Bug#165603: man-db: me too
On Tue, Oct 29, 2002 at 07:21:01AM -0800, Jeff Bailey wrote:
> On Tue, Oct 29, 2002 at 02:50:29PM +0000, Colin Watson wrote:
> > Thanks a lot! Not to rush you, but any luck so far, or can I help? I've
> > had five duplicates so far, so I have plenty of motivation ...
>
> Sorry, I haven't looked at it yet. I just finished my weekend from hell
> (I had 2 speaking engagements, plus another conference, but had Marcus
> Brinkmann visiting from Germany). It's on my todo list for this week,
> though.
>
> If only I could put off that dayjob thing... =)
I know the feeling. :-)
I think I've found a patch from CVS that clears up this segfault. First
off, a gdb transcript to explain why I think this patch is relevant:
(gdb) run foo
Starting program: /home/cjwatson/src/debian/man-db/man-db-2.4.0/src/apropos foo
foo (1) - No manpage for this program, utility or function.
Program received signal SIGSEGV, Segmentation fault.
0x400c0968 in re_string_context_at (input=0x80bdfe8, idx=25, eflags=0, newline_anchor=0) at regex_internal.c:540
warning: Source file is more recent than executable.
540 c = re_string_byte_at (input, idx);
(gdb) bt
#0 0x400c0968 in re_string_context_at (input=0x80bdfe8, idx=25, eflags=0, newline_anchor=0) at regex_internal.c:540
#1 0x400c0749 in re_string_reconstruct (pstr=0xbffff700, idx=26, eflags=0, newline=0) at regex_internal.c:439
#2 0x400bba7e in re_search_internal (preg=0x804f260, string=0x80bdfa4 "memory management-related functions for use with TIFF files", length=59, start=0, range=70,
stop=134995944, nmatch=0, pmatch=0x0, eflags=0) at regexec.c:665
#3 0x400bb191 in __regexec (preg=0x804f260, string=0x19 <Address 0x19 out of bounds>, nmatch=0, pmatch=0x0, eflags=0) at regexec.c:210
#4 0x08049901 in parse_whatis (page=0xbffffab0 "foo", lowpage=0x8050e00 "foo", whatis=0x80bdfa4 "memory management-related functions for use with TIFF files")
at whatis.c:413
#5 0x08049a8e in apropos (page=0xbffffab0 "foo", lowpage=0x8050e00 "foo") at whatis.c:488
#6 0x08049d94 in search (page=0xbffffab0 "foo") at whatis.c:563
#7 0x0804a103 in main (argc=2, argv=0xbffff974) at whatis.c:711
(gdb) up
#1 0x400c0749 in re_string_reconstruct (pstr=0xbffff700, idx=26, eflags=0, newline=0) at regex_internal.c:439
439 pstr->tip_context = re_string_context_at (pstr, offset - 1, eflags,
(gdb) list
434 offset = idx;
435 }
436
437 if (offset != 0)
438 {
439 pstr->tip_context = re_string_context_at (pstr, offset - 1, eflags,
440 newline);
441 /* Are the characters which are already checked remain? */
442 if (offset < pstr->valid_len)
443 {
(gdb) p offset
$1 = 26
(gdb) p pstr->valid_len
$2 = 0
(gdb) q
Here's the patch. I've backported it such that it applies against glibc
2.1.3-4, since there were some whitespace changes in between. If you
want to take those too for simplicity, I can put together a combined
patch. The crucial bits are chunks 4 and 5 of regex_internal.c; I think
the rest are dependent.
Note that the test suite is still running here.
---------------------
PatchSet 46089
Date: 2002/11/13 04:00:08
Author: roland
Log:
2002-11-11 Isamu Hasegawa <isamu@yamato.ibm.com>
* posix/regex_internal.c (re_string_skip_chars): Also return the last
wide character.
(re_string_reconstruct): Calculate the context by itself when the
offset points out of the valid range.
(re_string_context_at): Use wide character when MB_CUR_MAX > 1.
* posix/regex_internal.h (WIDE_NEWLINE_CHAR): New macro.
(IS_WIDE_WORD_CHAR): New macro.
(IS_WIDE_NEWLINE): New macro.
