[PATCH] Support native mach-o names in MachOFormat.h

Nick Kledzik kledzik at apple.com
Mon Aug 19 16:13:55 PDT 2013


On Aug 19, 2013, at 3:54 PM, Charles Davis <cdavis5x at gmail.com> wrote:
> On Aug 19, 2013, at 4:22 PM, Nick Kledzik wrote:
> 
>> Wow, that was fast Charles!
> Yeah, I've been sitting on those patches for a while.
>> 
>> I did not know that Support/MachO.h even existed!
> Outside of LLDB, hardly anyone does. They added it, and AFAIK they're the only ones that use it right now. (These patches will obviously change that.)
I asked Greg Clayton (of lldb) about the mach-o constants.  He said we just fleshed out existing files and just followed the naming conventions already in use in those files.  He is the one that suggested I add a #ifndef so that he could build lld with a #define to catch all uses of the old names.

>>  Why does Object/MachOFormat.h and it both define the same things differently?
> I don't know, and that's why I wanted to get rid of one. I suspect that whoever added Object/MachOFormat.h probably didn't know about Support/MachO.h either.

>>   There was some recent changes to ELF headers to layer them.  Is there something analogous we need to do for mach-o?
> We used to do something similar--we had the MachOObject class that does about the same thing as the ELFFile class (i.e. it was a low-level interface to a Mach Object), except without the heavy template meta-programming. MachOObjectFile was then just a thin wrapper around it, just as ELFObjectFile is now just a wrapper around ELFFile. But Rafael Espíndola got rid of that a while back, saying something about the InMemoryStruct template not working right. (He also wanted to add the kind of template meta-programming to Mach-O that ELF was using for its structs, but that got shot down.)

I’d like to hear from Michael Spencer about how he thinks the mach-o layering should work.  For my work in lld, I do * not* want to use templated endian swapping structs.  That is how I wrote Apple’s ld64 linker long ago.  But for lld, which is not a native mach-o linker, there is going to be a conversion layer.  So I plan to do any swapping manually in the conversion layer.  

-Nick


>> 
>> On Aug 19, 2013, at 3:14 PM, Charles Davis <cdavis5x at gmail.com> wrote:
>>> On Aug 19, 2013, at 4:02 PM, Eric Christopher wrote:
>>> 
>>>> Probably wouldn't take much to fix up all the name uses and leave out
>>>> the conditional?
>>> I have two patches; one does what Nick's patch does, but to <llvm/Support/MachO.h>, and also updates everything using it in LLVM to use the new names, and one that strips out the now-redundant <llvm/Object/MachOFormat.h> header and changes everything using *that* to use the new names.
>>> 
>>> Chip
>>> <0001-Support-MachO-Add-a-bunch-of-defines.patch><0002-Move-everything-depending-on-Object-MachOFormat.h-ov.patch>
>>> 
>>>> 
>>>> -eric
>>>> 
>>>> On Mon, Aug 19, 2013 at 2:57 PM, Nick Kledzik <kledzik at apple.com> wrote:
>>>>> [+patch]
>>>>> 
>>>>> include/llvm/Object/MachOFormat.h is odd in that it makes up new names for the public mach-o format structs and constants (e.g. it defines LCT_Segment instead of LC_SEGMENT). LLVM does not handle ELF this way.  LLVM defines SHT_PROGBITS just like the standard ELF header (although llvm puts them into a namespace).  This current set up makes it hard to reconcile the public mach-o documentation with llvm source code.
>>>>> 
>>>>> This patch adds all the standard mach-o names for structs and constants in the llvm::object::mach_o namespace.  All the existing names remain, but I put them inside #ifndef LLVM_NO_OLD_MACH_O_NAMES, so that if someone wanted to switch to the standard names they could define LLVM_NO_OLD_MACH_O_NAMES, do a full build, and update whatever failures are seen.
>>>>> 
>>>>> This issue came up during a code review of lld, as to why lld was defining it own mach-o constants instead of using the ones in llvm.
>>>>> 
>>>>> -Nick
>>>>> 
>>>>> 
>>>>> 
>>>>> _______________________________________________
>>>>> llvm-commits mailing list
>>>>> llvm-commits at cs.uiuc.edu
>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>>> 
>>>> 
>>>> _______________________________________________
>>>> llvm-commits mailing list
>>>> llvm-commits at cs.uiuc.edu
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits





More information about the llvm-commits mailing list