About insert-last-word and "command args | less"

About insert-last-word and "command args | less"

[Vincent Lefevre]

Hi,

I often do something like

  svn log file | m

(where m is an alias for less). But when I use insert-last-word, I get
the "m" instead of "file". Is there a way to ignore the one-character
words?

Alternatively, is there a way to make

  svn log file |

behave a bit like "< file", i.e. using READNULLCMD instead of getting
another prompt (and use "... | \" to get the prompt)?

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

Re: About insert-last-word and "command args | less"

[Bart Schaefer]

On Oct 14,  1:56pm, Vincent Lefevre wrote:
} 
} I often do something like
} 
}   svn log file | m
} 
} (where m is an alias for less). But when I use insert-last-word, I get
} the "m" instead of "file". Is there a way to ignore the one-character
} words?

Use smart-insert-last-word.

autoload -U smart-insert-last-word
zle -N insert-last-word smart-insert-last-word
zstyle :insert-last-word match '*[[:alnum:]]*[[:alnum:]]*'

That example says that a word must contain at least two alphanumeric
characters, so neither 'm' nor '|' qualifies and "file" is inserted.

Re: About insert-last-word and "command args | less"

[Vincent Lefevre]

On 2006-10-14 14:27:41 -0700, Bart Schaefer wrote:
> Use smart-insert-last-word.
> 
> autoload -U smart-insert-last-word
> zle -N insert-last-word smart-insert-last-word
> zstyle :insert-last-word match '*[[:alnum:]]*[[:alnum:]]*'
> 
> That example says that a word must contain at least two alphanumeric
> characters, so neither 'm' nor '|' qualifies and "file" is inserted.

Thanks, but this is buggy with setopt HIST_IGNORE_DUPS:

prunille:~> zsh -f
prunille% setopt HIST_IGNORE_DUPS
prunille% autoload -U smart-insert-last-word
prunille% zle -N insert-last-word smart-insert-last-word
prunille% zstyle :insert-last-word match '*[[:alnum:]]*[[:alnum:]]*'
prunille% true
prunille% :
prunille% echo | :
prunille% true
prunille% : true
prunille%

On this new line, I start by typing ^]_ and I get "true" as expected,
but on the second ^]_ I get "echoue" with the cursor over the "u".

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

Re: About insert-last-word and "command args | less"

[Bart Schaefer]

On Oct 15,  2:50am, Vincent Lefevre wrote:
}
} Thanks, but this is buggy with setopt HIST_IGNORE_DUPS:

Hrm.  It's not *just* HIST_IGNORE_DUPS, because it works fine as long
as you're not inserting the word in command position.  In fact, it
works fine as long as $BUFFER is not empty when you begin using it.

This implies that the bug is in one of up-history or down-history.
smart-insert-last-word sets NUMERIC and assumes that thereafter a
call to .up-history followed by one to .down-history will return the
editor to the current line.  This appears not to work when $BUFFER
starts out empty and there are duplicate history items to skip.

The simple workaround is to change

  setopt extendedglob

to

  setopt extendedglob nohistignoredups

in smart-insert-last-word, on the second line.

Re: About insert-last-word and "command args | less"

[Vincent Lefevre]

On 2006-10-14 21:04:51 -0700, Bart Schaefer wrote:
> The simple workaround is to change
> 
>   setopt extendedglob
> 
> to
> 
>   setopt extendedglob nohistignoredups
> 
> in smart-insert-last-word, on the second line.

OK. But is it possible
  * to skip a history line when all its words match the pattern?
  * to always get a word that wasn't proposed before (when one uses
    ^]_ several times) by skipping history lines when needed?

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

smart-insert-last-word bugs

[Vincent Lefevre]

On 2006-10-14 21:04:51 -0700, Bart Schaefer wrote:
> The simple workaround is to change
> 
>   setopt extendedglob
> 
> to
> 
>   setopt extendedglob nohistignoredups
> 
> in smart-insert-last-word, on the second line.

