OCaml Magic

> moved /usr/share/file to /usr/share/file.old to trigger the bug.

May you provide more details? I just did that and got (in the toploop):

# let c = Magic.make [];;
/etc/magic, 4: Warning using regular magic file `/usr/share/file/magic'
Exception: Magic.Failure "Magic.load".

which is what is expected. I can also do

# Magic.load c ["/usr/share/file/magic.mime"];;

without problem.

> let filenames = (if filenames = [] then ["/usr/share/file/magic"]
> else filenames) in

I removed this hack.

I used your new files in MLDonkey on Debian and moved /usr/share/file
to /usr/share/file.old to trigger the bug.

This is the gdb backtrace:

, 4: Warning using regular magic file `/usr/share/file/magic.mime'

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0xb7bd2ae0 (LWP 6774)]
0x0832342d in raise_magic_failure (msg=0x8353aff "Magic.load") at
src/utils/lib/magiclib_stub.c:57
57 raise_with_string(*exn, (char *) msg);
(gdb) bt
#0 0x0832342d in raise_magic_failure (msg=0x8353aff "Magic.load") at
src/utils/lib/magiclib_stub.c:57
#1 0x0832355b in ocaml_magic_load_default (c=-1215808820) at
src/utils/lib/magiclib_stub.c:301
#2 0x0820f74b in camlMagiclib__load_101 ()
#3 0xb7883acc in ?? ()
#4 0x0820fa85 in camlMagic_magic__magic_cookie_mime_70 ()
#5 0x0820fa95 in camlMagic_magic__magic_cookie_mime_70 ()

I will test it, but your code

(* FIXME: "/usr/share/file/magic" is a temporary choice until the
toplevel segfault is solved. *)
let filenames = (if filenames = [] then ["/usr/share/file/magic"]
else filenames) in

is not sufficient because, for example, on Debian Sarge the magic
database is in /usr/share/misc/file/ and on FreeBSD 7.0 it can
be found in /usr/share/misc/.

May you check again with OCaml 3.10 and the version 0.7.1 of ocaml-magic?

This time on Debian Etch here is a gdb backtrace using debug libraries
libc6-dbg
and http://people.debian.org/~smimram/mldonkey/libmagic1_4.17-2_i386.deb
I reproduced the bug because I removed /usr/share/file/*.mgc

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 16384 (LWP 29959)]
0xa7c58a0d in *__GI_memset (dstpp=0x85d45a0, c=<value optimized out>,
len=128) at ../sysdeps/i386/memset.c:65
65 ../sysdeps/i386/memset.c: Datei oder Verzeichnis nicht gefunden.
in ../sysdeps/i386/memset.c
(gdb) bt
#0 0xa7c58a0d in *__GI_memset (dstpp=0x85d45a0, c=<value optimized out>,
len=128) at ../sysdeps/i386/memset.c:65
#1 0xa7e7816d in apprentice_file (ms=0x8553dc8, magicp=0xafca3ccc,
nmagicp=0xafca3cc8,
fn=0x8550073 "/usr/share/file/magic", action=0) at apprentice.c:567
#2 0xa7e79615 in file_apprentice (ms=0x8553dc8, fn=0x8550073
"/usr/share/file/magic", action=0)
at apprentice.c:183
#3 0xa7e776c2 in magic_load (ms=0x8553dc8, magicfile=0x0) at magic.c:156
#4 0x0832e129 in ?? ()

Debug5 is reached, before debug6 is reached the segfault occurs


#define LOAD(fname) \
fprintf(stdout, "debug3\n"); \
magic_t cookie = COOKIE_VAL(c); \
fprintf(stdout, "debug4\n"); \
\
if (cookie == NULL) caml_invalid_argument("Magiclib.load"); \
fprintf(stdout, "debug5\n"); \
if (magic_load(cookie, fname) < 0) \
raise_on_error("Magiclib.load: ", cookie); \
fprintf(stdout, "debug6\n"); \
CAMLreturn(Val_unit)

The bug is still present when using the updated files from September or
the original ones.
They make no difference. this time I tested on a fresh installed Slackware
11.0

# file -v
file-4.17
magic file from /etc/file/magic

# ls -la /etc/file/
total 480
-rw-r--r-- 1 root root 443077 2006-08-01 05:13 magic
-rw-r--r-- 1 root root 31136 2006-08-01 05:13 magic.mime

open("/etc/file/magic.mgc", O_RDONLY|O_LARGEFILE) = -1 ENOENT (No such
file or directory)
open("/etc/file/magic", O_RDONLY|O_LARGEFILE) = 13
fstat64(13, {st_mode=S_IFREG|0644, st_size=443077, ...}) = 0
mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0)
= 0x404df000
read(13, "# Magic\n# Magic data for file(1)"..., 4096) = 4096
--- SIGSEGV (Segmentation fault) @ 0 (0) ---

After compiling magic.mgc: "cd /etc/file && file -C" the segfault
disappears
so Ocaml-Magic in its current state fails with the text version of the
libmagic database.

# ls -la /etc/file/
total 1428
-rw-r--r-- 1 root root 443077 2006-08-01 05:13 magic
-rw-r--r-- 1 root root 966016 2006-11-19 01:16 magic.mgc
-rw-r--r-- 1 root root 31136 2006-08-01 05:13 magic.mime

Logged In: YES
user_id=559957

This bug is also present on Debian:
http://bugs.debian.org/378831

Logged In: YES
user_id=220888

same here... is there any workaround before fix ?