[PATCH] Fix escaping of \034 (^\) in paths for \w, \W, and \s

[Koichi Murase]

I have tried the latest devel branch to see the behavior of the
escaping of directory names (\w, \W) in the prompt implemented in the
recent commit b4e5e550, which was originally discussed in the thread
[1].  It now mostly works as expected, but I noticed that the escaping
of \034 is incomplete.

[1] https://lists.gnu.org/archive/html/bug-bash/2022-01/msg00051.html

----

Bash Version: 5.2 (devel branch)
Commit b6a567e7f13406952cbb1d1adb2f00b2260a871e
Commit b4e5e5505cc4495c6237c32a65ba62ebcc497b31

Description:

  The escaping of the invisible characters in the directory names in
  the prompt escapes \w, \W and the path in \s has been implemented in
  the function `sh_backslash_quote_for_double_quotes'
  (lib/sh/shquote.c:323) in commit b4e5e550.  The primary purpose of
  the function `sh_backslash_quote_for_double_quotes' is to quote
  special characters in double quotes with a backslash.  When the
  escaping of the invisible characters are turned on by the argument
  FLAG, a backslash in the visible representation of $'\034', '^\', it
  not properly escaped by another backslash.  This may break the
  escaping of the succeeding special character.

Repeat-By:

    bash-dev$ PS1='\w\$ '
    ~$ mkdir $'\034$(echo hello)'
    ~$ cd !$
    ^\hello$

  Here, we expect '^\$(echo hello)' for the visible representation of
  the directory name. Here the command substitution in the directory
  name is unexpectedly evaluated replaced by its stdout `hello'.

Fix:

  In the attached patch `0001-fix-prompt-charvis-A.patch', the
  escaping of the invisible characters is processed before the
  backslash quoting so that the backslash quoting is also applied to
  the results of the escaping of invisible characters.  With this
  patch, the string is processed twice from the beginning to the end
  because the escaping of the invisible characters and the quoting of
  the special characters in double quotes are processed separately.

  To process both quoting in a single pass properly, I have prepared
  another alternative patch `0001-fix-prompt-charvis-A.patch' where I
  have extended the function `sh_charvis (lib/sh/strvis.c)' to accept
  an extra argument FLAGS.  When FLAGS has the bit (FLAGS & 1),
  `sh_charvis' quote special characters of double quotes by a
  backslash.

  However, I think the first patch is better because it is cleaner and
  also because I do not think that the performance can be a problem
  for the results of \w, \W, and \s. [ Note that the function
  `sh_backslash_quote_for_double_quotes' is currently called with
  (flags & 1) (enable the invisible-character escaping) only for \w,
  \W and \s ]

--
Koichi

0001-fix-prompt-charvis-A.patch
Description: Binary data

0001-fix-prompt-charvis-B.patch
Description: Binary data