In addition to this bug, here's another one: after I type a command
and use completion, smart-insert-last-word removes the space between
the command and the argument.

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

Re: About insert-last-word and "command args | less"

[Bart Schaefer]

On Oct 15, 12:39pm, Vincent Lefevre wrote:
}
} OK. But is it possible
}   * to skip a history line when all its words match the pattern?

This is difficult because of handling a numeric argument, e.g. when
called as "zle smart-insert-last-word 7" [%] it's supposed to back up 
exactly 7 commands and take the last word from that one.  The pattern
is not matched against the words until after the history line to
examine is chosen.

I believe the patch below does this the way you want ... however, it
makes a liar of the documentation, which implies that line-selection
takes precedence over interesting-word-selection; so it's not suitable
to be committed without some more work.

[%] As contrasted with "zle smart-insert-last-word -n 7" which is
supposed to select the 7th-from-last word of the previous command.

}   * to always get a word that wasn't proposed before (when one uses
}     ^]_ several times) by skipping history lines when needed?

That would require both that we solve the previous problem *and* store
somewhere all previous words that have been offered *and* do some kind
of negated match (e.g., the $lastcmd[(I)$pattern] change in the patch
below would have to be made back into a loop with two comparisons on
every word).


Index: smart-insert-last-word
===================================================================
diff -c -r1.2 smart-insert-last-word
--- smart-insert-last-word	15 Mar 2003 17:43:15 -0000	1.2
+++ smart-insert-last-word	15 Oct 2006 18:16:30 -0000
@@ -35,7 +35,7 @@
 #   bindkey '\e=' insert-last-assignment
 
 emulate -L zsh
-setopt extendedglob
+setopt extendedglob nohistignoredups
 
 # Not strictly necessary:
 # (($+_ilw_hist)) || integer -g _ilw_hist _ilw_count _ilw_cursor _ilw_lcursor
@@ -64,27 +64,34 @@
 _ilw_hist=$HISTNO
 _ilw_count=$NUMERIC
 
+if [[ -z "$numeric" ]]
+then
+    zstyle -s :$WIDGET match pattern ||
+	pattern='*[[:alpha:]/\\]*'
+fi
+
 zle .up-history || return 1      # Retrieve previous command
 lastcmd=( ${${(z)BUFFER}:#\;} )  # Split into shell words
+if [[ -n "$pattern" ]]
+then
+    integer n=0 found=$lastcmd[(I)$pattern]
+    while (( found == 0 && ++n ))
+    do
+        zle .up-history || return 1
+        lastcmd=( ${${(z)BUFFER}:#\;} )
+        found=$lastcmd[(I)$pattern]
+    done
+    (( found-- && ( NUMERIC += n ) ))
+fi
 zle .down-history                # Return to current command
 CURSOR=$cursor                   # Restore cursor position
 NUMERIC=${numeric:-1}            # In case of fall through
 
 (( NUMERIC > $#lastcmd )) && return 1
 
-if [[ -z "$numeric" ]]
+if [[ -n "$pattern" ]]
 then
-    integer i=1
-    zstyle -s :$WIDGET match pattern ||
-	pattern='*[[:alpha:]/\\]*'
-    while ((i <= $#lastcmd)); do
-	if [[ $lastcmd[-i] == $~pattern ]]; then
-	    NUMERIC=$i
-	    break
-	else
-	    ((++i))
-	fi
-    done
+    NUMERIC=$(( $#lastcmd - found ))
 fi
 LBUFFER[lcursor+1,cursor+1]=$lastcmd[-NUMERIC]
 _ilw_cursor=$CURSOR


-- 

Re: smart-insert-last-word bugs

[Bart Schaefer]

On Oct 15,  5:58pm, Vincent Lefevre wrote:
}
} ... another one: after I type a command and use completion,
} smart-insert-last-word removes the space between the command and the
} argument.

