[bug] configure script is getting incorrect value for SIZEOF_OFF_T when using clang
Dominique Pellé
I've compiled Vim-7.3e (2539:9397d2d76340) with the clang compiler on Linux.
All tests pass. However, I see something wrong when doing:
$ vim -c 'saveas! /tmp/foo'
Vim prints:
"/tmp/foo" [New File] 0 lines, 585937717901131776 characters written
Obviously the number of characters written is not correct (should be 0).
Valgrind also reports:
==18757== Conditional jump or move depends on uninitialised value(s)
==18757== at 0x496935B: _itoa (_itoa.c:442)
==18757== by 0x496CC85: vfprintf (vfprintf.c:1613)
==18757== by 0x498B70B: vsprintf (iovsprintf.c:43)
==18757== by 0x49741DA: sprintf (sprintf.c:34)
==18757== by 0x80D850C: msg_add_lines (fileio.c:5246)
==18757== by 0x80DB5EC: buf_write (fileio.c:4885)
==18757== by 0x80A24B0: do_write (ex_cmds.c:2706)
==18757== by 0x80A256B: ex_write (ex_cmds.c:2519)
==18757== by 0x80BA4D8: do_one_cmd (ex_docmd.c:2656)
==18757== by 0x80B739F: do_cmdline (ex_docmd.c:1122)
==18757== by 0x80B7F7E: do_cmdline_cmd (ex_docmd.c:728)
==18757== by 0x8100D64: exe_commands (main.c:2807)
==18757== by 0x80FE1FC: main (main.c:885)
==18757== Uninitialised value was created by a stack allocation
==18757== at 0x80D8494: msg_add_lines (fileio.c:5238)
And while compiling, I see this warning which points to the same problem:
fileio.c:5248:13: warning: conversion specifies type 'long long' but
the argument has type 'off_t' (aka 'long')
[-Wformat]
"%ldL, %lldC", lnum, nchars
~~~^~ ~~~~~~
The number of characters is printed in fileio.c:5248:
5233 void
5234 msg_add_lines(insert_space, lnum, nchars)
5235 int insert_space;
5236 long lnum;
5237 off_t nchars;
5238 {
5239 char_u *p;
5240
5241 p = IObuff + STRLEN(IObuff);
5242
5243 if (insert_space)
5244 *p++ = ' ';
5245 if (shortmess(SHM_LINES))
5246 sprintf((char *)p,
5247 #ifdef LONG_LONG_OFF_T
!5248 "%ldL, %lldC", lnum, nchars
5249 #else
5250 /* Explicit typecast avoids warning on Mac OS X 10.6 */
5251 "%ldL, %ldC", lnum, (long)nchars
5252 #endif
5253 );
Adding debug printf, I see that we have with clang:
* sizeof(off_t) == 4
* sizeof(long long) == 8
Yet in src/auto/config.h, I see this (which is thus incorrect):
/* Defined to the size of off_t */
#define SIZEOF_OFF_T 8
The configure script is getting SIZEOF_OFF_T incorrectly when using
clang compiler.
In src/auto/config.log, I see:
configure:11536: checking size of off_t
configure:11541: clang -o conftest -g -O2 -L/usr/local/lib
conftest.c -lm -lncurses -lnsl -lselinux -lacl -lattr -lgpm >&5
configure:11541: $? = 0
configure:11541: ./conftest
configure:11541: $? = 0
configure:11556: result: 8
Why do we compute size of 'off_t' in configure script, rather than
simply use sizeof(off_t)?
PS: You can install the clang compiler on Ubuntu with "sudo apt-get
install clang".
I had to add #include <wchar.h> in vim.h to be able to compile Vim successfully
with clang.
-- Dominique
Bram Moolenaar
Dominique Pelle wrote:
That seems to be wrong.
> * sizeof(long long) == 8
>
> Yet in src/auto/config.h, I see this (which is thus incorrect):
>
> /* Defined to the size of off_t */
> #define SIZEOF_OFF_T 8
>
> The configure script is getting SIZEOF_OFF_T incorrectly when using
> clang compiler.
