Merge remote-tracking branch 'origin/github/pr/45'

[grml.org.git] / zsh / zshtalk

--- Log opened Sat Sep 10 14:31:07 2005
16:03 @caphuso> ok...
16:03 @caphuso> so, let's start with clarifying a few things
16:04 @caphuso> zsh is a unix shell, hopefully known to anybody in here
16:04 @caphuso> from the bourne shell family
16:05 @caphuso> it has dozens of nifty features, making it an excellent shell
      for both interactive use and scripting
16:05 @caphuso> some of those features will be covered in this talk
16:05 @caphuso> which is basically all it is about
16:06 @caphuso> so... let's start with why you would want to use zsh
16:06 @caphuso> there are a few catches... you probably don't want to use zsh
      if all you do is really simple stuff in your shell, e.g. because you're a
      gnome/kde/xfce user
16:07 @caphuso> you also don't want to use zsh if you are afraid of complex
      programs
16:07 @caphuso> and zsh is _very_ complex
16:07 @caphuso> and a third reason to not use zsh is that you're working in a
      unicode environment, for which there is no support yet
16:07 @caphuso> big drawback
16:08 @caphuso> Nei: that's not using, that's pretending to use :)
16:08 @caphuso> right now development is happening to add unicode support, but
      the current state is that it's incomplete and very buggy, so you need to
      wait a few months
16:09 @caphuso> now the reasons why you would want to use zsh
16:09 @caphuso> it offers almost anything you expect from a shell, but has tons
      of improvements
16:10 @caphuso> the first of those improvements that made me go "wow! wanna
      have!" is that it has sane wordsplitting (by default)
16:10 @caphuso> that means that a variable FOO="hello bar" will, when you use
      it like $FOO not be split into two words
16:11 @caphuso> in bash etc. you almost always need to use double quotes, zsh
      makes that unnecessary
16:11 @caphuso> next: prompts
16:12 @caphuso> most shells have prompts, which are more or less customizable,
      but zsh is better :)
16:12 @caphuso> you have the regular $prompt (the one to the left), $rprompt
      (at the right side of the command line)
16:13 @caphuso> that's standard stuff though
16:13 @caphuso> you might know PROMPT_COMMAND from bash
16:13 @caphuso> zsh has two functions, precmd and preexec, which do basically
      the same, but run before a prompt is displayed and between you hitting
      enter and executing the command line
16:14 @caphuso> the latter allows you e.g. to set the xterm title to the name
      of the program you just started, very cool
16:14 @caphuso> and you can do even scarier things:
      http://apheriodic.net/phil/prompt/
16:14 @caphuso> (it's lengthy, you probably want to read it later)
16:15 @caphuso> another standard feature: aliases
16:15 @caphuso> sorry, http://aperiodic.net/phil/prompt/
16:16 @caphuso> i did say i didn't beta test this stuff :-)
16:17 @caphuso> back to aliases... zsh has sh style aliases, so they are
      limited. OTOH, you don't have to deal with the csh silliness of
      undebuggable !!!!!!!
16:17 @caphuso> but it has two major achievements over e.g. bash: alias -g and
      alias -s
16:18 @caphuso> alias -g is used to define global aliases, which means that
      those aliases are expanded everywhere on the command line, not just at
      the beginning
16:18 @caphuso> example: alias -g L='| less'; grep -r foo . L
16:19 @caphuso> the grep command will get expanded to "grep -r foo . | less",
      which saves a lot of typing
16:19 @caphuso> BUT global aliases are dangerous
16:19 @caphuso> it can happen that something gets expanded that really
      shouldn't
16:19 @caphuso> so be careful, e.g. by sticking to only very few, or by using
      "safe" names
16:20 @caphuso> alias -s defines suffix aliases
16:20 @caphuso> example: alias -s pdf=xpdf
16:21 @caphuso> if you now "execute" a single file name ending in .pdf (e.g.