I think this is the generic problem with user-defined widgets and
autoremoval, to wit, whether to do autoremoval is a property of the
*next* widget executed (*after* the one that did the auto-insertion)
and there's no way to tell zle what the autoremoval property should
be for a user-defined widget.

This can be worked around by arranging that the first thing the user-
defined widget does is execute a built-in widget that thas the desired
autoremoval behavior.  So perhaps a reasonable fix would be to create
a couple of built-in widgets that do nothing *except* autoremoval. E.g.,

       zle remove-preceding-auto-suffix
and
       zle retain-preceding-auto-suffix

(choose other names as appropriate).

Re: About insert-last-word and "command args | less"

[Bart Schaefer]

On Oct 15, 12:39pm, Vincent Lefevre wrote:
}
}   * to skip a history line when all its words match the pattern?
}   * to always get a word that wasn't proposed before (when one uses
}     ^]_ several times) by skipping history lines when needed?

Maybe what you really want here is not smart-insert-last-word, but a
variant of _history_complete_word that considers only the rightmost
"interesting" word of each command in the history?

Re: About insert-last-word and "command args | less"

[Vincent Lefevre]

On 2006-10-15 11:49:43 -0700, Bart Schaefer wrote:
> Maybe what you really want here is not smart-insert-last-word, but a
> variant of _history_complete_word that considers only the rightmost
> "interesting" word of each command in the history?

Perhaps (not sure if smart-insert-last-word was supposed to do that).

Also, I've seen another bug in smart-insert-last-word: When I recall
a previous history line, e.g. with up-line-or-history or with
history-incremental-search-backward, then kill the last word and use
insert-last-word, instead of getting the last word of the lastest
history line, I get the last word of the history line that was
entered just before the recalled history line.

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

Re: smart-insert-last-word bugs

[Bart Schaefer]

On Oct 15, 11:46am, Bart Schaefer wrote:
}
} ... there's no way to tell zle what the autoremoval property should
} be for a user-defined widget.
} 
} ... perhaps a reasonable fix would be to create } a couple of built-in
} widgets that do nothing *except* autoremoval.

Here's a patch.  I'm not entirely certain whether we also need versions
of these that have the ZLE_MENUCMP or ZLE_ISCMP flags, or whether instead
of ZLE_NOTCOMMAND these should have ZLE_MENUCMP.

I chose to call them auto-suffix-remove and auto-suffix-retain.  Better
suggestions welcome.

Index: Src/Zle/iwidgets.list
===================================================================
--- iwidgets.list	31 May 2006 14:11:59 -0000	1.6
+++ iwidgets.list	15 Oct 2006 23:05:04 -0000
@@ -14,6 +14,8 @@
 "accept-line", acceptline, 0
 "accept-line-and-down-history", acceptlineanddownhistory, 0
 "argument-base", argumentbase, ZLE_MENUCMP | ZLE_KEEPSUFFIX | ZLE_LASTCOL | ZLE_NOTCOMMAND
+"auto-suffix-remove", handlesuffix, ZLE_NOTCOMMAND
+"auto-suffix-retain", handlesuffix, ZLE_KEEPSUFFIX | ZLE_NOTCOMMAND
 "backward-char", backwardchar, 0
 "backward-delete-char", backwarddeletechar, ZLE_KEEPSUFFIX
 "backward-delete-word", backwarddeleteword, ZLE_KEEPSUFFIX
Index: Src/Zle/zle_utils.c
===================================================================
--- zle_utils.c	1 Oct 2006 02:38:53 -0000	1.10
+++ zle_utils.c	15 Oct 2006 22:11:16 -0000
@@ -865,6 +865,15 @@
     return 0;
 }
 
