Hacker News Reader

Comment by chasil | Hacker News Reader

On the one hand, I have no idea why NT chose CMD.EXE as their command shell instead of pdksh; most would ascribe this to the influence of OS/2. NT was built with a POSIX interface, and Korn would have been a much better fit; one objection might be that NT forks processes very slowly, which is a major aspect of Korn.

On the other hand, I don't believe Dave Cutler would have ever allowed /usr/src to ship with the source code to the kernel.

View on HN

Ill take a stab at that. command.com scripts were very similar to cmd.exe scripts. That was important to get people to actually use the thing. You can see the same thing today with powershell. PS is in many ways better. Except it does not run the old scripts very well. Even something simple like 'dir /s' does not work in powershell you have to do it the powershell way. If they had moved to sh style command eventually they would have been fine. But... their existing base would have been very unhappy with the results in the short term.

At the same time, Xenix had a huge influence on Microsoft.

I can't believe that anyone with taste would want CMD.EXE, and the criticism is deep.

https://blog.nullspace.io/batch.html

Oh no doubt it is a rubish language to program in. The thing is their existing base had a set of scripts that already worked. These are not 5 line scripts either. One place I worked at the thing was probably 20k lines of batch script. To hand them a new shell is akin to saying 'throw everything you did out and redo it'. Those scripts were probably home grown by people who had other things to do (you work with what you have). Or by expensive consultants. They 'must work' or 'why am I upgrading to this?' It is not necessarily hard to do. It just an extra expense most businesses are going to skip if they can.

Some reasons why CMD.EXE is so bad:

(1) COMMAND.COM cut a lot of corners, in significant part due to being written in assembly and designed to run on machines with very low memory (DOS 1.0 would run on machines with only 32KB of RAM)

(2) Backward compatibility meant a lot of the questionable decisions in (1) had to be maintained

(3) At some point Microsoft stopped investing in it, and didn't do anything further to add new features or file off more rough edges

I think (1) and (2) were kind of inevitable, but (3) is a decision Microsoft has made much more recently

I wish they'd open-source CMD.EXE and start accepting PRs for it – I think some of the worst things about it (e.g. the "Terminate batch job (Y/N)?" prompt which is impossible to turn off without patching the executable) would get fixed. But I doubt they'll do that because they just want people to focus on PowerShell instead–and before PowerShell, it was the same thing for VBScript

At this point ConHost.exe is open source [0] so it is maybe not a stretch to expect Microsoft to open source CMD.EXE at some point.

Though with PowerShell being cross-platform and already open source, I personally don't think there's enough to gain in some sort of better open source CMD.EXE fork. I'd be interested in being proved wrong on that, but I'm also happy enough with PowerShell these days I'm not in a hurry to return to CMD.EXE.

[0] https://github.com/microsoft/terminal/tree/main/src/host

PowerShell is rather heavy and takes a lot longer to start than CMD.EXE does

I think there are plenty of little annoyances in CMD.EXE which could be fixed. That prompt I mentioned is the most obvious one, but I'm sure there are more.

Ideally, a command processor would be composed of an LR-parsed language that could easily be represented as a grammar for yacc. Neither Korn/pdksh nor CMD.EXE can meet such a requirement.

DOS batch compatibility could easily be met with a secondary command shell.

NT's primary command shell could have reworked AWK's grammar into an interactive shell.

It's a shame that the NT command shell did not recieve the attention that it deserved. DCL would have been preferable (if you like FORTRAN).

> Ideally, a command processor would be composed of an LR-parsed language that could easily be represented as a grammar for yacc

There are plenty of parser generators other than yacc, many of which accept different types of grammars from LR (e.g. LL parsers, PEG parsers). I'm not sure why it should matter whether a command processor's grammar is LR or LL or whatever.

The POSIX shell (and its precursors and derivatives) are not LR-parsed languages, and they suffer from ambiguities and inconsistencies. The extent of the distrust is clear in the video and slides below.

I don't know as much about the descent of the COMMAND.COM of MS-DOS with the influence of OS/2 into NT CMD.EXE, but I can say with certainty that it was no better.

We needed something clean and maintainable, and we didn't get it.

https://archive.fosdem.org/2018/schedule/event/code_parsing_...

The problem the slides are raising is that the POSIX shell's grammar is ad hoc: despite the POSIX standard containing an LR yacc grammar, it is annotated with English-language rules which violate the LR formalism.

But not being LR-parsed in particular seems like a red herring – if it had been designed to be cleanly LL-parsed, I don't think anyone would be complaining.

> The problem the slides are raising is that the POSIX shell's grammar is ad hoc...

...and that's the point. A shell language with no inconsistencies or ambiguities is what we needed, and we didn't get it.

A language with a formal yacc grammar would have helped us get it. Neither the POSIX shell nor CMD.EXE are ever going to meet such a requirement.

I don't know if DCL was clean enough to have a yacc grammar. If it was, then it would have been a better choice.

I don't know why you are defending such a precept when the weaknesses are so very well known.

EDIT: A yacc grammar might be preferred as yacc itself is part of POSIX.2, assuring many implementations.

You can find lex and yacc here:

https://pubs.opengroup.org/onlinepubs/9699919799/utilities/

However, had any formal grammar been used, it would have been far better than what we have now.

There are languages with perfectly clean grammars which can't be parsed by yacc because they aren't LALR(1).

I agree that a command processor should have a grammar that can be expressed in a well-known formalism, and its parser generated by a parser generator.

I agree that both POSIX shell and CMD.EXE are flawed because that isn't true.

What I'm disagreeing with, is that it is important that the grammar formalism be LALR(1) in particular, and that the parser generator be yacc in particular.

Suppose I have a Packrat parser generator. [0] And my command processor has a nice clean PEG grammar. [1] And I use the Packrat parser generator to generate the parser of my command processor. That grammar quite possibly isn't LALR(1), and hence yacc in particular won't be able to generate a parser for it. But what's the problem with that? If it is a problem at all, it is a very different problem than the problem that CMD.EXE and POSIX shell have

[0] e.g. https://github.com/arithy/packcc

[1] https://en.wikipedia.org/wiki/Parsing_expression_grammar

> I have no idea why NT chose CMD.EXE as their command shell

Because the OS is a sequel and replacement to DOS, and so it has a DOS-like shell.

Speaking as a former professional VMS, DOS, Windows and NT sysadmin:

I preferred CMD to any Unix shell, and I still do. I had boxes to run. I am not a programmer and I don't want programmers' tools.

It is much more use to me to be able to do:

move *.00? oldlogs

And have it pick all files matching the wildcard and move them to the folder -- without needing to be told it's a folder -- than it is for me to be able to write shell scripts.

I think you'll need to provide a better example, since the equivalent in most unix shells is...

  mv *.00? oldlogs

Just sayin'. =)

I see your =) and raise you this:

    ren *.htm *.html

Reply on news.ycombinator.com