sheetjs-clone/docbits/62_colrow.md

121 lines
4.2 KiB
Markdown
Raw Normal View History

#### Row and Column Properties
<details>
<summary><b>Format Support</b> (click to show)</summary>
**Row Properties**: XLSX/M, XLSB, BIFF8 XLS, XLML, SYLK, DOM, ODS
**Column Properties**: XLSX/M, XLSB, BIFF8 XLS, XLML, SYLK, DOM
</details>
Row and Column properties are not extracted by default when reading from a file
and are not persisted by default when writing to a file. The option
`cellStyles: true` must be passed to the relevant read or write function.
_Column Properties_
The `!cols` array in each worksheet, if present, is a collection of `ColInfo`
objects which have the following properties:
```typescript
type ColInfo = {
2017-09-24 23:40:09 +00:00
/* visibility */
hidden?: boolean; // if true, the column is hidden
2017-09-24 23:40:09 +00:00
/* column width is specified in one of the following ways: */
wpx?: number; // width in screen pixels
width?: number; // width in Excel's "Max Digit Width", width*256 is integral
wch?: number; // width in characters
2017-09-24 23:40:09 +00:00
/* other fields for preserving features from files */
level?: number; // 0-indexed outline / group level
2017-09-24 23:40:09 +00:00
MDW?: number; // Excel's "Max Digit Width" unit, always integral
};
```
_Row Properties_
The `!rows` array in each worksheet, if present, is a collection of `RowInfo`
objects which have the following properties:
```typescript
type RowInfo = {
/* visibility */
hidden?: boolean; // if true, the row is hidden
/* row height is specified in one of the following ways: */
hpx?: number; // height in screen pixels
hpt?: number; // height in points
level?: number; // 0-indexed outline / group level
};
```
_Outline / Group Levels Convention_
The Excel UI displays the base outline level as `1` and the max level as `8`.
Following JS conventions, SheetJS uses 0-indexed outline levels wherein the base
outline level is `0` and the max level is `7`.
<details>
2017-09-24 23:40:09 +00:00
<summary><b>Why are there three width types?</b> (click to show)</summary>
There are three different width types corresponding to the three different ways
spreadsheets store column widths:
2017-09-24 23:40:09 +00:00
SYLK and other plain text formats use raw character count. Contemporaneous tools
like Visicalc and Multiplan were character based. Since the characters had the
same width, it sufficed to store a count. This tradition was continued into the
BIFF formats.
SpreadsheetML (2003) tried to align with HTML by standardizing on screen pixel
count throughout the file. Column widths, row heights, and other measures use
pixels. When the pixel and character counts do not align, Excel rounds values.
XLSX internally stores column widths in a nebulous "Max Digit Width" form. The
Max Digit Width is the width of the largest digit when rendered (generally the
"0" character is the widest). The internal width must be an integer multiple of
the the width divided by 256. ECMA-376 describes a formula for converting
between pixels and the internal width. This represents a hybrid approach.
Read functions attempt to populate all three properties. Write functions will
try to cycle specified values to the desired type. In order to avoid potential
conflicts, manipulation should delete the other properties first. For example,
when changing the pixel width, delete the `wch` and `width` properties.
</details>
<details>
2017-09-24 23:40:09 +00:00
<summary><b>Implementation details</b> (click to show)</summary>
_Row Heights_
Excel internally stores row heights in points. The default resolution is 72 DPI
or 96 PPI, so the pixel and point size should agree. For different resolutions
they may not agree, so the library separates the concepts.
Even though all of the information is made available, writers are expected to
follow the priority order:
1) use `hpx` pixel height if available
2) use `hpt` point height if available
_Column Widths_
Given the constraints, it is possible to determine the MDW without actually
inspecting the font! The parsers guess the pixel width by converting from width
to pixels and back, repeating for all possible MDW and selecting the MDW that
minimizes the error. XLML actually stores the pixel width, so the guess works
in the opposite direction.
Even though all of the information is made available, writers are expected to
follow the priority order:
1) use `width` field if available
2) use `wpx` pixel width if available
3) use `wch` character count if available
</details>