LC_SEGMENT command 0 filesize field greater than vmsize field

[Michel Fortin]

On 2010-11-13 07:16:27 -0500, Jacob Carlborg <doob at me.com> said:

> On 2010-11-12 14:41, Michel Fortin wrote:
>> On 2010-11-12 04:08:56 -0500, Jacob Carlborg <doob at me.com> said:
>> 
>>> Now this really starts to annoy me. I don't know if it's a bug in Mac
>>> OS X or in DMD but I'm getting this error when building some libraries
>>> (Tango, for example):
>>> 
>>> /usr/bin/ranlib: object: libtango.a(tango-coreBitArray-release.o)
>>> malformed object (LC_SEGMENT command 0 filesize field greater than
>>> vmsize field)
>>> ar: internal ranlib command failed
>> 
>> I've seen something similar too with some transitory test cases while
>> playing with DMD to generate Objective-C data structures. I'd tend to
>> believe there's a bug in the backend. I didn't know this bug could be
>> trigged by compiling code with regular DMD, it could have been something
>> I trigged playing with the compiler to make it generate extra Objectve-C
>> segments/sections.
>> 
>> If you want to inspect the object file, I suggest you extract
>> tango-coreBitArray-release.o from the libtango.a archive and play with
>> otool, objdump, dumpobj, and perhaps obj2asm. The first two are provided
>> with Apple's developer tools so they and tend to better reflect the
>> linker in their understanding of the object file format; the later two
>> come with DMD and better reflect Walter's implementation.
> 
> I used the dumpobj tool to inspect the LC_SEGMENT and just as the error 
> message says, the filesize field is greater than the vmsize field.

backend/machobj.c line 765-766:

    segment_cmd.vmsize = vmaddr;
    segment_cmd.filesize = foffset - segment_cmd.fileoff;

I'd put an assert just after those two lines to make sure vmsize is 
greater or equal to filesize. Running dmd in the debugger should make 
it break on the failed assertion and perhaps you'll be able to 
investigate. It seems likely that the problem comes from the loop just 
above.

-- 
Michel Fortin
michel.fortin at michelf.com
http://michelf.com/