SDBM_File under MS-Windows95/98 does not work correctly. (APR#1302)

Full_Name: SUZUKI, Norio
Version: ActivePerl 620
OS: Windows 98
Submission from: (NULL) (210.198.167.27)


I found the same SDBM_File's BUG as Mr. Dave Mullen reported in modules/495.
When running the script shown below under MS-Windows95(or 98), you can find
some keys are missing after the 29th key is inserted.

---^
use Fcntl;
use SDBM_File;
unlink ('sdbm.pag','sdbm.dir');

%check = ();
$s0 = 'a'x 125;

for $n (1..30) {
    $s = $s0.sprintf("%03d",$n);
    tie(%sdbm, 'SDBM_File', 'sdbm', O_RDWR|O_CREAT, 0666);
    $sdbm{$s}++;
    $check{$s}++;
    $n_sdbm  = keys %sdbm;
    $n_check = keys %check;
    untie %sdbm;
    $size = -s 'sdbm.pag';
    print STDERR "n:$n,keys of \%check:$n_check,keys of sdbm:$n_sdbm, size of
sdbm.pag:$size\n";
    last unless $n_sdbm == $n_check;
}

---$

When I asked this to Mr. WATANABE, Hirofumi (a patch author of jperl), he
showed a patch and told me that Unix file system suported a gap or hole, 
but MS-Windows95/98 didn't.

> notes of lseek(2)
> Seeking far beyond the end of  a  file,  then  writing,  may
> create a gap or "hole", which occupies no physical space and
> reads as zeros.

--- AP509_source/ext/SDBM_File/sdbm/sdbm.c.orig	Fri Jan 22 09:40:54 1999
+++ AP509_source/ext/SDBM_File/sdbm/sdbm.c	Fri Jan 22 10:08:49 1999
@@ -270,2 +270,6 @@
 	char twin[PBLKSIZ];
+#if defined(DOSISH) || defined(WIN32)
+	char zer[PBLKSIZ];
+	long oldtail;
+#endif
 	char *pag = db->pagbuf;
@@ -292,2 +296,19 @@
  */
+
+#if defined(DOSISH) || defined(WIN32)
+		/*
+		 * Fill hole with 0 if made it.
+		 * (hole is NOT read as 0)
+		 */
+		oldtail = lseek(db->pagf, 0L, SEEK_END);
+		memset(zer, 0, PBLKSIZ);
+		while (OFF_PAG(newp) > oldtail) {
+			if (lseek(db->pagf, 0L, SEEK_END) < 0 ||
+			    write(db->pagf, zer, PBLKSIZ) < 0) {
+
+				return 0;
+			}
+			oldtail += PBLKSIZ;
+		}
+#endif
 		if (hash & (db->hmask + 1)) {

• SDBM_File under MS-Windows95/98 does not work correctly. (APR#1302) by kipp
• Re: SDBM_File under MS-Windows95/98 does not work correctly. (APR#1302) by Jarkko Hietaniemi