[gnu.bash.bug] Bash-1.03 and exported PATH and history substitutions

gnb@melba.bby.oz.au (Gregory Bond) (09/14/89)

1) If you have a path line like
	PATH=:/bin:/usr/bin:~/bin:/usr/local/bin

   then when it is exported, the ~ is not expanded.  This confuses
   subsequent invocations of /bin/sh or other programs that parse
   paths, for example make.  Csh does expand components of the path it
   exports.

2) the present simple-minded history substitution has some surprises
   for csh users.  I have noted in the past the lack of delimiter
   escaping or :gs global substitutions or "&" pattern replacement.
   This one annoys me a lot: you can't add slashes with ^.^./^ as this
   is treated as "!!:s/././/" which does nothing at all.  Nor does
   ^x^xx^lly (or the equivalent ":s/x/xx/lly") do what is expected
   (i.e. turn txt into txxtlly).  Some serious rewrite of the history
   editing code would need to be done to make bash csh-compatible in
   this area.  Is this a desired end, or is there some other model of
   history editing bash is trying to emulate?  I have started looking
   at these problems and may eventually produce a better history
   editor.

Greg.

BTW: I have patches to add FIGNORE to the 1.03 filename completion (a
la SunOS csh set fignore ()) if anyone is interested.

bfox@AUREL.CALTECH.EDU (Brian Fox) (09/15/89)

   Date: Thu, 14 Sep 89 19:05:36 +1000
   From: Gregory Bond <munnari!melba.bby.oz.au!gnb@uunet.uu.net>

   1) If you have a path line like
	   PATH=:/bin:/usr/bin:~/bin:/usr/local/bin

      then when it is exported, the ~ is not expanded.  This confuses
      subsequent invocations of /bin/sh or other programs that parse
      paths, for example make.  Csh does expand components of the path it
      exports.

Csh expands words beginning with whitespace.  It does not have to expand
the components of PATH, which are separated by colons.  I could special
case the PATH variable, which I think is ugly beyond belief. 

      This one annoys me a lot: you can't add slashes with ^.^./^ as

Yes, you can.  Please try again.  Also, please specify the version of
Bash (or better yet, readline) which is giving you the problem.  In that
way, people won't assume that the bug exists in all versions of Bash,
but perhaps only in the version that you have, or only on the hardware
that you have Bash compiled on.

      is treated as "!!:s/././/"
No, it is treated as "!!:s^.^./"

      ^x^xx^lly (or the equivalent ":s/x/xx/lly") doesn't do turn
      txt into txxtlly.

This is a bug.  However, I found that "!!:s/foo/bar/more text" changed
"hellofooness" into "hellobarnessmore text", which is what I believe it
is supposed to do.  So perhaps the problem only exists in quick
substitution, with the `^' character.  (5 minutes pass) Okay, I fixed
it.  Patches for :g and for the substitution fix follow.  These patches
apply to bash 1.03's history.c file.

      Some serious rewrite of the history editing code would need to be
      done to make bash csh-compatible in this area.

Actually, I am of the opinion that a serious rewrite of Csh would be
required to give it the line editing features of Bash.  In the course of
writing this letter I have added "!:gs/foo/bar", so I would not suppose
that a "serious" rewrite is neccessary.

      Is this a desired end, or is there some other model of history
      editing bash is trying to emulate?

I provided Csh style history substitution as a bonus, and as a quick
hack.  I thought that people who had been used to using Csh would
appreciate a little flavor of home, even though all of the history
functions could be performed interactively, with the readline editing
features.  I had hoped that any time I spent working on the command line
interface would be to enhance readline, so that features missing from my
quick history hack could be better performed in the Emacs idiom.

I do not expect to spend any substantial amount of time hacking on the
arcane and non-intuitive Csh !-style history interface.  I have already
spent too much time as it is.
  I have started looking at

      these problems and may eventually produce a better history editor.

If you produce a better history editor, which is clean, and provides the
same interface as mine, I will be more than happy to make it a part of
the Readline library.  Perhaps next time, you can spend the half hour
adding the features that you want instead of me!

Brian Fox

bfox@AUREL.CALTECH.EDU (Brian Fox) (09/15/89)

Whoops, I forgot the patches...


----------------------------------------
File: /usr/gnu/src/lib/readline/history-diffs
----------------------------------------
*** history.c.~10~	Thu Aug 31 14:10:07 1989
--- history.c	Fri Sep 15 08:53:04 1989
***************
*** 661,667 ****
    int i, word_spec_error = 0;
    int cc, modified = 0;
    char *word_spec, *event;
!   int starting_index, only_printing = 0;
  
    char *get_history_word_specifier (), *rindex ();
  
--- 661,667 ----
    int i, word_spec_error = 0;
    int cc, modified = 0;
    char *word_spec, *event;
!   int starting_index, only_printing = 0, substitute_globally = 0;
  
    char *get_history_word_specifier (), *rindex ();
  
***************
*** 696,701 ****
--- 696,702 ----
  	       history_expansion_char, history_expansion_char,
  	       string);
        string = format_string;
+       l += 4;
        goto grovel;
      }
  
***************
*** 853,866 ****
  		  goto next_special;
  
  		  /* :s/this/that substitutes `this' for `that'. */
  		case 's':
  		  {
  		    char *this, *that, *new_event;
! 		    int delimiter = string[i + 2];
  		    int si, l_this, l_that, l_temp = strlen (temp);
  
! 		    if (!delimiter) break;
  
  		    i += 3;
  
  		    /* Get THIS. */
--- 854,882 ----
  		  goto next_special;
  
