: shell, then run hexdump and FileCheck that (is hexdump output
consistent on all *nix platforms?)

* Use REQUIRES: shell, then run sed or perl to format the file with lines and
opcode values and FileCheck that (is sed/perl on all platforms?)

* Since lit uses Python, it seems like I should be able to use python to
transform the file into something FileCheck can process.  But I don’t know how
to invoke python from a RUN: line in a cross platform way.

* Add a C tool to llvm that dumps out a dependency file in a human and FileCheck
readable way.  Seems like overkill to add a directory to llvm/tools/ that
contains a tool only useful for dumping a Xcode file.  Comparing the file bytes
is all I really need.

* Write an llvm-hexdump tool in llvm/tools/ and use that in my test case.

Thoughts?

-Nick

On Thu, Oct 2, 2014 at 6:27 PM, Nick Kledzik <kledzik at apple.com> wrote:
> For lld to mimic the darwin linker it needs to support an option that
> dumps a binary file which contains a list of the input and output files.
> This file is used by Xcode to dynamically update its dependency graph.
>
> I’m a bit stumped on how to test this. Some ideas I have:
>
> * Use REQUIRES: shell, then run hexdump and FileCheck that (is hexdump
> output consistent on all *nix platforms?)
>
I doubt that hexdump output is consistent across all Unix systems.

> * Use REQUIRES: shell, then run sed or perl to format the file with lines
> and opcode values and FileCheck that (is sed/perl on all platforms?)
>
> * Since lit uses Python, it seems like I should be able to use python to
> transform the file into something FileCheck can process.  But I don’t know
> how to invoke python from a RUN: line in a cross platform way.
>
I don't see any reason to use Perl, because it may not be available on some
platforms, and Perl and Python are mostly functionally equivalent.

I'd do with Python. I'd write a binary-to-text converter as a Python
script, put it into test/ directory, and then let lit to run the script by
appending a magic comment to the script like this:

  # RUN: lld -flavor darwin -output %t.binary ...
  # python %s %t.binary | FileCheck %s
  #
  # CHECK: <expected text output here>

* Add a C tool to llvm that dumps out a dependency file in a human
and
> FileCheck readable way.  Seems like overkill to add a directory to
> llvm/tools/ that contains a tool only useful for dumping a Xcode file.
> Comparing the file bytes is all I really need.
>
> * Write an llvm-hexdump tool in llvm/tools/ and use that in my test case.
>
> Thoughts?
>
> -Nick
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20141002/c2b1c26a/attachment.html>

On Oct 2, 2014, at 6:52 PM, Rui Ueyama <ruiu at google.com> wrote:
> 
> I'd do with Python. I'd write a binary-to-text converter as a
Python script, put it into test/ directory, and then let lit to run the script
by appending a magic comment to the script like this:
> 
>   # RUN: lld -flavor darwin -output %t.binary ...  
>   # python %s %t.binary | FileCheck %s
>   #
>   # CHECK: <expected text output here>
I know how to make this work on unix based systems.  But, will adding:
  # RUN: python %s/path/to/hexdump.py %t
work on Windows?  The lit infrastructure does a lot to make the RUN lines work
like a very limited shell even on Windows.

-Nick



-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20141002/989ef96d/attachment.html>