Members:
posix/regex_internal.c:1.14->1.15
posix/regex_internal.h:1.17->1.18
Index: posix/regex_internal.c
===================================================================
RCS file: /cvs/glibc/libc/posix/regex_internal.c,v
retrieving revision 1.14
retrieving revision 1.15
diff -p -u -u -r1.14 -r1.15
--- posix/regex_internal.c 6 Nov 2002 19:24:20 -0000 1.14
+++ posix/regex_internal.c 13 Nov 2002 04:00:08 -0000 1.15
@@ -66,7 +66,8 @@ static void re_string_construct_common (
re_string_t *pstr,
RE_TRANSLATE_TYPE trans, int icase);
#ifdef RE_ENABLE_I18N
-static int re_string_skip_chars (re_string_t *pstr, int new_raw_idx);
+static int re_string_skip_chars (re_string_t *pstr, int new_raw_idx,
+ wint_t *last_wc);
#endif /* RE_ENABLE_I18N */
static re_dfastate_t *create_newstate_common (re_dfa_t *dfa,
const re_node_set *nodes,
@@ -333,21 +334,24 @@ build_wcs_upper_buffer (pstr)
Return the index. */
static int
-re_string_skip_chars (pstr, new_raw_idx)
+re_string_skip_chars (pstr, new_raw_idx, last_wc)
re_string_t *pstr;
int new_raw_idx;
+ wint_t *last_wc;
{
mbstate_t prev_st;
int rawbuf_idx, mbclen;
+ wchar_t wc = 0;
/* Skip the characters which are not necessary to check. */
for (rawbuf_idx = pstr->raw_mbs_idx + pstr->valid_len;
rawbuf_idx < new_raw_idx;)
{
- int remain_len = pstr->len - rawbuf_idx;
+ int remain_len;
+ remain_len = pstr->len - rawbuf_idx;
prev_st = pstr->cur_state;
- mbclen = mbrlen ((const char *) pstr->raw_mbs + rawbuf_idx, remain_len,
- &pstr->cur_state);
+ mbclen = mbrtowc (&wc, (const char *) pstr->raw_mbs + rawbuf_idx,
+ remain_len, &pstr->cur_state);
if (BE (mbclen == (size_t) -2 || mbclen == (size_t) -1 || mbclen == 0, 0))
{
/* We treat these cases as a singlebyte character. */
@@ -357,6 +361,7 @@ re_string_skip_chars (pstr, new_raw_idx)
/* Then proceed the next character. */
rawbuf_idx += mbclen;
}
+ *last_wc = (wint_t) wc;
return rawbuf_idx;
}
#endif /* RE_ENABLE_I18N */
@@ -436,12 +441,12 @@ re_string_reconstruct (pstr, idx, eflags
if (offset != 0)
{
- pstr->tip_context = re_string_context_at (pstr, offset - 1, eflags,
- newline);
/* Are the characters which are already checked remain? */
if (offset < pstr->valid_len)
{
/* Yes, move them to the front of the buffer. */
+ pstr->tip_context = re_string_context_at (pstr, offset - 1, eflags,
+ newline);
#ifdef RE_ENABLE_I18N
if (MB_CUR_MAX > 1)
memmove (pstr->wcs, pstr->wcs + offset,
@@ -466,11 +471,26 @@ re_string_reconstruct (pstr, idx, eflags
if (MB_CUR_MAX > 1)
{
int wcs_idx;
- pstr->valid_len = re_string_skip_chars (pstr, idx) - idx;
+ wint_t wc;
+ pstr->valid_len = re_string_skip_chars (pstr, idx, &wc) - idx;
for (wcs_idx = 0; wcs_idx < pstr->valid_len; ++wcs_idx)
pstr->wcs[wcs_idx] = WEOF;
+ if (pstr->trans && wc <= 0xff)
+ wc = pstr->trans[wc];
+ pstr->tip_context = (IS_WIDE_WORD_CHAR (wc) ? CONTEXT_WORD
+ : ((newline && IS_WIDE_NEWLINE (wc))
+ ? CONTEXT_NEWLINE : 0));
}
+ else
#endif /* RE_ENABLE_I18N */
+ {
+ int c = pstr->raw_mbs[pstr->raw_mbs_idx + offset - 1];
+ if (pstr->trans)
+ c = pstr->trans[c];
