Hooks have always been a mess. There was no organization when they were
added, as when the project first started there was no need for them.

To make things consistent, break out disabling hooks into
__lp_disable_hooks(), and make prompt_on() be the enabler of hooks. Then
have lp_activate() disable hooks at the start, and enable hooks at the
end.

Some small changes around how state is detected, but no functional
difference here.

Based on the work done in ecdde384, the
prompt_bang and prompt_subst options, which handle '!' and '$'/'`'
characters respectively, can be left untouched. When running
__lp_escape(), those options are checked, and if set, their characters
are escaped.

prompt_percent however, must still be enabled always. The shell sequence
escapes, '%{' and '%}' only work when the option is enabled. Maybe
someday we can detect when the option is off and disable all terminal
formatting, but it doesn't seem worth the effort.

Backslashes must be escaped even on Zsh to prevent confusion with a
backslash directly preceding a character that would give is special
meaning.

For some reason, Bash ignores a '\$' sequence when the promptvars option
is off, printing it as '$'. It does not do the same for a backtick
escaped: '\`' is printed as '\`'. To prevent this issue, detect if the
option is on, and if so, escape both '$' and '`'.

Running 'shopt' is actually faster than either a [[ ]] or (( ))
expression.

PR/679

Extension to #496 to add [AWS Vault](https://github.com/99designs/aws-vault)
support. AWS Vault does not set the AWS_PROFILE/AWS_DEFAULT_PROFILE
variables, but does add AWS_VAULT. Since this is philosophically the
same as showing an AWS profile on the prompt, I simply added AWS_VAULT
to the detection chain.

PR/678

Also remove some trailing space.

Closes: nojhan/liquidprompt#675

This was missed in the PR for the Ruby Virtualenv section.
The order in the docs for Terraform and Kubecontext was backwards from
the order in the code.
Moved parts of the PS1 building to different lines to make it easier to
read and add new sections.

Resolves #676

PR/673

LP_KUBECONTEXT_DELIMITER_SUFFIX removes a suffix (dev-cluster and - becomes ->
dev). But sometimes the boilerplate part is a prefix, as when using AWS
EKS (Elastic Kubernetes Service).

`aws eks update-kubeconfig --name $CLUSTER_NAME` sets the cluster's ARN
as the context name, most of which is boilerplate:
`arn:aws:eks:$REGION:$ACCOUNT_ID:cluster/$CLUSTER_NAME`.

PR/671

This breaks several VCS functions (mercurial, fossil, svn).

Rarely, sensors or acpi can return no temperature data. In this case,
the data function would return false and 0, but the detect function
would error if set -u was enabled.

Harden all of the four temperature functions against this, and greatly
increase the test coverage to cover these four functions, as well as
adding a specific case for the tools returning no data.

This also allows for the temperature functions to return values less
than 0, if somehow a device has all temp sensors returning less than
that.

Mention in the function documentation that it is possible for the
function to return a not set string.

Fixes 670

Fixes https://github.com/nojhan/liquidprompt/issues/600.

Partially through the work for 2.1, it seemed like it was time to update
the Changelog.

Adding the work done in commit 9f878488
to the Powerline full theme.

Adding the work done in commit 0e552ba7
to the Powerline full theme.

Adding the work done in commit 0660342d
to the Powerline full theme.

In random older versions of Bash, accessing arrays when set -u is set
will crash or return incorrect values. This is a continuation of commit
e121179d, but for the powerline themes.

Zsh 5.0 continues to be a huge pain. As noted in the shell wiki, all
variables assigned to another variable as part of a 'local' or 'typeset'
line must be quoted, or things will break. The reason I caught these
many instances was a failing test on Zsh-5.0.8 because of the bad line
in __lp_pwd_tilde().

Undeclared local variables are the worst. When declaring a local
variable, Zsh treats it as set (to the empty) string, while Bash (more
correctly, in my opinion) treats it as unset. This meant that this check
of if the first powerline section had happened yet would always return
true on Zsh, showing an ugly arrow at the start of the prompt. On
Bash-3.2 (with set -u) it would just crash, which is how I noticed the
bug.

The solution is to never rely on a local being set or not, but to look
at the actual value. In this case check for the empty string (with a set
-u safety).

While not useful in CI scripts, this lets a developer run the tests
against their own shells if they wish.

