Hi,
Siteshwar Vashisht wrote on Sun, Jan 22, 2012 at 07:29:23PM +0530:
> I am working on a shell, and the idea is to show suggestions for
> command line options on pressing tab. So let's say on pressing
> "ls<TAB>", I will get all the command line options for ls
> command. Now a good idea is to fetch these options from manual
> pages.
OUCH.
No, please don't.
This is a misguided, terrible idea in so many ways.
1) Do not bloat the shell.
If there are two programs in userland that you don't want to
be large, complicated, fragile, and slow, it's init(8) and
the shell. The shell is a critical component of the system,
it is not the right place for bells and whistles. And you
don't want yet more non-standards-compliant features in the
shell. Shells diverge too much from each other and have too
much bloat already.
2) Do not bloat *any* utility program with unrelated bells
and whistles. That's just not the Unix way.
Write small, focussed tools that make life easy when used
in concert.
For finding command line options, there is man(1).
Most programs have a one-line usage() message, too.
There is no need to do the same thing again, in a third place.
3) Options typically are a dash and one character.
If any frequently needed option is called
--hey-look-i-invented-a-fantastically-long-name,
that's a bug in the program; fix it, call it -l
if it's really needed for every second invocation.
If you need it twice a year, fine, type it in.
No problem to solve here, move on.
4) Whatever you do, it will be crude guesswork.
Different systems document options in different ways.
Heck, different *programs* document options in different ways.
There are different languages for producing documentation,
both the source and the output look different.
Some systems install manual source code only, some
install preformatted pages only, some install them gzip'ed,
some don't, some cache them, some don't, some install them
in this directory, some in another, some use man.conf(5),
some use manpath(1), i could go on and on...
Even if documentation authors try to follow some usual
way, many get it wrong and the results are hard to parse;
after all, most documentation authors are programmers in
the first place and don't know or care that much about the
finer points of formatting documentation. In any case,
you don't want crude guesswork in the shell.
Or even if you accept that, it will maybe work half of
the time on your own computer, but almost nowhere else.
5) How are you going to find out whether you are even looking
at the right manual? There may be different versions of
ls(1): The native one, the UCB one, the GNU one, and then
there may be a POSIX manual which doesn't correspond to any
actual implementation. Different packages are installed
everywhere, and the file system layout is different
everywhere, as well.
6) You are trying to do semantic analysis with groff(7).
Bad idea, don't do that, that's not at all what Roff
is about, and Roff is doing a terrible job in that area -
simply because it's beside the point, it was designed to
produce high-quality typesetting output.
If you want to do semantic analysis, use tools written
for semantic analysis, like the mandoc(3) libraries
written by Kristaps Dzonsons. But *PLEASE*, don't link
those into a shell...
7) Even though the mdoc(7) language, allowing concise
semantic annotations, was invented more than 20 years
ago (and already used for the Net/1 4.4BSD release),
most manuals still use the old-fashioned man(7) format
nowadays - which doesn't even provide specific markup
for command line options! So, what are you even going
to parse for? You have almost nothing to work on in the
first place, not even mandoc(3) will be able to help you
much with man(7) input, even if you have the source code...
8) ...
I'm sure i could go on like this.
But why should i?
Why kill a dead horse again and again to get it even more dead?
This is utterly hopeless.
Just drop the whole project.
Yours,
Ingo
