[Toybox] One for Elliott.

[Rob Landley]

On 5/13/20 11:10 AM, enh wrote:
> yeah, i actually left a comment in the source about this one.
> 
> basically, binutils readelf lies --- it takes two passes over the
> array, counting non-null entries the first time and then only
> admitting to the existence of the non-null entries. at the same time
> (and it's hard to know which way round the causality goes here) some
> versions of some of binutils' linkers would leave a bunch of null
> entries at the end, as you see from this genuine example that _wasn't_
> specifically picked for that purpose.
> 
> i resolved in favor of telling the exact truth (on the theory that
> that's kind of why anyone's running readelf in the first place). but i
> can see an argument for doing what the other guy does too, and can
> send you the patch if you'd rather do that. at least we have the code
> comment to explain the otherwise unnecessary first pass over the array
> :-)

I'm happy to go with the simpler implementation, and "binutils' readelf lies to
hide binutils ld's bug" sounds very gnu.

I could run the output through "head -n xx" to make them consistent in the test....

>> -Dynamic section at offset 0xd98 contains 33 entries:
>> +Dynamic section at offset 0xd98 contains 29 entries:

Modulo that.

>>    Tag                Type                 Name/Value
>>   0x0000000000000001 (NEEDED)             Shared library: [libc.so]
>>   0x0000000000000001 (NEEDED)             Shared library: [libm.so]
>> @@ -30,7 +30,3 @@
>>   0x000000006ffffff0 (VERSYM)             0x528
>>   0x000000006ffffff9 (RELACOUNT)          4
>>   0x0000000000000000 (NULL)               0x0
>> - 0x0000000000000000 (NULL)               0x0
>> - 0x0000000000000000 (NULL)               0x0
>> - 0x0000000000000000 (NULL)               0x0
>> - 0x0000000000000000 (NULL)               0x0

Hmmm... maybe it would be better to add a test binary that binutils _doesn't_
lie about? I can "toyonly" this one with a comment about binutils lying to hide
trailing NULL entries...

Anyway, "not accidental or regression, known issue, should have TODO".

Thanks,

Rob