      foo.pdf), the command line will get rewritten to xpdf foo.pdf
16:22 @caphuso> one limitation: won't work for stuff like "alias -s tar.gz="tar
      tzf"
16:22 @caphuso> the dot in tar.gz confuses it :-(
16:22 @caphuso> dp___: yes
16:23 @caphuso> i've mentioned that csh style aliases aren't supported
16:23 @caphuso> that's ok, because functions are a much nicer way of doing
      their work
16:23 @caphuso> functions in zsh work basically the same as in any other sh
      like shell, with only few improvements
16:24 @caphuso> one borrowed from ksh is that they can be made autoloadable
16:24 @caphuso> to reduce memory usage
16:24 @caphuso> you can define functions by simply putting their body into a
      file named like the function
16:25 @caphuso> that allows you to have a function work as a script
16:25 @caphuso> functions are pretty boring, so we get to the next feature:
      history
16:26 @caphuso> that is, remembering what commands you ran for later reuse
16:26 @caphuso> handling of that is horribly adjustable to your exact needs,
      with various options
16:27 @caphuso> zsh's history mechanism also allow you to use the csh style
      bang history expansion (stuff like !-2:t and so)
16:27 @caphuso> pretty cool if you can get used to it
16:27 @caphuso> a unique feature of zsh is that you can have multiple histories
16:28 @caphuso> fc -p/fc -a/fc -P dealwith the "history stack"
16:29 @caphuso> "fc -p" will clear out the current history and start
      remembering a new one, until you run fc -P, which will restore the old
      history
16:30 @caphuso> i use that to "bind" certain histories to specific directories
16:30 @caphuso> e.g. my ~/.seti directory, in which everything seti@home
      related is stored, has its own history, which gets automatically loaded
      when i cd into it
16:31 @caphuso> also useful for source trees in which you often do the same
      commands like "cvs update", ...
16:32 @caphuso> big topic: parameter expansion
16:32 @caphuso> that's basically just $FOO or ${FOO%bar}
16:33 @caphuso> florolf: fc -p fileinwhichthehistoryyouwanttouseisstoredin
16:33 @caphuso> it's also the most frightening part of zsh
16:34 @caphuso> and it makes people who think that perl looks like line noise
      vomit
16:35 @caphuso> i'll show a few examples later
16:35 @caphuso> next big thing is pattern matching
16:35 @caphuso> this includes globbing (*.txt etc.) and comparisions [[ $foo ==
      * ]]
16:35 @caphuso> extremely sophisticated in zsh
16:36 @caphuso> again, some interesting examples later
16:36 @caphuso> documentation is also a very good reason to use zsh
16:36 @caphuso> it is by far the best documented shell around
16:37 @caphuso> you have a huge reference manual (as man pages, texinfo, HTML,
      pdf, ...)
16:37 @caphuso> $ man zshall 2> /dev/null | wc 19882 132867 1035768
16:37 @caphuso> oops, should be two lines
16:37 @caphuso> Nei: i do!
16:38 @caphuso> this reference manual is very complete and well written, but
      terse
16:38 @caphuso> so we also got the user-friendly user guide
16:39 @caphuso> you can get it from zsh.sunsite.dk
16:39 @caphuso> if you want to get the most out of zsh, then a printout of it
      will be your best friend in the next few weeks
16:39 @caphuso> it's slightly outdated, but not much, and misses a few things
16:39 @caphuso> but still incredibly cool
16:40 @caphuso> if you got money to spend, then you might buy "from bash to z
      shell", which is a real book. i haven't read it myself, so i can't say
      how good it is
16:41 @caphuso> now completion
16:41 @caphuso> this is where zsh excells at, because really absolutely nothing
      comes even close
16:41 @caphuso> if you ever thought that bash or tcsh would do a good job, be
      surprised
16:42 @caphuso> the zsh completion system is better in any respect
16:42 @caphuso> Nei: huh?
16:42 @caphuso> it's highly configurable without forcing you to rewrite
      functions
16:43 @caphuso> Nei: indeed, fish sucks. it's not really programmable
      completion what it offers
16:44 @caphuso> it's basically all built in
16:44 @caphuso> i'll point out some impressive examples later
16:44 @caphuso> next topic: zle
16:44 @caphuso> zle is the line editor zsh uses (what readline is for bash)
16:45 @caphuso> i.e. the thing that let's you have a way to delete words, move
      around lines, copy, paste, ...
16:46 @caphuso> it's good enough that you can play tetris with it!
16:46 @caphuso> this obviously involves several features, including the ability
      to use multiple lines for your command
16:47 @caphuso> it's also extensible by simply writing shell functions
16:47 @caphuso> so you can implement everything you need
16:47 @caphuso> there will be some examples later on
16:47 @caphuso> just one more feature: dynamically loadable modules
16:47 @caphuso> (assuming your OS supports those)
16:48 @caphuso> there are a couple of modules that extend zsh's features
16:48 @caphuso> the line editor for example is such a module
16:48 @caphuso> there are also modules for ftp or tcp networking
16:49 @caphuso> a math module to do stuff like print $(( sin(5) ))
16:49 @caphuso> oh, yes, zsh does floating point math!
