JdeBP

@gumnos@mastodon.bsd.cafe

Here's part of the bug.

https://github.com/freebsd/freebsd-src/blob/main/usr.bin/grep/file.c#L210

There's no handling of EOF if it occurs as the very first read() of the file.

A zero-length regular file and /dev/null will also result in two read()s returning 0 before grep then exits.

This code was introduced in commit 3ed1008b891eb34df96a6ae4f87e5a3270be759c in 2010.

#OpenBSD's version of this code is older and uses the normal C stdio library routines and getline(3) rather than rolling its own layer over read().

#FreeBSD #grep

@gumnos@mastodon.bsd.cafe

But did you also do a

type grep

to ensure that you were running /usr/bin/grep and not something from ports, nor some 'helpful' alias/function that someone has snuck in to /etc/profile and /etc/csh.cshrc?

It wouldn't be the first time in history for the latter. Best to eliminate it straightaway so that the blame can be squarely hung upon /usr/bin/grep.

#grep #FreeBSD

@gumnos@mastodon.bsd.cafe

Not reproducible, as expected, on my older #FreeBSD here.

Make sure to double check for embarrassing shell scripts, aliases, directories in $PATH, and shell functions before writing that bug report. I've seen that so many times over the years. (-:

#grep

@ianthetechie@fosstodon.org

I was about to write the same thing.

And as @steeph@todon.eu pointed out, getopt in shell script does not allow one to do several things, including take advantage of the Z shell's ability to automatically construct command-line completions for anything that supports --help .

@rl_dane@polymaths.social
#getopt #UnixShells #zsh

@shlinkio@fosstodon.org

I have this vague memory that when I first set up my github repositories, one could not turn off pull requests. I turned off almost everything that could be turned off.

I've just turned them off on all of my repositories.

I wasted too much of my life combatting robot-generated bulk e-mail. I'm going to nip robot-generated bulk pull requests in the bud before they become a problem.

@ska@social.treehouse.systems
#GitHub #AIslop

@cks@mastodon.social

The opposite is fun, too: when there's a system #DesktopBus broker running but it's not a systemd system, and logging in through a chain of intertwined mechanisms tries to auto-start #systemd.

@JdeBP@tty0.social

#PAM #login

@cks@mastodon.social

It's not even a hot take. It's actual history.

STEVIE came from the days when people were re-inventing Joy vi for other platforms and systems with (gasp!) arrow keys and console-paradigm I/O.

It was less than a decade until people were thinking that Joy vi could be improved and were actively trying to make things that were better.

Watcom vi, for another example, came out in the early 1990s and that had windows, and uses for function keys.

#vi #STEVIE #vim #OpenWatcom

@feld@friedcheese.us @lattera@bsd.network @drwho@masto.hackers.town

Digging into the talk page of the article on the FAT filesystem and Linux, one finds that, back in 2009, #Wikipedia people thought the same thing. (-:

https://en.wikipedia.org/wiki/Wikipedia:Articles_for_deletion/UMSDOS

It was past tense even then. Linux 2.6 had come out some 6 years beforehand.

Amusingly, the Linux Documentation Project has a UMSDOS how-to written in the present tense that has not been updated since 2001.

#FAT #Linux #UMSDOS #LinuxDocumentationProject

@feld@friedcheese.us @lattera@bsd.network @drwho@masto.hackers.town

Indeed so. Linux had a filesystem type named umsdos that layered a raft of non-native semantics on top of the FAT on-disc data structures with the use of magic '--LINUX.---' files.

I think that we can safely say in the 2020s that, across the world of Windows NT, BSDs, Linux, et al., use of the FAT filesystem format has been relegated to edge cases like some boot volumes (e.g. the EFISP) and some data interchange, and isn't in general use for system volumes.

#FreeBSD

@cks@mastodon.social

I suspect that this has changed over the years. Kay Sievers's original proposal in 2010 indeed had the active device as the last one. But Lennart Poettering writing in 2012, after things had settled down amongst the kernel developers, wrote that the active device was the first one. I vaguely remember inspecting the code that constructed the string to check, at the time. I'll try to remember to have another look.

https://0pointer.de/blog/projects/serial-console.html

#Linux #console #vt #manual #nosh

@cks@mastodon.social

I wrote a console(4) manual page some years ago, to fill in a gap.

http://jdebp.info/Softwares/nosh/guide/commands/freebsd-console.xml

#FreeBSD #console #manual #nosh

@argv_minus_one@mastodon.sdf.org

People have brought up the TOPS-20 command processor already.

Just for completeness, I'll mention DCL on VMS, which had long parameter names (e.g. /COPIES to PRINT and /FULL to DIRECTORY).

There's also MCR on VMS (and RSX), with HELLO and BYE for logging on and off.

