KornShell – Moving forward | Hacker News Reader

KornShell – Moving forward

situ.im

58 pointssiteshwar31 comments

I think ksh was the first shell I was exposed to vi command-line editing in on a Sun workstation a long time ago -- I haven't thought about it in years as it seems bash is so prevalent, but good to so some folks still keeping it alive.

You'll still find it as the default shell on at least some BSDs, and I think AIX.

On OpenBSD, the default shell is their fork of pdksh. On NetBSD, it is Almquist Shell, and on FreeBSD, it is Tcsh (IIRC).

On FreeBSD it's tcsh for root and ash for non-root, oddly enough.

Interesting. Thanks.

AIX defaults to ksh88. My first job out of college a long time ago was to maintain and enhance AIX's various installers, which involves some 100KLOC (!!) of ksh scripts. I was one of the first people on the team to move newer projects to Perl- this was early 00's and somewhat fashionable back then, and the only other scripting language supported on AIX at the time. In retrospect, it's possible everything might have been better off left as ksh :-)

The AIX installp format is my favorite package manager. It was fairly easy to debug it until it hit the ODM. The only time when AIX utility debugging got weird was when multibos went off the rails.

I started using ksh on AIX and it has been my favorite shell since then. I've written a few thousand SLOC worth of utilities and customization scripts in KornShell.

Commercial UNIXen that bundled CDE had two ksh versions - ksh88 installed as the POSIX shell, and ksh93 installed as the CDE graphical shell as /usr/dt/bin/dtksh.

The dtksh binary bundled Motif X/Windows support, and scripts could easily and quickly build complex GUI applications. This actually remains the fastest and least resource intensive way to build a quick GUI.

https://sourceforge.net/projects/cdesktopenv/

The dtksh was written at Novell by Pendergrast, who wrote a textbook on it.

Modern ksh93 should make this a supported compile option.

I've never used it, but I'm deeply grateful for it because, from what I can tell, without Ksh we wouldn't have Zsh, which I use every day and really enjoy.

If you like zsh, you're going to love fish†

† The qualifying clause is probably unnecessary here.

Maybe even if not! I didn't care that much for zsh but I love fish.

I love zsh, but am not a huge fan of fish. Lots of non-standard things that don't play nicely with existing workflows.

I prefer bash/zsh/etc because of the ecosystem around them. If I wanted to throw that away and start from a different base I'd probably use Powershell or something like it to be honest.

I love zsh, but abandoned ship of fish long ago.

I find fish interesting, because I thought it is what I wanted (throwing away the pain of sh syntax), but now I just have to keep switching between sh and fish and I keep forgetting which is which. I have to write too many scripts for other people, and I can at best assume they have bash.

I've gone back to bash, might try out zsh.

Why not just write scripts in bash (or, hell, Python?) and use fish as your shell?

Just because now it's two languages for loops to remember, I kept trying to write fish loops in bash scripts, and bash loops in fish shells.

Some people are better at this kind of thing than me, but I find it hard when languages are close in notation (I never confuse C++ and python, but I confuse python/ruby, or bash/fish).

Zsh is a bit of a bear to configure -- the documentation is relatively chatty and verbose, and there is an enormous variety of options and configuration points available.

IMO its real benefit is in its scripting language, which is mostly a Bash superset with really nice quality-of-life features, namely a vast set of parameter expansion flags, super-advanced globbing options, native support for numerical-indexed arrays, and... it doesn't split arrays on whitespace unless you explicitly ask it to! You can freely iterate over filenames with whitespace, special characters, etc., and you technically don't need to quote any of it. A good Zsh programmer will be able to almost totally grep, awk, bc, etc., and all the overhead of spawning a subshell.

My first shell was ksh on a DEC Unix many years ago. That was the first scripting I ever did.

OpenBSD still uses a ksh fork as a default shell. mksh is also widely-available and pretty great -- I use it as my shell on Linux boxes, too.

its not a fork of ksh, its a fork of an open source ksh clone called pdksh that predates AT&T ksh open sourcing, and is frankly kinda crap. OpenBSD has improved it, but compared to real ksh its lacking quite a lot, most painfully sane substrings.

Right. In the spur of the moment I wrote "a fork of ksh" as in "a shell (mostly) conforming to ksh' specs", which is technically incorrect -- the worst kind of incorrect.

ksh was not my first shell, but as much as you, it is where I use vi command-line editing a lot. It's one of the first shells I really got to grip with as a young budding system administrator. I didn't know any vim really before I went on a Sun scripting for sysadmin course. My Ksh scripting book has served me well over the many years.

At work, in the scope of a migration from RHEL 6 to rhEL 7, I have spend a lot of time migrating scripts from ksh (pdksh) to bash in order to remove this dependency. I have not seen any advantage of ksh over bash. There are many differences (ksh is less strict), but IMO there is no point in keeping 2 mostly similar but incompatible shells when less and less people are using shell. I think ksh should die. Maybe I am missing something ?

The biggest benefit of real ksh, to me, is that variables don't disappear in weird cases, such as:

    cat file |\
    while read stuff
       do other stuff; set variable
    done

In ksh93, variables set within the while loop stay around outside of it. Whereas bash will run the while loop in a forked sub process.

That just sounds like (practically speaking, if not totally accurately) local scoping.

There are TONS of advantages to ksh93 over bash.

-Floating point math, not just integer

-Coprocess syntax that is correct and compliant

-Genuine Korn retains variables in control structures

-Lots more, detailed in the Korn & Bolsky ksh93 book

The language specification for ksh93 is large and powerful, and bash doesn't approach it.

There's basically nothing to see here. One or two minor bug fixes, a few cleanups, and a new build system replacing one that already worked fine.

In my experience replacing the build system and doing nothing else is usually a good indicator for a project that will be soon abandoned. If that's the first priority it means there's no real drive to implement anything new.

> a new build system replacing one that already worked fine.

Have you actually tried the legacy build system before making that comment ? If not, I would suggest you to try that. It's by far the worst build system I ever dealt with. Just to give you an example, if there's a compilation failure in the middle of build process, build system will not stop, it will continue to compile everything that it should not compile and then exit with status 0. There was an absolute need to replace it with something newer.

Replacing the build system is just the first step towards improving the code base and it's not the end of it.

yes i built ast quite a few times. nmake always worked fine.

I was doing lot of refactoring around ksh93 code and legacy build system was not really helpful in detecting problems. It was best to change it.

I hadn't been aware of this project, but I am thankful it exists as a mksh user. ksh rocks; I love how low-bloat and fast it is. Feature bloat — e.g., distribution package management helper scripts that come preinstalled — slowed down bash. Enough was enough.

To the remark about build system, I would like to offer some speculation: I have (or had?) worked with its maintainer Kurtis by way of common employer on a few projects. He is a fantastic engineer and extremely careful and diligent; from that, I would grant significant deference. His team maintained one of the most sprawling pieces of legacy shell script infrastructure I have seen in my career.

Here's to hoping for a blossoming revival of the original essentials.

Reply on news.ycombinator.com