Re: Move new totalorder from libm to libc

[Florian Weimer <fw at deneb dot enyo dot de>]

* Gabriel F. T. Gomes:

> On Sat, 14 Dec 2019, Florian Weimer wrote:
>
>>Should we define the new totalorder in libc, so that it is harder to
>>get the wrong version at run time due to underlinking?
>
> I didn't understand how underlinking could cause a wrong version to be
> selected.  Isn't totalorder only provided by libm?  Where would a user
> program get the old version from?  I can only imagine it getting undefined
> references during link time.

If you build a shared object with -shared, you won't get an undefined
symbol error (with typical toolchains):

$ cat t.c
#include <math.h>

void *p = totalorder;
$ gcc -D_GNU_SOURCE -shared t.c
$ eu-readelf --symbols=.dynsym a.out

Symbol table [ 3] '.dynsym' contains 7 entries:
 1 local symbol  String table: [ 4] '.dynstr'
  Num:            Value   Size Type    Bind   Vis          Ndx Name
    0: 0000000000000000      0 NOTYPE  LOCAL  DEFAULT    UNDEF 
    1: 0000000000000000      0 NOTYPE  WEAK   DEFAULT    UNDEF _ITM_deregisterTMCloneTable
    2: 0000000000000000      0 NOTYPE  WEAK   DEFAULT    UNDEF __gmon_start__
    3: 0000000000000000      0 NOTYPE  GLOBAL DEFAULT    UNDEF totalorder
    4: 0000000000000000      0 NOTYPE  WEAK   DEFAULT    UNDEF _ITM_registerTMCloneTable
    5: 0000000000000000      0 FUNC    WEAK   DEFAULT    UNDEF __cxa_finalize@GLIBC_2.2.5 (2)
    6: 0000000000004020      8 OBJECT  GLOBAL DEFAULT       19 p
$ eu-readelf -d a.out

Dynamic segment contains 25 entries:
 Addr: 0x0000000000003e50  Offset: 0x002e50  Link to section: [ 4] '.dynstr'
  Type              Value
  NEEDED            Shared library: [libc.so.6]
  INIT              0x0000000000001000
[…]

It's much harder to drop -lc from the linker command line, which is
why the symbol would still get a version if it were defined in libc.