16:49 @caphuso> no more kludging with bc
16:50 @caphuso> ok, this first quick overview is done, from now on i'll touch a
      couple of features more detailed
16:50 @caphuso> any questions so far?
16:51 @caphuso> ok... i'll paste a small ~/.zshrc, that's just enough to get
      you started
16:51 @caphuso> zmodload zsh/complist
16:51 @caphuso> autoload compinit && compinit
16:51 @caphuso> HISTFILE=~/.zsh_history
16:51 @caphuso> SAVEHIST=1000
16:51 @caphuso> HISTSIZE=1000
16:51 @caphuso> setopt APPEND_HISTORY
16:51 @caphuso> setopt nobeep
16:51 @caphuso> prompt="%# "
16:51 @caphuso> rpompt=" %~"
16:51 @caphuso> nothing fancy in there, just setting up of the history and
      loading of the completion system
16:52 @caphuso> cow: printf
16:52 @caphuso> e1f`: thanks
16:52 @caphuso> RPROMPT=" %~"
16:53 @caphuso> ok, globbing
16:53 @caphuso> zsh's globbing makes find(1) useless for most jobs
16:54 @caphuso> e.g. you can do stuff like *(.) to match only regular files or
      *(/) to match only directories
16:55 @caphuso> Stara: one moment
16:56 @caphuso> http://84.129.253.97/~caphuso/talk
16:56 @caphuso> theclaw: because rprompt is wrong, needs to be upper case
16:57 @caphuso> theclaw: hysterical reasons?
16:57 @caphuso> dunno
16:57 @caphuso> back on topic: it can also match files larger or smaller than a
      certain size, older or newer than {a,m,c}time, files being block devices
      etc.
16:58 @caphuso> those can be combined with and, or and inverted
16:58 @caphuso> there's a way to deal with symlinks (reference or not)
16:58 @caphuso> permissions can also be matched
16:58 @caphuso> *(f:a+x,o+w:)
16:59 @caphuso> the most impressive thing is the e glob qualifier (this is what
      those things are called)
16:59 @caphuso> *(e:'file $REPLY | grep -q JPEG':)
16:59 @caphuso> this one will match all files of which file says that they are
      JPEGs
17:00 @caphuso> i had to use that one to dig through a browser cache, where all
      files were named 234346 or 456745234
17:00 @caphuso> glob qualifiers aren't the only thing zsh adds
17:01 @caphuso> it does all the regular stuff, like foo? or *bar*
17:01 @caphuso> [a-z]
17:01 @caphuso> but also things like (foo|bar|baz)
17:02 @caphuso> or foo*~*bar* to match everything that starts with foo but
      doesn't contain bar
17:02 @caphuso> it can recurse down directory trees with **
17:03 @caphuso> "print -l /**/*" will list _any_ file or directory on your
      system
17:03 @caphuso> be careful when using that one with external commands, you run
      very quickly into ARG_MAX problems
17:03 @caphuso> florolf: no, to any subdirectory
17:04 @caphuso> no matter how deep into the tree they are
17:04 @caphuso> well, compare "print /**" with "print /**/*"
17:04 @caphuso> you'll see the difference in a few minutes >:-)
17:05 @caphuso> globbing is just a special use of the more general pattern
      matching
17:05 @caphuso> the other use is in comparisons and replacements
17:06 @caphuso> in those the glob qualifiers are obviously useless (a number
      that's -eq'ed doesn't have permissions or a file size)
17:07 @caphuso> but it has a need for other things, most of which are well
      known from regular expressions
17:07 @caphuso> theclaw: (e:'file $REPLY | grep -q JPEG':) is the glob
      qualifier
17:07 @caphuso> $REPLY is a special parameter that contains the name of the
      file being matched
17:08 @caphuso> back on topic: e.g. (#s) or (#e) for what ^ and $ are in
      regexps
17:09 @caphuso> beginning of line/end of line
17:09 @caphuso> (#b) or (#m) to enable backreferences
17:09 @caphuso> (#i) to match case insensitive
17:10 @caphuso> and... *tadaa* (#a) to match approximately
17:10 @caphuso> (the latter isn't typically available in regular expressions)
17:11 @caphuso> approximate matching means that certain errors are ignored,
      e.g. "(#a1)foo*" would still match the string "ofobar"
17:12 @caphuso> there's quite a bit more on this topic, and explaining it in
      detail would take a few hours
17:12 @caphuso> (the topic == pattern matching)
17:12 @caphuso> i suggest strongly that you read the appropriate chapter in the