>
> In src/auto/config.log, I see:
>
> configure:11536: checking size of off_t
> configure:11541: clang -o conftest -g -O2 -L/usr/local/lib
> conftest.c -lm -lncurses -lnsl -lselinux -lacl -lattr -lgpm >&5
> configure:11541: $? = 0
> configure:11541: ./conftest
> configure:11541: $? = 0
> configure:11556: result: 8
>
> Why do we compute size of 'off_t' in configure script, rather than
> simply use sizeof(off_t)?
That doesn't always work in #ifdef.
Perhaps the AC_SYS_LARGEFILE macro interferes? If it's not used
properly the sizeof(off_t) may differ depending on some flags.
> PS: You can install the clang compiler on Ubuntu with "sudo apt-get
> install clang".
> I had to add #include <wchar.h> in vim.h to be able to compile Vim
> successfully with clang.
Why? Hmm, perhaps it's needed in charset.c, because towlower and
towupper are used there. I'll do that. Does it work now?
--
I started out with nothing, and I still have most of it.
-- Michael Davis -- "Tonight Show"
/// Bram Moolenaar -- Br...@Moolenaar.net -- http://www.Moolenaar.net \\\
/// sponsor Vim, vote for features -- http://www.Vim.org/sponsor/ \\\
\\\ download, build and distribute -- http://www.A-A-P.org ///
\\\ help me help AIDS victims -- http://ICCF-Holland.org ///
Dominique Pellé
I now understand why sizeof(off_t) was wrong. Vim did not compile
with clang and I quickly added #include <wchar.h> to make it compile
I included it *before* #include "auto/config.h" in Vim.h which was a
mistake since auto/config.h #defines _FILE_OFFSET_BITS 64.
Including <wchar.h> after "auto/config.h" makes sizeof(off_t) == 8
Anyway, your recent changeset 2540:8a156630208b makes it compile
and run without problem now with clang.
I think #include <wchar.h> was needed to compile because with clang,
I see this in auto/config.h:
/* #undef HAVE_WCTYPE_H */
Whereas with gcc, I see this:
#define HAVE_WCTYPE_H 1
Somehow, wctype.h header file does not compile!? (but that's not Vim's fault).
Another remark when using clang. I see several of this kind of compilation
warnings which are easy to fix (see attached patch):
ex_cmds.c:2083:18: warning: format string is not a string literal
(potentially insecure) [-Wformat-security]
fprintf(fp_out, _("# You may edit it if you're careful!\n\n"));
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Regards
-- Dominique
.
Bram Moolenaar
Dominique Pelle wrote:
> > Why? Hmm, perhaps it's needed in charset.c, because towlower and
> > towupper are used there. I'll do that. Does it work now?
>
> I now understand why sizeof(off_t) was wrong. Vim did not compile
> with clang and I quickly added #include <wchar.h> to make it compile
> I included it *before* #include "auto/config.h" in Vim.h which was a
> mistake since auto/config.h #defines _FILE_OFFSET_BITS 64.
> Including <wchar.h> after "auto/config.h" makes sizeof(off_t) == 8
>
> Anyway, your recent changeset 2540:8a156630208b makes it compile
> and run without problem now with clang.
>
> I think #include <wchar.h> was needed to compile because with clang,
> I see this in auto/config.h:
>
> /* #undef HAVE_WCTYPE_H */
>
> Whereas with gcc, I see this:
>
> #define HAVE_WCTYPE_H 1
>
> Somehow, wctype.h header file does not compile!? (but that's not
> Vim's fault).
I'm glad there is a simple explanation and that it's fixed now.
> Another remark when using clang. I see several of this kind of compilation
> warnings which are easy to fix (see attached patch):
>
> ex_cmds.c:2083:18: warning: format string is not a string literal
> (potentially insecure) [-Wformat-security]
> fprintf(fp_out, _("# You may edit it if you're careful!\n\n"));
> ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Thanks, I'll include the patch.
--
hundred-and-one symptoms of being an internet addict:
48. You get a tatoo that says "This body best viewed with Netscape 3.1 or
higher."