[TUHS] question about mkfs(8) and stdio

[TUHS] question about mkfs(8) and stdio

[Kenneth Stailey]

Did stdio buffering change over time?

If you look at an old BSD mkfs for cylinder-group style file systems (not 7th
Ed filesystems) you will see the super block backup loop does not fflush(stdio)
between printf()s

        printf("super-block backups (for fsck -b#) at:");
        for (cylno = 0; cylno < sblock.fs_ncg; cylno++) {
                initcg(cylno);
                if (cylno % 10 == 0)
                        printf("\n");
                printf(" %d,", fsbtodb(&sblock, cgsblock(&sblock, cylno)));
        }

but I have memories that the superblocks were printed out one at a time as if
fflush(stdout) was called between them rather than one line at a time with
line-buffered stdio.

At some point I thought "SysV must have broke this" since newfs would print out
a complete row of superblock numbers at once with a big delay between the rows
rather than each superblock number with a short delay between each number.

But when I go searching the oldest BSD code has no fflush(stdout) the way
modern FreeBSD does:

        for (cylno = 0; cylno < sblock.fs_ncg; cylno++) {
                initcg(cylno, utime);
                if (mfs)
                        continue;
                j = snprintf(tmpbuf, sizeof(tmpbuf), " %ld%s",
                    fsbtodb(&sblock, cgsblock(&sblock, cylno)),
                    cylno < (sblock.fs_ncg-1) ? "," : "" );
                if (i + j >= width) {
                        printf("\n");
                        i = 0;
                }
                i += j;
                printf("%s", tmpbuf);
                fflush(stdout);
        }

Did stdio buffering change over time so that line buffering became the default?

Thanks,
Ken


__________________________________
Do you Yahoo!?
Protect your identity with Yahoo! Mail AddressGuard
http://antispam.yahoo.com/whatsnewfree

[TUHS] question about mkfs(8) and stdio

[Michael Sokolov]

Kenneth Stailey <kstailey at yahoo.com> wrote:

> Did stdio buffering change over time?

Line buffering was a Berkeley innovation.  stdout became line-buffered by
default (when it is a terminal) in 4.0BSD.  4.2BSD added setlinebuf(3) to allow
people to make stdout or stderr line-buffered when they want to.  4.3BSD
extended it to work on any stream, not just stdout or stderr.

> If you look at an old BSD mkfs for cylinder-group style file systems (not 7th
> Ed filesystems)

Ahmm, you call that old?  To me it's new...  It's a (wonderful) 4.2BSD
innovation.

> but I have memories that the superblocks were printed out one at a time as if
> fflush(stdout) was called between them rather than one line at a time with
> line-buffered stdio.

I use 4.3BSD-* systems every day and have been for the past several years, and
you can take my word for it that on all 4.3BSD-* systems, including Quasijarus,
plain 4.3, and Ultrix the alternate superblock list output from mkfs/newfs
appears one line at a time on the tty.

> At some point I thought "SysV must have broke this"

I'm curious, where does SysV fit into this?  It's the wonderful 4.2BSD
filesystem a BSD-only thing that Missed'em-five people treated as a satanic
manifestation?

> since newfs would print out
> a complete row of superblock numbers at once with a big delay between the rows

That's exactly what it does.  BTW mkfs = newfs.  mkfs was/is the original UNIX
filesystem creator.  It was almost completely rewritten in 4.2BSD to create the
new filesystems.  At the same time the newfs program was written as a user-
friendly front-end to mkfs (it merely exec'ed mkfs with a bunch of options).
The situation remained in 4.3.  In 4.3-Tahoe/Quasijarus mkfs.c and newfs.c are
compiled and linked into one binary called newfs, CSRG was forced to do this in
order to support disk labels.

MS

[TUHS] question about mkfs(8) and stdio

[John Holden]

> you will see the super block backup loop does not fflush(stdio) between
> printf()s

Was there a 'setbuf(stdout, NULL)' at the beginning? This would force immediate
output from any printf's (less efficient than a fflush)