      user guide
17:13 @caphuso> if you want to look at some real life example, look for the
      compaudit function
17:13 @caphuso> one example of mine:
17:14 @caphuso> setopt functionargzero
17:14 @caphuso> readme liesmich () { less -- (#ia3)${~0}*(OL[1,3]) }
17:14 @caphuso> touch LiesMich README readme.txt Read_ME.doc lies-mich
17:14 @caphuso> readme; liesmich
17:14 @caphuso> first line sets the functionargzero option, which does nothing
      but setting $0 inside a function to the name of the function
17:14 @caphuso> normally it'd be /bin/zsh or something like that
17:15 @caphuso> the next line defines _two_ functions (readme and liesmich),
      that are the same
17:15 @caphuso> (liesmich is german and means readme)
17:16 @caphuso> the line noise after -- is a typical zsh glob
17:16 @caphuso> it consists of three parts ${~0}, (#ia3) and (OL[1,3])
17:17 @caphuso> the first is a simple parameter expansion of $0, but turned
      into something eligable for pattern matching
17:17 @caphuso> this means that the names of the two functions will be used to
      match files
17:17 @caphuso> the (#ia3) makes the whole thing 1) case insensitive and 2)
      allows up to 3 mistakes
17:18 @caphuso> the (OL[1,3]) at the end sorts the resulting list by size
      (biggest first) and uses only the first three files
17:18 @caphuso> the mere * in there does what it always does, it matches
      anything
17:19 @caphuso> the touch command creates 4 empty files
17:19 @caphuso> and running the two functions in the last line results in less
      displaying first README, Read_ME.doc and readme.txt, then LIESMICH and
      lies-mich
17:21 @caphuso> in lesser shells you'd need three times as much code, and
      probably several external programs
17:21 @caphuso> questions?
17:22 @caphuso> now the real horror: parameter expansion
17:22 @caphuso> things like this:
17:22 @caphuso> for a in **/*(.);
      mv $a ${a:h}/${${a:t}//(#m)((#s)?|(_| )?)/${(U)MATCH}}
17:22 @caphuso> or this:
17:22 @caphuso> function foo () {
      print -l ${(q)${${(f)"$(<$1)"}:#(#s)[[:space:]]#\#*}%%[[:space:]]##\#*
      }
17:22 @caphuso> }
17:22 @caphuso> parameter expansion in zsh can be nested
17:23 @caphuso> if you want to remove foo at the beginning and bar at the end
      of a parameter, you'd normally do tmp=${FOO%bar}; FOO=${tmp#foo}
17:24 @caphuso> in zsh you can use ${${FOO%bar}#tmp}
17:24 @caphuso> err, ${${FOO%bar}#foo}
17:24 @caphuso> and those things can grow really big
17:24 @caphuso> especially when you start using arrays or associative arrays
      (hashes in perl lingo)
17:26 @caphuso> _the_ killer example is _pids
17:26 @caphuso> it's a function of the completion system that completes PIDs
17:26 @caphuso> you need several hours with the reference manual, the user
      guide and 2l coffee
17:26 @caphuso> maybe even more
17:27 @caphuso> the sad thing is, anything i'd try to teach you about parameter
      expansion would be wasted
17:27 @caphuso> so, for the beginning, stick to the standard stuff
17:27 @caphuso> work your way up from there, while reading the user guide
17:28 @caphuso> ZHOST=localhost
17:28 @caphuso> ZPORT=1234
17:28 @caphuso> ZHOME=/tmp
17:28 @caphuso> ZDATA="this is a stupid example"
17:28 @caphuso> : ${(AA)zopt:=${(z)${(ps:=:)${(f)$(<configfile)}}}}
17:29 @caphuso> this is just one small example, showing that all this insanity
      can be used for good reasons
17:29 @caphuso> the first 4 lines are the content of a simple config file
17:29 @caphuso> and the line starting with : reads that file in and creates an
      associative array $zopt with the keys ZHOST, ZPORT, ZHOME and ZDATA
17:30 @caphuso> the values of those keys are simply the things behind the =
17:30 @caphuso> doing it the old fashioned way would be more cumbersome
17:31 @caphuso> but here we simply read in the whole file with $(<configfile)
17:31 @caphuso> split on newlines and create an array with ${(f)...}
17:31 @caphuso> split on = with ${(ps:=:)...}
17:32 @caphuso> so we have an array with the elements ZHOST, localhost, ZPORT,
      1234, ZHOME, /tmp, ...