I tested this with:
bash 3.2.0(1)-release (x86_64-unknown-linux-gnu)
bash 4.0.28(2)-release (x86_64-unknown-linux-gnu)
bash 4.2.53(1)-release (x86_64-unknown-linux-gnu)
bash 4.4.0(1)-release (x86_64-unknown-linux-gnu)
bash 4.4.19(1)-release (x86_64-unknown-linux-gnu)
bash 5.0.17(1)-release (x86_64-pc-linux-gnu)
bash 5.1.0(1)-release (x86_64-pc-linux-gnu)
zsh 5.0.8 (x86_64-unknown-linux-gnu)
zsh 5.1.1 (x86_64-unknown-linux-gnu)
zsh 5.2 (x86_64-unknown-linux-gnu)
zsh 5.3.1 (x86_64-unknown-linux-gnu)
zsh 5.4.2 (x86_64-unknown-linux-gnu)
zsh 5.5.1 (x86_64-redhat-linux-gnu)
zsh 5.8 (x86_64-pc-linux-gnu)

I found a few bugs, fixed in the following commits.

While (again) running the theme preview script does not really test that
it is working, nor does it do like the old test script did and actually
validate that output of the themes are what we expect, it will now catch
runtime errors, which could be incredibly helpful.

Parameter substitution is (like most standards) weakly defined, and here
is a case where it was not tested on Zsh. Simple fix of wrapping the
whole string (the '(' was the bad character) in quotes.

Also add a test case to catch a regression.

Fixes #667

When the LP_COLORMAP config option contained a first element that was
the empty string, lp_activate() would think it was empty, and set
LP_COLORMAP back to the default value.

This was identified by @milouse in a comment here:
https://github.com/nojhan/liquidprompt/pull/628#discussion_r611182040

The issue is not severe, as any value set in the config will always be
taken (as it is set last). But a value set in the environment, either
manually or in a shell RC file would be overridden (as long as the first
element was the empty string).

Commit e121179d has a good explaination
of why the ${array[@]+"${array[@]}"} form is used. This unfortunately
must be used _anywhere_ an array could be empty or have any empty
strings as values.

Fixes: https://github.com/nojhan/liquidprompt/issues/249

The problem is simple enough: the extra backslashes or percents that
__lp_escape() adds show up in the title, as the title is changed in the
previous commit to no longer be printed as part of PS1. The fix is not
so simple.

This patch adds "raw" copy of the data strings for username, hostname,
and current path with no color or shell character escaping. This is
designed for putting into the terminal title, but it could also be used
for comparisons. lp_path_format() already had a no formatting return var
in lp_path, but now it is not shell escaped.

The title is changed from duplicating the prompt to instead showing just
the core bits in the central brackets: username, hostname, current path,
followed by the prompt mark. As well as the user prefix, postfix, and
MARK_PREFIX. This is a change from the long standing default, but it is
customizable both by themes and users, and I have long seen complaints
of the current title layout being less than helpful.

There is one small issue though: to prevent LP_PS1_PREFIX,
LP_PS1_POSTFIX, and LP_MARK_PREFIX that have color from messing up the
title, the whole title string is still passed into lp_formatted_title()
to strip color out. This is less than ideal, but faster than treating
those 3 variables by themselves.

 configuation -> configuration

Rationale:

* In general, it is not proper for liquidprompt to change shell
  options that it does not absolutely have to change.

* LP_PS1_{PREFIX,POSTFIX} are a thing, and are in fact an effective
  means to combine liquidprompt with other prompt-touching logics
  (such as emacs-vterm).  So it is not good to prevent those other
  logics from actually working.

This commit restores the bash version of __lp_escape to what it did
prior to 62f02708 (escaping dollar signs in addition to backslashes)
and removes all interference with the promptvars shell option.

NOTE: I don't know much about shells other than bash, so left the
parallel logic for zsh as is.

This feature depends on the removal of the prompt escape sequences that
came before. Part of that is the removal of adding the title to $PS1:
now it is just print to the screen. This could improve speed slightly,
as the shell doesn't need to parse that text. But it does prevent usage
of prompt escape codes in the title.

Which causes a rare issue where __lp_escape() adds duplicate backslashes
or percents to escape a string for the prompt, but now are not needed in
the title. It will just print double in the title, so it isn't a huge
issue, but is something that should be fixed. I'm not sure how to handle
it though, since the current architecture requires escaping the strings
before formatting is added to them.

I also needed to rework how the runtime feature was handled, since it
monopolized the DEBUG trap in Bash. Now it can handle multiple features
that need to run after the user has submitted the command, but before it
runs.

The feature itself is simple: lookup the latest command in the history,
and print the normal title postpended by that command to the screen,
formatted as a terminal title. Note that Bash's $BASH_COMMAND is not
great, as it only contains the current _command_, not the whole command
line.

Resolves #609

This was forgotten before.