Re: Question about POSIX regex support

[Gallagher James]

On Jun 21, 2007, at 6:07 PM, Bruno Haible wrote:

Hello,

Gallagher James wrote:

System specifics: Fedora Core 6, intel, gcc/++ 4.1.1.

And what's the CPU? x86 or x86_64? 32-bit or 64-bit?

x86_64.

Problem 1: With a string "123abcdef" and a regex of "abc", after
calling regex(...) the resulting pmatch[] has values very different
than before. In the past the result of this call was that pmatch
[0].rm_so was 3 and pmatch[0].rm_eo was 6. Now I get pmatch[0].rm_so
= 3, pmatch[0].rm_eo = 0, pmatch[1].rm_so = 6 and pmatch[1].rm_eo =
0. It seems that the new result is not correct.

Problem 2: Using valgrind I'm getting an Invalid write of 8 bytes on
lines 950 and 951 of regexec.c.

Both symptoms hint towards mismatching definitions of 'regoff_t'. Namely,

it looks like the regoff_t that your program is using/seeing is 32-bit

wide, whereas the regoff_t that gnulib's regex module is using is 64-bit

wide.

Yup, that was it.

Here's a checklist:

1) Verify the values of sizeof (regoff_t) in three different configurations:

     - with the system header (-I/usr/include)

32

- with the gnulib regex.h and with config.h (-Ilib - DHAVE_CONFIG_H)

64

     - with the gnulib regex.h and without config.h (-Ilib)

32

     Test program:

           #if HAVE_CONFIG_H
           # include <config.h>
           #endif
           #include <stdio.h>
           #include <regex.h>

int main () { printf ("%d\n", 8 * sizeof (regoff_t)); return 0; }

2) Verify that you have the -I option in place so that your program uses

     the regex.h from gnulib, not the one from /usr/include.

3) Verify that every source file that includes gnulib headers starts out

     with   #include <config.h>.

Problem fixed.

Thanks very much!

James

Bruno

--
James Gallagher                jgallagher at opendap.org
OPeNDAP, Inc                   406.723.8663