17:32 @caphuso> and then we put that into an associative array with
      ${(AA)zopt:=...}
17:33 @caphuso> Nei: no, start reading from the inner to the outer, not from
      the right to the left
17:34 @caphuso> third big topic: ZLE
17:34 @caphuso> micikas: #emacs on the 24th
17:35 @caphuso> i already mentioned the multi line editing support
17:35 @caphuso> this means you get a real multi line buffer, just like you'd do
      in a text editor
17:35 @caphuso> it's also saved and restored in multiple lines when using the
      history
17:36 @caphuso> this is quite convenient, because your throw away shell scripts
      can grow
17:36 @caphuso> (because of the increased legibility)
17:37 @caphuso> tcsh users will know run-help
17:38 @caphuso> hitting "ESC, h" (or meta-h) will fire up the man page of
      whatever command you have on the command line, let you read it, then go
      back and continue editing where you left of
17:39 @caphuso> push-line is a more general solution for that kind of problem:
      it simply pushes the current line on the stack, type in another command
      line, execute that, then turb back to where you left of
17:39 @caphuso> bound to ctrl-q by default
17:40 @caphuso> so, but some basics first
17:40 @caphuso> the two commands with which you deal with zle are bindkey and
      zle
17:40 @caphuso> bindkey is used to handle keymaps
17:41 @caphuso> emacs users will know the concept of keymaps
17:41 @caphuso> basically just tables containing key-function pairs
17:41 @caphuso> one of those keymaps zsh uses is the emacs keymap, it contains
      the standard emacs like stuff
17:42 @caphuso> ctrl-a == beginning-of-line, "ESC, f" == forward-word
17:42 @caphuso> viins is the keymap for the vi insert mode, vicmd the one for
      the command mode
17:42 @caphuso> you can switch between emacs and vi with bindkey -e and bindkey
      -v
17:43 @caphuso> you are free to bind any key to any function, in any table
17:43 @caphuso> so you can add emacs bindings to vi, and run the vi functions
      in emacs mode
17:43 @caphuso> btw, what i named "functions" are called "editing widgets"
17:44 @caphuso> Nei: oh, yes, i've changed that
17:44  * caphuso lost track of default zsh settings years ago
17:44 @caphuso> some of those widgets are built into zle
17:45 @caphuso> some others are provided by those loadable modules mentioned
      earlier
17:45 @caphuso> the most interesting ones though are those written as shell
      functions
17:46 @caphuso> a simple example: run-with-sudo () { LBUFFER="sudo $LBUFFER" }
17:46 @caphuso> zle -N run-with-sudo
17:46 @caphuso> bindkey '^X^E' run-with-sudo
17:47 @caphuso> $LBUFFER is one of the zle special parameters, it contains the
      part of the command line left to the cursor
17:47 @caphuso> it simply gets "sudo " prepended to it
17:48 @caphuso> zle -N ... makes zle aware that run-with-sudo is from now on a
      bindable widget
17:48 @caphuso> and the bindkey command binds it to ctrl-x ctrl-e
17:48 @caphuso> now you can type at any moment C-x C-e and you'll have your
      current command line with sudo at the beginning
17:48 @caphuso> if you like me forget that way too often, you'll love this
17:49  * caphuso can't count how often he saw E: Could not open lock file
      /var/lib/apt/lists/lock - open (13 Permission denied)
17:49 @caphuso> of course, those shell functions can grow to any degree of
      "complicatedness" you like
17:50 @caphuso> there's a few interesting ones in the zsh distribution, look
      for smart-insert-last-word or the *-word-match family
17:50 @caphuso> also a whole chapter in the user guide
17:51 @caphuso> ok... now the very last topic: completion
17:51 @caphuso> probably the one most interesting to all of you, because it's
      the one you can start using without actually knowing anything ;-)
17:51 @caphuso> simply hitting your tab key in random places is sufficient
17:52 @caphuso> the completion system is loaded with "autoload compinstall &&
      compinstall"
17:52 @caphuso> after that, you're ready to go
17:53 @caphuso> until you find that some things don't behave like you want them
      to
17:53 @caphuso> that's why you get the chance of tweaking and customizing
17:53 @caphuso> in order to be able to do that, we need to learn about context