+ pstr->tip_context = (IS_WORD_CHAR (c) ? CONTEXT_WORD
+ : ((newline && IS_NEWLINE (c))
+ ? CONTEXT_NEWLINE : 0));
+ }
}
if (!MBS_CASE_ALLOCATED (pstr))
{
@@ -537,10 +557,32 @@ re_string_context_at (input, idx, eflags
return ((eflags & REG_NOTEOL) ? CONTEXT_ENDBUF
: CONTEXT_NEWLINE | CONTEXT_ENDBUF);
}
- c = re_string_byte_at (input, idx);
- if (IS_WORD_CHAR (c))
- return CONTEXT_WORD;
- return (newline_anchor && IS_NEWLINE (c)) ? CONTEXT_NEWLINE : 0;
+ if (MB_CUR_MAX == 1)
+ {
+ c = re_string_byte_at (input, idx);
+ if (IS_WORD_CHAR (c))
+ return CONTEXT_WORD;
+ return (newline_anchor && IS_NEWLINE (c)) ? CONTEXT_NEWLINE : 0;
+ }
+ else
+ {
+ wint_t wc;
+ int wc_idx = idx;
+ while(input->wcs[wc_idx] == WEOF)
+ {
+#ifdef DEBUG
+ /* It must not happen. */
+ assert (wc_idx >= 0);
+#endif
+ --wc_idx;
+ if (wc_idx < 0)
+ return input->tip_context;
+ }
+ wc = input->wcs[wc_idx];
+ if (IS_WIDE_WORD_CHAR (wc))
+ return CONTEXT_WORD;
+ return (newline_anchor && IS_WIDE_NEWLINE (wc)) ? CONTEXT_NEWLINE : 0;
+ }
}
/* Functions for set operation. */
Index: posix/regex_internal.h
===================================================================
RCS file: /cvs/glibc/libc/posix/regex_internal.h,v
retrieving revision 1.17
retrieving revision 1.18
diff -p -u -u -r1.17 -r1.18
--- posix/regex_internal.h 6 Nov 2002 20:35:39 -0000 1.17
+++ posix/regex_internal.h 13 Nov 2002 04:00:08 -0000 1.18
@@ -30,6 +30,7 @@
/* The character which represents newline. */
#define NEWLINE_CHAR '\n'
+#define WIDE_NEWLINE_CHAR L'\n'
/* Rename to standard API for using out of glibc. */
#ifndef _LIBC
@@ -355,6 +356,8 @@ typedef struct bin_tree_t bin_tree_t;
#define IS_WORD_CHAR(ch) (isalnum (ch) || (ch) == '_')
#define IS_NEWLINE(ch) ((ch) == NEWLINE_CHAR)
+#define IS_WIDE_WORD_CHAR(ch) (iswalnum (ch) || (ch) == L'_')
+#define IS_WIDE_NEWLINE(ch) ((ch) == WIDE_NEWLINE_CHAR)
#define NOT_SATISFY_PREV_CONSTRAINT(constraint,context) \
((((constraint) & PREV_WORD_CONSTRAINT) && !IS_WORD_CONTEXT (context)) \
---------------------
PatchSet 46090
Date: 2002/11/13 04:00:16
Author: roland
Log:
.
Members:
ChangeLog:1.6893->1.6894
Index: ChangeLog
===================================================================
RCS file: /cvs/glibc/libc/ChangeLog,v
retrieving revision 1.6893
retrieving revision 1.6894
diff -p -u -u -r1.6893 -r1.6894
--- ChangeLog 12 Nov 2002 22:02:38 -0000 1.6893
+++ ChangeLog 13 Nov 2002 04:00:16 -0000 1.6894
@@ -1,3 +1,14 @@
+2002-11-11 Isamu Hasegawa <isamu@yamato.ibm.com>
+
+ * posix/regex_internal.c (re_string_skip_chars): Also return the last
+ wide character.
+ (re_string_reconstruct): Calculate the context by itself when the
+ offset points out of the valid range.
+ (re_string_context_at): Use wide character when MB_CUR_MAX > 1.
+ * posix/regex_internal.h (WIDE_NEWLINE_CHAR): New macro.
+ (IS_WIDE_WORD_CHAR): New macro.
+ (IS_WIDE_NEWLINE): New macro.
+
2002-11-12 Andreas Jaeger <aj@suse.de>
* sysdeps/x86_64/strchr.S: Don't use one register for two
Thanks,
--
Colin Watson [cjwatson@flatline.org.uk]
Reply to: