spurious highlight when using match on character at end of soft-wrapped line

[lacygoill]

Steps to reproduce

Run this shell command:

vim -Nu NONE -S <(cat <<'EOF'
    vim9script
    &l:breakindent = true
    &l:linebreak = true
    set breakat+=]
    printf('%s]%s[xxx]%s', repeat('x', 70), repeat('x', 50), repeat('x', 70))
        ->setline(1)
    normal! 6e
    matchaddpos('ErrorMsg', [[1, 126]])
EOF
)

Notice that on the second soft-wrapped line, the highlight is applied from the cell occupied by the last character, up to the very last cell of the window.

Expected behavior

Maybe it's working as intended, but I think it's still distracting and misleading for the user. There are no real characters on the cells after the last ].

Shouldn't the highlight be applied only on the cell whose coordinates were passed to matchaddpos()? That is:

[1, 126]

And not:

[1, 126]
[1, 127]
[1, 128]
...

Operating system

Ubuntu 20.04.3 LTS

Version of Vim

8.2 Included patches: 1-3692

Screenshot

Additional Context

I found this issue while using the matchparen plugin:

vim -Nu NORC -S <(cat <<'EOF'
    vim9script
    &l:breakindent = true
    &l:linebreak = true
    set breakat+=]
    printf('%s]%s[xxx]%s', repeat('x', 70), repeat('x', 50), repeat('x', 70))
        ->setline(1)
    normal! 4e
EOF
)

---

Same issue with a text property:

vim9script
&l:breakindent = true
&l:linebreak = true
set breakat+=]
printf('%s]%s[xxx]%s', repeat('x', 70), repeat('x', 50), repeat('x', 70))
    ->setline(1)
normal! 6e
prop_type_add('test', {highlight: 'ErrorMsg'})
prop_add(1, 126, {length: 1, type: 'test'})

[lacygoill]

I found this issue while using the matchparen plugin

Which gives a confusing user experience when we move on a bracket whose matching pair is at the end of a soft-wrapped line:

[lacygoill]

The patch 8.2.3698 fixed the issue for a highlight installed by a match, but not for a highlight installed by a text property:

vim9script
&l:breakindent = true
&l:linebreak = true
set breakat+=]
printf('%s]%s[xxx]%s', repeat('x', 70), repeat('x', 50), repeat('x', 70))
    ->setline(1)
normal! 6e
prop_type_add('test', {highlight: 'ErrorMsg'})
prop_add(1, 126, {length: 1, type: 'test'})

In practice, this means that the Vim9 refactoring is still affected by the issue, because it uses text properties.

[lacygoill]

The patch 8.2.3700 fixed the second issue, but it created a third one which is worse because it occurs much more frequently:

vim9script
&l:linebreak = true
"\t[xxx]"->setline(1)
prop_type_add('test', {highlight: 'ErrorMsg'})
prop_add(1, 2, {length: 1, type: 'test'})

Notice that all the cells covered by the tab character are highlighted, because it's in front of the opening bracket. Before the patch, only the bracket was highlighted.

I have no idea why 'linebreak' affects this issue. The line is too short to be wrapped anyway, so it shouldn't matter; but it does.

If it can't be fixed, I think it would be better to simply revert the patch, because it makes the matchparen refactoring much worse. This is especially noticeable when reading Vim's help pages, where lines are indented with tabs, and often start with brackets.

[lacygoill]

The patch 8.2.3700 introduced a regression:

vim9script
&linebreak = true
&wrap = true
printf('%s+(%s)', 'x'->repeat(&columns / 2), 'x'->repeat(&columns / 2))->setline(1)
prop_type_add('test', {highlight: 'ErrorMsg'})
prop_add(1, (&columns / 2) + 2, {length: 1, type: 'test'})

Notice that all the empty cells after the + character on the first soft-wrapped line are highlighted; only the opening parenthesis on the next line should be.

[lacygoill]

Not sure whether it's the same issue, but here is another case where some cells are highlighted even though they are not occupied by any character in the buffer:

vim9script
printf('%s (:%s)', 'x'->repeat(&columns / 2), 'y'->repeat(&columns / 2))
    ->setline(1)
&linebreak = true
matchadd('ErrorMsg', '([^)]*)')

Notice that, on the first line of the screen, every cell after the colon is highlighted. But none of them is occupied by a character.

---

Same behavior with a syntax rule:

vim9script
printf('%s (:%s)', 'x'->repeat(&columns / 2), 'y'->repeat(&columns / 2))
    ->setline(1)
&linebreak = true
syntax on
syntax match Test /([^)]*)/
highlight link Test ErrorMsg

Same behavior with a text property:

vim9script
printf('%s (:%s)', 'x'->repeat(&columns / 2), 'y'->repeat(&columns / 2))
    ->setline(1)
&linebreak = true
prop_type_add('Test', {highlight: 'ErrorMsg'})
prop_add(1, searchpos('(', 'nW')[1], {length: getline(1)->strlen(), type: 'Test'})

---

As a workaround, we need to invoke matchaddpos(), and explicitly pass to the latter the coordinates of cells which are occupied by a character in the buffer:

vim9script
printf('%s (:%s)', 'x'->repeat(&columns / 2), 'y'->repeat(&columns / 2))
    ->setline(1)
&linebreak = true
for col: number in range(searchpos('(', 'nW')[1], getline(1)->strlen())
    matchaddpos('ErrorMsg', [[1, col]])
endfor