17:54 @caphuso>
      :completion:FUNCTION:COMPLETER:COMMAND-OR-MAGIC-CONTEXT:ARGUMENT:TAG
17:54 @caphuso> this is what a context looks like: 6 fields, seperated with :
17:54 @caphuso> the first field is simply "completion"
17:55 @caphuso> FUNCTION is of no interest, unless you get your hands really
      dirty
17:55 @caphuso> COMPLETER is the name of the shell function that creates
      matches at the moment of the context lookup. i'll explain those later
17:58 @caphuso> COMMAND-OR-MAGIC-CONTEXT is either the name of the command you
      are trying to complete for (e.g. in case of "make cle<TAB>" it'd be make"
      or a magic context like -redirect- (after > or <) or -math- (when you're
      at a point of arithmetic expansion)
17:58 @caphuso> ARGUMENT is also less useful, unless you already know enough to
      be bored by me talking
17:59 @caphuso> and TAGS are basically short yet descriptive names of what
      exactly you are completion at the moment
17:59 @caphuso> e.g. in "make <TAB>" the tag would be "targets", because you're
      trying to complete Makefile targets
18:00 @caphuso> "kill <TAB" has the tags "processes" and "jobs"
18:00 @caphuso> you can look them up with "ctrl-x, h"
18:01 @caphuso> so, some hands on examples: :completion:*:*:xpdf:*:*
18:01 @caphuso> this specifies: whenever we're completing for xpdf
18:01 @caphuso> :completion:*:approximate:rm:*:*
18:01 @caphuso> whenever we're completion after rm and are using the
      _approximate completer (description follows later)
18:02 @caphuso> more specific settings win over the general ones
18:02 @caphuso> :completion:*:*:*:*:processes
18:02 @caphuso> :completion:*:*:kill:*:processes
18:02 @caphuso> the first one says: whenever we complete the tag "processes"
18:03 @caphuso> the second: whenever we complete processes after kill
18:03 @caphuso> if i had two clashing settings for those two context, then
      after kill only the second one would be used
18:03 @caphuso> this is just common sense
18:03 @caphuso> the system would be useless if it worked otherwise :)
18:04 @caphuso> now how to use those contexts
18:04 @caphuso> setting and lookup of that stuff is done with the builtin
      zstyle
18:04 @caphuso> you can use that outside of the completion system to store and
      retrieve data too, some of the zle functions do that for example
18:05 @caphuso> we'll only deal with setting, not with retrieving
18:05 @caphuso> zstyle ':completion:*:*:(^rm):*:*' ignored-pattern '*.(o|~)'
      '#*#(D)'
18:06 @caphuso> he're we set the style 'ignored-pattern' (you can find a list
      and descriptions of styles in the manual) to '*.(o|~)' '#*#(D)'
18:06 @caphuso> when completing after anything except rm
18:06 @caphuso> ^rm means "don't match rm"
18:07 @caphuso> the two globs match *.o files (C programmers will know), *~
      (backup files) and #*# files (emacs autosave files). the (D) qualifier
      turns on globbing of .* files
18:08 @caphuso> ignored-pattern means: when completion, then ignore everything
      that matches the value of this style
18:08 @caphuso> so, after any command that's not rm, we won't get those files
      offered as completions
18:09 @caphuso> this is just one simple example of how extensible the zsh
      completion system is configurable
18:09 @caphuso> there are a few dozen styles
18:10 @caphuso> you'll need to read the manual, i really don't wanna talk till
      tomorrow morning :-)
18:11 @caphuso> now the completers i wanted to explain quickly
18:11 @caphuso> a completer is simply a shell function that looks up the
      current context and creates a list of possible completions, offering
      those to the user
18:11 @caphuso> of course, nothing is that simple in zsh
18:11 @caphuso> the first completer available is "_complete"
18:12 @caphuso> it does basically what bash does
18:12 @caphuso> nothing fancy at all
18:12 @caphuso> just looking up stuff, then straight completion mozi<tab> ->
      mozilla
18:13 @caphuso> _expand is the first exception, because it doesn't really
      generate possible completions
18:13 @caphuso> it expands certain things on the command line, e.g. $PARAMETERS
      or $(print -c *)
18:13 @caphuso> $PATH<TAB> would become