+/* user control of auto-suffixes -- see iwidgets.list */
+
+/**/
+int
+handlesuffix(UNUSED(char **args))
+{
+  return 0;
+}
+
 /***************/
 /* undo system */
 /***************/
Index: Doc/Zsh/zle.yo
===================================================================
--- zle.yo	31 May 2006 14:11:59 -0000	1.20
+++ zle.yo	15 Oct 2006 22:56:54 -0000
@@ -1665,6 +1665,24 @@
 Execute the current line, and push the next history
 event on the the buffer stack.
 )
+tindex(auto-suffix-remove)
+item(tt(auto-suffix-remove))(
+If the previous action added a suffix (space, slash, etc.) to the word on
+the command line, remove it.  Otherwise do nothing.  Removing the suffix
+ends any active menu completion or menu selection.
+
+This widget is intended to be called from user-defined widgets to enforce
+a desired suffix-removal behavior.
+)
+tindex(auto-suffix-retain)
+item(tt(auto-suffix-retain))(
+If the previous action added a suffix (space, slash, etc.) to the word on
+the command line, force it to be preserved.  Otherwise do nothing.
+Retaining the suffix ends any active menu completion or menu selection.
+
+This widget is intended to be called from user-defined widgets to enforce
+a desired suffix-preservation behavior.
+)
 tindex(beep)
 item(tt(beep))(
 Beep, unless the tt(BEEP) option is unset.

-- 

Re: About insert-last-word and "command args | less"

[Bart Schaefer]

On Oct 15, 11:14pm, Vincent Lefevre wrote:
} Subject: Re: About insert-last-word and "command args | less"
}
} On 2006-10-15 11:49:43 -0700, Bart Schaefer wrote:
} > Maybe what you really want here is not smart-insert-last-word, but a
} > variant of _history_complete_word that considers only the rightmost
} > "interesting" word of each command in the history?
} 
} Perhaps (not sure if smart-insert-last-word was supposed to do that).

Well, no, it really wasn't supposed to do that.  It was supposed to find 
an "interesting" word if there was one, and otherwise behave just like
insert-last-word.  I'm not sure the behavior of continuing to search up
through additional history lines is always wanted.
 
} Also, I've seen another bug in smart-insert-last-word: When I recall
} a previous history line, e.g. with up-line-or-history or with
} history-incremental-search-backward, then kill the last word and use
} insert-last-word, instead of getting the last word of the lastest
} history line, I get the last word of the history line that was
} entered just before the recalled history line.

Hrm.

This reveals two problems.  (1) The definition of the "current" history
line, where [builtin] insert-last-word begins, does not mean the line
currently displayed in the editor buffer.  (2) Parsing $BUFFER to find
words for insertion isn't safe, because BUFFER reflects edits made to
any history lines during the current zle session, but insert-last-word
looks at the original unedited lines.

I'm a bit annoyed by the remifications of this, namely that to emulate
insert-last-word it's necessary to load the zsh/parameter module and
pull lines out of the $history hash.

Here's a new copy of smart-insert-last-word (the diff is only 9 lines
shorter than the entire function) which attempts to address all of this.
It introduces a new zstyle "auto-previous" to control the behavior that
Vincent wants, and employs an "always" block and the side-effects of
assignment to HISTNO to handle invocations on history lines other than
the most recent.  It requires my auto-suffix-retain patch; comment out
that line if you want to try it with something less than the latest CVS
HEAD checkout.

If this looks OK, I'll commit it along with a documentation update.

