patch:

--- dbf_head.c.original 2010-09-28 12:46:04.000000000 +0200
+++ dbf_head.c  2010-09-28 12:46:15.000000000 +0200
@@ -200,9 +200,11 @@
                break;
            case 'D':
                dbf->db_flen = 8;
+               put_short(dbfield.dbf_flen, dbf->db_flen);
                break;
            case 'L':
                dbf->db_flen = 1;
+               put_short(dbfield.dbf_flen, dbf->db_flen);
                break;
            default:
                put_short(dbfield.dbf_flen, dbf->db_flen);

Some DBF tools report created file as corrupted.

Solution is to modify 
file: dbf_head.c
function:  get_dbf_field

before:
-------------------------------
        switch (dbf->db_type) {
            case 'N':
            case 'F':
                        dbf->db_flen = dbfield.dbf_flen[0];
                        dbf->db_fdc = dbfield.dbf_flen[1];
                        break;
            case 'L':
                        dbf->db_flen = 1;
                        break;
            case 'D':
                        dbf->db_flen = 8;
                        break;
            default:
                dbf->db_flen = get_short(dbfield.dbf_flen);
                        break;
        }
-------------------------------

after patch:
-------------------------------
        switch (dbf->db_type) {
            case 'N':
                dbfield.dbf_flen[0] = dbf->db_flen;
                dbfield.dbf_flen[1] = dbf->db_fdc;
                break;
            case 'D':
                dbf->db_flen = 8;
                put_short(dbfield.dbf_flen, dbf->db_flen);
                break;
            case 'L':
                dbf->db_flen = 1;
                put_short(dbfield.dbf_flen, dbf->db_flen);
                break;
            default:
                put_short(dbfield.dbf_flen, dbf->db_flen);
        }
-------------------------------

sorry for above, correct function is: put_dbf_field
and...

before patch:
-------------------------------
       switch (dbf->db_type) {
            case 'N':
                dbfield.dbf_flen[0] = dbf->db_flen;
                dbfield.dbf_flen[1] = dbf->db_fdc;
                break;
            case 'D':
                dbf->db_flen = 8;
                break;
            case 'L':
                dbf->db_flen = 1;
                break;
            default:
                put_short(dbfield.dbf_flen, dbf->db_flen);
        }
-------------------------------


after patch should be:
-------------------------------
        switch (dbf->db_type) {
            case 'N':
                dbfield.dbf_flen[0] = dbf->db_flen;
                dbfield.dbf_flen[1] = dbf->db_fdc;
                break;
            case 'D':
                dbf->db_flen = 8;
                put_short(dbfield.dbf_flen, dbf->db_flen);
                break;
            case 'L':
                dbf->db_flen = 1;
                put_short(dbfield.dbf_flen, dbf->db_flen);
                break;
            default:
                put_short(dbfield.dbf_flen, dbf->db_flen);
        }
-------------------------------

Thanks, I'll try and commit it in a few days.

Great, waiting...

It was tested on production server.

In Clipper some RDD like NTX discard this header value but some like ADS relay on it and report corrputed file.

Some other tools may relay on it too.

Thank you for your bug report. This issue has been fixed
in the latest released version of the package, which you can download at
http://pecl.php.net/get/dbase