Original bug ID: 3944
Reporter: anonymous
Assigned to: @xavierleroy
Status: closed (set by @xavierleroy on 2006-01-04T09:28:04Z)
Resolution: fixed
Priority: normal
Severity: minor
Version: 3.09.0
Category: ~DO NOT USE (was: OCaml general)
Monitored by: smimram jcristau

Bug description

Hi,

Recently the compilation of missinglib (an ocaml library without bindings) failed on sparc (in Debian) because the test program was producing a bus error [1].

Gdb gives the following:

(gdb) run
Starting program: /tmp/missinglib-0.4.10.debian/test/runtests.opt
........................................................
Program received signal SIGBUS, Bus error.
0x0005f698 in caml_format_float ()
(gdb) where
#0 0x0005f698 in caml_format_float ()
#1 0x0006a3ec in caml_c_call ()

0x0005f680 <caml_format_float+240>: call 0x5d110 <caml_stat_alloc>
0x0005f684 <caml_format_float+244>: nop
0x0005f688 <caml_format_float+248>: mov %o0, %l0
0x0005f68c <caml_format_float+252>: mov %l0, %o0
0x0005f690 <caml_format_float+256>: mov %i0, %o1
0x0005f694 <caml_format_float+260>: call 0x7d814
0x0005f698 <caml_format_float+264>: ldd [ %i1 ], %o2
0x0005f69c <caml_format_float+268>: call 0x5d3d4 <caml_copy_string>
(gdb) p (double*)$i1
$1 = (double *) 0xa3f74

The function caml_format_float is apparently called to print the duration of the tests. Notice that 0xa3f74 is not a multiple of 8, which I guess is the cause of this issue. This corresponds to the source code line

sprintf(dest, String_val(fmt), Double_val(arg));

So, here %i1 contains the float argument (arg) of the function. Maybe the float allocation should always return a 8-byte aligned cell? Aligning floats (i.e. C doubles) on values (here long) is not always enough since in this case sizeof(double) > sizeof(value) (8 > 4). For example, caml_copy_float does not to take care of alignment.

Thanks.

[1] http://buildd.debian.org/fetch.php?&pkg=missinglib&ver=0.4.10.debian-2&arch=sparc&stamp=1136143511&file=log&as=raw