Don't forget the whole "* commands" system on the BBC MIcro, either.

#VMS #CLI #TOPS20 #DCL #DIGITAL #retrocomputing #MCR #RSX #BBCMicro

@ska@social.treehouse.systems

My schools did not teach C or any language like it.

I learned about -- and #getopt from a combination of FidoNet, Usenet, and a 1985 book on Unix by Eric Foxley.

The habit of using -- probably cemented when I wrote a load of command-line utilities for DOS and OS/2 in the 1990s which all supported an end-of-options marker in their command-tail parsing library.

https://groups.google.com/g/comp.os.os2.utilities/c/415LDKL3_Lw/m/P9tPDLd7VAUJ

I made use of what my own tools could do. (-:

@cazabon@mindly.social
#os2clu #dosclu

@jas@fosstodon.org

My first suspect for a #login not supporting -- would be something with a 1980s history pre-dating standard #getopt, such as Solaris, which is ironic given that #inetutils has its only -- present in conditionally compiled code targetting Solaris.

#FreeBSD, #NetBSD, and #OpenBSD login all use getopt(), pervasive in these worlds for decades, as do the util-linux login (used by Debian et al.), and the #Illumos and #BusyBox logins.

#suckless login supports -- via ARGBEGIN.

@ska@social.treehouse.systems @cazabon@mindly.social

@ska@social.treehouse.systems

Looking at the commit and the code as it still stands today, it is interesting that only on Solaris does it even try to use -- in the arguments to login to signal the end of options, and even then only in limited circumstances.

#FreeBSD telnetd, for comparison, always puts -- in before the supplied account name.

I wonder how long it will be before the lesson is properly learned.

@jas@fosstodon.org
#getopt #login #telnetd #inetutils

@feld@friedcheese.us

If you look around you should still be able to find Luke Mewburn's paper on #MewburnRC .

Similarly, what Linux people erroneously call "sysvinit" is actually van Smoorenburg init+rc created (by Miquel van Smoorenburg) for Minix in 1992.

It's not from #Unix System 5 at all, as that had obsoleted R3's init+rc system years before 1992 and gone on to the AIX SRC, and the SAF, SAC, ttymon, SMF, and stuff still in #Illumos today.

http://jdebp.info/FGA/unix-service-access-facility.html

@amin@polymaths.social @fbievan@polymaths.social @rl_dane@polymaths.social

#rc #init

@rl_dane@polymaths.social @fbievan@polymaths.social @amin@polymaths.social

Don't use rc.local. It has been obsolete since 1983. #FreeBSD has actually discontinued it twice in its history.

FreeBSD has had Mewburn rc for most of this century. It's what #NetBSD uses and #OpenBSD has a similar but NIH system. At minimum, use Mewburn rc scripts.

http://jdebp.info/FGA/rc.local-is-history.html

The rc(8) manual page has an example script. There's tutorial doco as well.

https://docs.freebsd.org/en/articles/rc-scripting/

#rc #MewburnRC

@alexanderdyas@mindly.social

tr -d works on more systems than tr --delete will.

https://illumos.org/man/1/tr

https://man.netbsd.org/tr.1

You can also use that function with the Korn and Watanabe shells.

#tr

In light of recent reports I thought about adding AF_VSOCK support (if that is meaningful) to the #nosh toolset's #ifconfig .

It turns out that there's no vsock device on my Debian development machine.

I do not have the time at the moment to look into this more thoroughly.

@lobsters@mastodon.social

For what it's worth:

The hyperlinked article is purely about Linux and its two major C libraries (GNU and #musl), and does not cover #Unix, historical or otherwise, at all.

There are systems still around where grantpt and unlockpt are important. On #NetBSD with certain options set they are not actually no-ops. They are definitely not no-ops on #Illumos-based operating systems, where (for starters) STREAMS stuff happens on pseudo-terminal front-end devices.

@rqm@exquisite.social

@rl_dane@polymaths.social hit the nail on the head.

This isn't really specific to completion. It's just general overlong input line editing behaviour.

The PD #KornShell and its derivatives (e.g. the #MirBSD Korn Shell) only have line editing with a single line that sideways scrolls.

ksh93, however, has a
set -o multiline
option for switching to a multiple-line line editing mode. (The Z and Bourne Again shells have similar.)

You might like the Watanabe shell. It's in ports.

@magicant@pawoo.net
#UnixShells

@rl_dane@polymaths.social

I was curious as to what this meant, so I fired up PD ksh and tried what I thought long command-prompt entries meant.

The PD #KornShell uses a column width greater than the terminal width (which I set to 50 columns here, just to make things easier) because it SPC-pads everything to the length of the longest string; and ends up double-spacing most rows in the table as a consequence.

I couldn't figure out how to get it to specifically clip rows, though.

@rqm@exquisite.social
#UnixShells