marianoguerra/shrubbery/column does not have a README file
Column
An indentation column.
Not a number. A tab advances to a stop that depends on what came before it,
so two columns reached through different mixes of tabs and spaces are only
PARTIALLY ordered: one is more indented than another only when it extends
the other's sequence. When neither extends the other, comparing them is an
error the user has to be told about — see ColumnOrder::Incomparable.
Plain is the case that happens on every line of every file anyone has
written, and is kept apart so it costs no allocation. Tabbed holds the
alternating run lengths — chars, tabs, chars, tabs, …, chars — LEFTMOST
first, so a comparison is a straight walk from the start of the line and an
append is a concatenation. (The reference stores the same list reversed; the
order here is the one that makes both operations read forwards.)
Every entry is in half-columns. A run of n tabs is stored as 2n.
That uniformity is what makes the +0.5 convention safe: a | at column N
takes part in comparisons as N.5, and storing tab counts in the same unit
means the half-column tolerance in cmp can never be triggered by a
difference of one tab, which it would if the two units were mixed.
Column::add_chars
This column with n more character widths on its right.
Column::add_tab
This column with one tab on its right.
The tab starts a new run, so everything before it becomes a fixed prefix
that a later column must reproduce exactly to be comparable at all.
Column::cmp
Compare two columns.
Walks both run lists from the start of the line. A list that runs out while
the other continues is the smaller column — it is a prefix of the other,
which is exactly what "less indented" means here.
Where the two diverge, one case is tolerated and the rest are errors. The
tolerated case is a half-column difference in the LAST run of the shorter
side: that is a | sitting half a column before a tab, and the reference
answers every comparison about it with false rather than complaining.
Everything else is a genuine mixed-tab ambiguity, and the user is told.
Column::half_next
Half a column further right, or the next whole column if already halfway.
The reference's column-half-next. It is not simply half_up: applying it
twice must not reach 3.0 by way of 2.5, or a | under a | would line up
with a block's content.
Column::half_up
This column plus half a character width — the | convention.
A | at column N takes part in comparisons as N.5, which is what lets a
block's content and an alternative's | sit at the same visual column and
still be ordered against each other.
Column::le
self <= other, defined as the reference defines it: NOT other < self.
That is not the same as Lt || Eq. When the pair is Unordered — a |
half a column before a tab — other < self is false, so <= is true, while
Lt || Eq would be false. The reference leans on that, so the definition
has to be the negated one.
Column::lt
self < other, in the reference's sense: Unordered and Incomparable
are both false here, and only Incomparable is worth reporting.
Column::next
The next whole column after this one.
Floors first: the column after 2.5 is 3, not 3.5. That is what makes a
block opened by a | at 2.5 demand content at 3.
Column::of_chars
A column n whole character widths from the start of the line.
Column::plus
other placed to the RIGHT of self: self is the base, other the
offset added to it.
The reference's column+, with the arguments in the order the phrase
"column a plus b" suggests rather than the order the Racket takes them.
Column::sub
self - other in half-columns, or 0 when the two have different tab
structure.
Zero rather than an error because a difference across different tab
structures has no meaning to report — the reference makes the same choice.
Column::to_display
A readable column, for a diagnostic or a test failure.
4 for a plain column, 4.5 when the | half is on it, and
2+tab+1 when there is a tab in the middle — which is the form that
makes an "incomparable indentation" message explain itself.
ColumnOrder
The result of comparing two columns.
Five answers, not three. Incomparable is the user-facing "incomparable
indentation due to mixed tabs" error, and making it a constructor is what
stops a caller quietly defaulting it to Eq. Unordered is the reference's
half-a-column-before-a-tab tolerance: every comparison of the pair is false,
but it is not an error. Collapsing the two would either turn a tolerated
case into a failure or a failure into silence.
count_graphemes
Walk text from (lines, col), returning where it ends up.
The reference's count-graphemes, and the only place a column is advanced
over source text. Three things it must get right, all of which are silent
when wrong:
- \r\n is ONE line break, and a lone \r is one too.
- A tab starts a new run rather than advancing to a fixed stop.
- Everything else advances by one per EXTENDED GRAPHEME CLUSTER, not one per code point.
It iterates clusters rather than asking for the next boundary at each
position, which lets the segmenter make a single pass. That is sound because
\r\n, a lone \r, a lone \n and a tab are each a cluster of their
own under UAX #29, so no cluster ever straddles a case boundary here.
Source Files
Install
Download zipShrubbery notation for MoonBit: parser, source-faithful and reformatting printers, and structured diagnostics
shrubberyrhombusparserpretty-printersyntax
0.1.1
Apache-2.0
Updated 15 hours ago
27 downloadsDependencies
- marianoguerra/error-report0.1.0
- kawaz/grapheme0.10.4