      /home/caphuso/bin:/usr/local/bin:/usr/X11R6/bin:/usr/sbin:/sbin:/usr/game
      s:/usr/lib/surfraw:/usr/bin:/bin:/usr/bin/X11 on the command line
18:14 @caphuso> _correct does what the name implies: it corrects typos
18:14 @caphuso> _approximate is _correct on crack
18:14 @caphuso> it not only corrects typoes, it also generates completions
18:15 @caphuso> given a directory ~/public_html, then _correct would do the
      right thing for "cd ~/pbulic_html<TAB>", but not for "cd ~/pb<TAB>"
18:16 @caphuso> there are a few more, but those are the most interesting ones
18:17 @caphuso> using completers is a matter of setting the style "completer"
      properly
18:17 @caphuso> zstyle ':completion:*' completer _complete
18:18 @caphuso> (bash emulation again ;-)
18:18 @caphuso> zstyle ':completion:*' completer _complete _approximate
18:18 @caphuso> this means: first try _complete, if it can't generate any
      possible matches, then try _approximate
18:19 @caphuso> zstyle ':completion:*' completer _complete _correct
      _complete:-extended _complete:-substring
18:19 @caphuso> this means: first _complete, if it fails _correct, then
      _complete again, then _complete again
18:20 @caphuso> the trick is that by appending :-foo the completer will look up
      the context not under its regular name (_complete) but as (_complete-foo)
18:21 @caphuso> this allows to do the same completion with different settings
      because the context differ
18:21 @caphuso> zstyle ':completion:*:complete-extended:*' matcher-list 'm:{a-z}={A-Z}' 'r:|[+._-]=*'
18:21 @caphuso> zstyle ':completion:*:complete-substring:*' matcher-list 'm:{a-z}={A-Z} l:|=**'
18:22 @caphuso> we do that here... as COMPLETER field we have now not complete
      but complete-extended and complete-substring, and for both we set the
      style matcher-list to yet another kind of line noise
18:22 @caphuso> those matchers configure how zsh is doing comparison
18:22 @caphuso> s
18:22 @caphuso> the default way is to simply compare the string on the command
      line with all possible completions
18:23 @caphuso> foo<tab> with the possible completions of foobar, foobaz and
      moo_foo will offer the first two as possible completions
18:23 @caphuso> because only they match "foo*"
18:24 @caphuso> matching control can change that, make it ignore certain things
18:24 @caphuso> e.g. m:{a-z}={A-Z} tells it that lower case characters match
      upper case characters
18:24 @caphuso> echo $disp<TAB -> echo $DISPLAY
18:27 @caphuso> r:|[+._-]=* makes the characters +, ., _ and - special, so that
      "cd p_h<TAB> -> cd public_html" or "rm x.txt<TAB> -> rm xenomorph.txt"
      works
18:27 @caphuso> and the l:|=** does a simple substring match, so that
      "foo<TAB>" will also match barfoobaz
18:28 @caphuso> now the whole list of completers in this last example explained
      slowly
18:29 @caphuso> when the user hits TAB, the _complete is tried. if it fails
      to generate any matches, the word on the command line is tried to be
      corrected. if this doesn't help, case-insensitive matching with a few
      special characters is tried, and if that fails too, substring matching
18:29 @caphuso> and as soon as one completer succeeds with creating at least
      one match, the whole thing stops
18:30 @caphuso> now you might wonder why to do that so complicated
18:30 @caphuso> easy: 1) the more possible matches you try to generate, the
      slower the whole thing becomes
18:31 @caphuso> _complete is noticably faster than _complete:-extended, and
      i've seen _approximate run for several minutes, due to a large directory
18:31 @caphuso> so you want to be careful
18:31 @caphuso> 2) the more matches you create, the longer the list of possible
      completion becomes
18:32 @caphuso> when _complete might only show 5 matches, _approximate can go
      up to 50
18:32 @caphuso> or more
18:33 @caphuso> this doesn't only take longer, it also makes it hard to handle
18:33 @caphuso> so, that's the reasoning behind
18:33 @caphuso> ok
18:34 @caphuso> AFAICT, that's it
18:34 @caphuso> i hope it's been interesting enough, and would glad to see a
      few more enthusiastic zsh users in #zsh in the next few weeks
18:34 @caphuso> any questions?
--- Log closed Sat Sep 10 18:35:39 2005