  		  /* :s/this/that substitutes `this' for `that'. */
+ 		  /* :gs/this/that substitutes `this' for `that' globally. */
+ 		case 'g':
+ 		  if (string[i + 2] == 's')
+ 		    {
+ 		      i++;
+ 		      substitute_globally = 1;
+ 		      goto substitute;
+ 		    }
+ 		  else
+ 		    
  		case 's':
+ 		  substitute:
  		  {
  		    char *this, *that, *new_event;
! 		    int delimiter = 0;
  		    int si, l_this, l_that, l_temp = strlen (temp);
  
! 		    if (i + 2 < strlen (string))
! 		      delimiter = string[i + 2];
  
+ 		    if (!delimiter)
+ 		      break;
+ 
  		    i += 3;
  
  		    /* Get THIS. */
***************
*** 889,912 ****
  		      goto cant_substitute;
  
  		    /* Find the first occurrence of THIS in TEMP. */
! 		    for (si = 0; (si + l_this) <= l_temp; si++)
! 		      if (strncmp (temp + si, this, l_this) == 0) {
! 			new_event =
! 			  (char *)alloca (1 + (l_that - l_this) + l_temp);
! 			strncpy (new_event, temp, si);
! 			strncpy (new_event + si, that, l_that);
! 			strncpy (new_event + si + l_that,
! 				 temp + si + l_this,
! 				 l_temp - (si + l_this));
! 			new_event[(l_that - l_this) + l_temp] = '\0';
! 			temp = new_event;
  			goto hack_specials;
  		      }
  
- 		  cant_substitute:
  		    goto event_not_found;
  		  }
- 
  
  		  /* :# is the line so far.  Note that we have to
  		     alloca () it since RESULT could be realloc ()'ed
--- 905,945 ----
  		      goto cant_substitute;
  
  		    /* Find the first occurrence of THIS in TEMP. */
! 		    si = 0;
! 		    for (; (si + l_this) <= l_temp; si++)
! 		      if (strncmp (temp + si, this, l_this) == 0)
! 			{
! 			  new_event =
! 			    (char *)alloca (1 + (l_that - l_this) + l_temp);
! 			  strncpy (new_event, temp, si);
! 			  strncpy (new_event + si, that, l_that);
! 			  strncpy (new_event + si + l_that,
! 				   temp + si + l_this,
! 				   l_temp - (si + l_this));
! 			  new_event[(l_that - l_this) + l_temp] = '\0';
! 			  temp = new_event;
! 
! 			  if (substitute_globally)
! 			    {
! 			      si += l_that;
! 			      l_temp = strlen (temp);
! 			      substitute_globally++;
! 			      continue;
! 			    }
! 
! 			  goto hack_specials;
! 			}
! 
! 		  cant_substitute:
! 
! 		    if (substitute_globally > 1)
! 		      {
! 			substitute_globally = 0;
  			goto hack_specials;
  		      }
  
  		    goto event_not_found;
  		  }
  
  		  /* :# is the line so far.  Note that we have to
  		     alloca () it since RESULT could be realloc ()'ed
----------------------------------------

gnb@melba.bby.oz.au (Gregory Bond) (09/19/89)

   >From:  bfox%aurel.caltech.edu@murtoa.cs.mu.oz (Brian Fox)
   >Subject:  Bash-1.03 and exported PATH and history substitutions
   >Date:	Fri, 15 Sep 89 09:07:19 H

       Date: Thu, 14 Sep 89 19:05:36 +1000
       From: Gregory Bond <munnari!melba.bby.oz.au!gnb@uunet.uu.net>

       1) If you have a path line like
    	   PATH=:/bin:/usr/bin:~/bin:/usr/local/bin

          then when it is exported, the ~ is not expanded.  This confuses
          subsequent invocations of /bin/sh or other programs that parse
          paths, for example make.  Csh does expand components of the path it
          exports.

    Csh expands words beginning with whitespace.  It does not have to expand
    the components of PATH, which are separated by colons.  I could special
    case the PATH variable, which I think is ugly beyond belief. 

I agree, it's ugly.  However, some method of expanding ~name in
PATH-like variables would be nice.  I think a general-purpose
mechanism is required (for things like MANPATH etc), but I can't think
of any clean way to do this expansion using expr or test etc, nor can
I think of a method of constructing the PATH to expand at construction
time.  But some special operator is also ugly.  Perhaps a
string-concatenation operator: 
	PATH=:/bin:/usr/bin: + ~/bin + :rest:of:path
(In csh this happens because PATH is set using the csh space-separated
syntax.)

          This one annoys me a lot: you can't add slashes with ^.^./^ as

    Yes, you can.  Please try again.  Also, please specify the version of
    Bash (or better yet, readline) which is giving you the problem.  

Um, red faces time.  I checked this before posting, but must have
checked on the 1.02 shell (where it definitly didn't work). 


    I do not expect to spend any substantial amount of time hacking on the
    arcane and non-intuitive Csh !-style history interface.  I have already
    spent too much time as it is.

  	  I have started looking at
          these problems and may eventually produce a better history editor.

    If you produce a better history editor, which is clean, and provides the
    same interface as mine, I will be more than happy to make it a part of
    the Readline library.  Perhaps next time, you can spend the half hour
    adding the features that you want instead of me!

Better in the sense of more like csh - for many years of habit die
hard.  (The number of times I type <ESC> in bash and emacs to get file
completion... ).  Given the :gs stuff, the next thing I miss is & in
the replacement string.  I AM working on this, but it's a bit more
than 1/2 hour for someone who doesn't know the code.  This whole
history mechanism is nasty twisted job, full of odd surprises and dark
corners.  I can well understand the desire to leave it behind.

But, boy, I do love this shell.

Greg.  Bash fanatic.