The new implementation is less efficient but definitely correct.

This reverts commits ea7c165b59 and
3d81c83132.

When "have 6 expectations and 4 region_highlight entries", the pure-zsh
implementation printed them as follows:

    not ok 7 - cardinality check - have 6 expectations and 4 region_highlight entries: «expected_region_highlight=( $'1 1 builtin' $'3 6 comment' $'8 13 comment' $'15 15 default' $'16 21 comment' $'22 22 default' )»
    «region_highlight=( $'0 1 builtin' $'2 6 comment' $'7 13 comment' $'14 22 default' )»
    # expected_region_highlight  '22 22 default'
    # '1 1 builtin'              region_highlight
    # '3 6 comment'              '0 1 builtin'
    # '8 13 comment'             '2 6 comment'
    # '15 15 default'            '7 13 comment'
    # '16 21 comment'            '14 22 default'

Whereas the column(1)-based implementation prints them as follows:

    not ok 7 - cardinality check - have 6 expectations and 4 region_highlight entries: «expected_region_highlight=( $'1 1 builtin' $'3 6 comment' $'8 13 comment' $'15 15 default' $'16 21 comment' $'22 22 default' )» «region_highlight=( $'0 1 builtin' $'2 6 comment' $'7 13 comment' $'14 22 default' )»
    # expected_region_highlight  region_highlight
    # '1 1 builtin'              '0 1 builtin'
    # '3 6 comment'              '2 6 comment'
    # '8 13 comment'             '7 13 comment'
    # '15 15 default'            '14 22 default'
    # '16 21 comment'
    # '22 22 default'

Ultimately, this difference is down to the pure-zsh implementation
getting the arguments as a single list, whereas paste(1) gets two
separate lists.

Command position: non-executable files, non-cdable directories, CDPATH false positives

Fixes #461.

(q-) passes through newlines and NUL bytes verbatim.  Using (qqqq) ensures the
escaped string will be on a single line (as required by the TAP format) and be
readable even if it contains control characters.

Fixes #475.

Fixes #504.

Fixes #202.

Test expectations are updated.  For example, BUFFER='/bin' is now
highlighted as path_prefix because it's a prefix of '/bin/sh' which
would be valid.  However, BUFFER='/bin;' is now properly highlighted
as an error (unless AUTO_CD is set).

The next commits will change this behaviour.

Fixes #430.

Will be fixed in the next commit.

If the right column was longer, the excess entries were printed on the left column.

Part of issue #645.

Part of issue #645.

Fixes #609.

We already declare $expected_region_highlight in run_test_internal().
Therefore, it will always be declared.

It was broken by commit e6eea1f9b7e67136911536a2450c7130f44c0b04,
"test harness: Don't leak options from test files to the test harness.".

See the last commit,
bdb4e8b70e test harness: When the cardinality check fails, pretty-print \$expected_region_highlight and \$region_highlight.

* danielsh/whitespace-historical:
  Fix historical instances of one-space indentation.

Fixes an issue whereby the '# TODO "issue #687"' directive in the output of
opt-shwordsplit1.zsh was truncated, because the test itself had set the
SH_WORD_SPLIT option and that affected the evaluation of
«${(z)expected_region_highlight[i]}» in the test harness.

Furthermore, this patch also independently fixes the error under
zsh-5.0.8 and earlier that was fixed by the previous commit.

The output of test-data/opt-shwordsplit1.zsh on zsh 5.7 is:
.
    1..2
    ## opt-shwordsplit1
    # BUFFER=vim
    not ok 1 - [1,7] «$EDITOR» - expected (1 7 "function"), observed (1 7 "unknown-token"). # TODO "issue
    ok 2 - cardinality check # SKIP cardinality check disabled whilst regular test points are expected to fail

On zsh 5.0.8, tap_escape() choked when called on the arguments argv=('[1,7]'
'«vim»').  This patch fixes that.

As you may have noticed, under zsh 5.7 the diagnostic message of test point 1
is truncated.  That'll be fixed in the next commit.

The escaped caret was taken for a negated character class.  This caused test-data/arith1.zsh
to XPass: the arithmetic expansion was consumed by the 'while' loop.

This is not perfect: we don't try to detect cases such as «$((ls); (ls))»,
which look like arithmetic expansions but are in fact command substitutions.

Fixes part of #607.

Introduces #704.

The cardinality check shall —

- if the test sets \$expected_mismatch, be XFail;
- elif any test points is XFail, be skipped;
- else, be expected to pass.

To test this, change «6 * 9» to «6 + 9» in test-data/arith1.zsh that
will be added in the after-next (grandchild) commit.