--- 8< --- snip --- 8< ---
# smart-insert-last-word
# Inspired by Christoph Lange <langec@gmx.de> from zsh-users/3265;
# rewritten to correct multiple-call behavior after zsh-users/3270;
# modified to work with copy-earlier-word after zsh-users/5832.
# Edited further per zsh-users/10881 and zsh-users/10884.
#
# This function as a ZLE widget can replace insert-last-word, like so:
#
#   zle -N insert-last-word smart-insert-last-word
#
# With a numeric prefix, behaves like insert-last-word, except that words
# in comments are ignored when interactive_comments is set.
#
# Otherwise, the rightmost "interesting" word from any previous command is
# found and inserted.  The default definition of "interesting" is that the
# word contains at least one alphabetic character, slash, or backslash.
# This definition can be overridden by use of a style like so:
#
#   zstyle :insert-last-word match '*[[:alpha:]/\\]*'
#
# For example, you might want to include words that contain spaces:
#
#   zstyle :insert-last-word match '*[[:alpha:][:space:]/\\]*'
#
# Or include numbers as long as the word is at least two characters long:
#
#   zstyle :insert-last-word match '*([[:digit:]]?|[[:alpha:]/\\])*'
#
# That causes redirections like "2>" to be included.
#
# Note also that the style is looked up based on the widget name, so you
# can bind this function to different widgets to use different patterns:
#
#   zle -N insert-last-assignment smart-insert-last-word
#   zstyle :insert-last-assignment match '[[:alpha:]][][[:alnum:]]#=*'
#   bindkey '\e=' insert-last-assignment
#
# The "auto-previous" style, if set to a true value, causes the search to
# proceed upward through the history until an interesting word is found.
# If auto-previous is unset or false and there is no interesting word, the
# last word is returned.

emulate -L zsh
setopt extendedglob nohistignoredups

# Begin by preserving completion suffix if any
zle auto-suffix-retain

# Not strictly necessary:
# (($+_ilw_hist)) || integer -g _ilw_hist _ilw_count _ilw_cursor _ilw_lcursor

integer cursor=$CURSOR lcursor=$CURSOR
local lastcmd pattern numeric=$NUMERIC

# Save state for repeated calls
if (( HISTNO == _ilw_hist && cursor == _ilw_cursor )); then
    NUMERIC=$[_ilw_count+1]
    lcursor=$_ilw_lcursor
else
    NUMERIC=1
    _ilw_lcursor=$lcursor
fi
# Handle the up to three arguments of .insert-last-word
if (( $+1 )); then
    if (( $+3 )); then
	((NUMERIC = -($1)))
    else
	((NUMERIC = _ilw_count - $1))
    fi
    (( NUMERIC )) || LBUFFER[lcursor+1,cursor+1]=''
    numeric=$((-(${2:--numeric})))
fi
_ilw_hist=$HISTNO
_ilw_count=$NUMERIC

if [[ -z "$numeric" ]]
then
    zstyle -s :$WIDGET match pattern ||	pattern='*[[:alpha:]/\\]*'
fi

# Note that we must use .up-history for navigation here because of
# possible "holes" in the $history hash (the result of dup expiry).
# We need $history because $BUFFER retains edits in progress as the
# user moves around the history, but we search the unedited lines.

{
  zmodload -i zsh/parameter
  zle .end-of-history              # Start from final command
  zle .up-history || return 1      # Retrieve previous command
  local buffer=$history[$HISTNO]   # Get unedited history line
  lastcmd=( ${${(z)buffer}:#\;} )  # Split into shell words
  if [[ -n "$pattern" ]]
  then
      # This is the "smart" part -- search right-to-left and
      # latest-to-earliest through the history for a word.
      integer n=0 found=$lastcmd[(I)$pattern]
      if zstyle -t :$WIDGET auto-previous
      then
          while (( found == 0 && ++n ))
          do
              zle .up-history || return 1
              buffer=$history[$HISTNO]
              lastcmd=( ${${(z)buffer}:#\;} )
              found=$lastcmd[(I)$pattern]
          done
      fi
      (( found-- > 0 &&            # Account for 1-based index
        (numeric = $#lastcmd - found) ))
  fi
} always {
  HISTNO=$_ilw_hist                # Return to current command
  CURSOR=$cursor                   # Restore cursor position
  NUMERIC=${numeric:-1}            # In case of fall-through
}

(( NUMERIC > $#lastcmd )) && return 1

LBUFFER[lcursor+1,cursor+1]=$lastcmd[-NUMERIC]
_ilw_cursor=$CURSOR