[PATCH] llvm-objdump macho section extractor.

Tue Aug 9 11:51:32 PDT 2016

Revised patch again.

I removed the hex printing code, let the test handle that. Still using
hexdump and filecheck to verify the contents of the binary.
Added better error handling for file access, and now using all of LLVM's
file IO libraries.
No more error messages on those segment types.

PL

On Mon, Aug 8, 2016 at 2:36 PM, Kevin Enderby <enderby at apple.com> wrote:

>
> On Aug 8, 2016, at 2:29 PM, Puyan Lotfi <plotfi at apple.com> wrote:
>
>
> On Aug 8, 2016, at 1:53 PM, Kevin Enderby <enderby at apple.com> wrote:
>
>
> On Aug 8, 2016, at 11:27 AM, Puyan Lotfi <puyan.lotfi.llvm at gmail.com>
> wrote:
>
> Kevin
>
> I have a new patch attached.
>
> I am fine with only supporting LC_SEGMENT_64 or LC_SEGMENT_64  for the
> time being but I adding a message in case the extract code encounters a
> segment other than those types.
>
>
> I’m not understanding why you consider it an error to have load commands
> other than an LC_SEGMENT_64 or LC_SEGMENT_64 used with Mach-O files used
> with your option.  Is this not expected work with something an x86_64
> Mach-O compiled for a Mac OS X?  It would have other load commands for
> like LC_SYMTAB, LC_UUID, LC_FUNCTION_STARTS, etc.  Or is this option only
> to work in some GPU Mach-O files with only LC_SEGMENT_64 or LC_SEGMENT_64
> load commands?
>
>
> The latter is correct. I can add support for those other segments in
> another commit, but right now I haven't written the code to do that. I
> don't consider it an error, but I figured the extract should print if it
> encounters a segment it doesn't support.
>
>
> I guess maybe I don’t understand what you are saying.  There are only two
> segment load commands, the LC_SEGMENT_64 and LC_SEGMENT_64.  Are you using
> the word “segment” to mean “load command”?  If not what do you mean by a
> "encounters a segment it doesn't support”?
>
> Would you like me to add support for all of the possible segment types??
>
>
> What other “segment types” are there?  Sorry I might be lost I’m just not
> following you.
>
>
>
> I have added some additional flags to narrow down which segment to dump.
>
>
> Also, I added a flag and some code to print the section to stderr in a hex
> format for testing
>
>
> For test wouldn’t you just extract the sections using your option and test
> that it is the same as a file one you add to the test suite?
>
>
> I could, but how should I do that? I mean, should I just run cmp on both
> files?
>
>
> That’s what I would do.  As that is the functionality you are testing/
>
> Can I even do that if the test is run on a system without cmp? Sorry, I
> was just trying to do the thing where I could rely on FileCheck the most
> here.
>
>
> I’m not a test expert.  Maybe there is a way to test that to files are the
> same with FileCheck.  I guess you might want to see if that is possible.
>
>
> I don’t really understand the need for the code to dump the section to
> stderr.   There is already the option -section to dump Mach-O sections.
> Also seems odd that you use the ‘C’ printf style of coding here.  Maybe
> something more C++ like would fit better with the llvm code if you really
> want to keep this bit.
>
>
> Ah, yeah you're right what I am doing here is weird.
>
> and I've added a test case as well.
>
>
> Seems like this could be a bit smaller and test the extracted file and use
> your -extract-macho-sections-dir option.  Without that it seems like the
> test will write into the location of the input file.
>
>
> Ah right I think I forgot to add that flag to the test.
>
> PL
>
>
> Let me known how this looks to you.
>
> Thanks
>
> PL
>
> On Fri, Aug 5, 2016 at 4:22 PM, Kevin Enderby <enderby at apple.com> wrote:
>
>> Hi Puyan,
>>
>> Seems like the functionality would be easier to test if it didn’t write
>> the output files to the same directory as the input file.  And of course
>> you’ll want to add a test case as the llvm tradition with any changes.
>>
>> Also I’m not an llvm style expert but you can likely shorten the logic
>> and not use the a switch statement for your return value from
>> ExtractLoadCommand().  Also it looks like it should get an error with any
>> Mach-O file that has non LC_SEGMENT_64 or LC_SEGMENT_64 commands?  Is that
>> correct and what you want.
>>
>> You also might just want to return a bool for success of failure from
>> ExtractSectionData() and your other routines.
>>
>> Also it llvm seems to like the shortest bit if code so I don’t think you
>> need the “int Res” and could just use an if statement.
>>
>> Kev
>>
>> > On Aug 5, 2016, at 4:06 PM, Puyan Lotfi <puyan.lotfi.llvm at gmail.com>
>> wrote:
>> >
>> > Hi Kevin, All:
>> >
>> > I have a macho-extract tool kicking around locally that I threw
>> together a while back that simply dumps each section in a MachO to disk,
>> and I've decided it's probably high-time to upstream the functionality into
>> llvm-objdump.
>> >
>> > Let me know if this looks good enough to check in, or otherwise
>> feedback would be nice too.
>> >
>> > Thanks
>> >
>> > PL
>> >
>> >
>> > <llvm-objdump-macho-extract.patch>
>>
>>
> <llvm-objdump-macho-extract2.patch>
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160809/259a3d2b/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: llvm-objdump-macho-extract3.patch
Type: application/octet-stream
Size: 11498 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160809/259a3d2b/attachment.obj>