One subject that seems to come up every other month or so is how RichEdit tables work. So I might as well post the answer. Hopefully RichEdit tables will eventually be described in the Windows SDK. They are not directly related to Math in Office, but I had mathematical expressions in mind when designing RichEdit’s table facility. Both mathematics and tables are recursive. For example you can have a fraction in the numerator of another fraction, and you can have a table in the cell of another table. So implementing tables seemed like a useful project that might also reveal how to implement a WYSIWYG implementation of mathematics. In fact, MathML <mtable>’s have a lot in common with general tables.

Most people at the time (1999) were recommending that a table cell should be represented by a whole RichEdit instance, which would give great generality. But I wanted a model that was much smaller, faster and worked with the built-in Find/Replace functionality and the RTF file converters. To this end, we needed a model, like Word’s, that was part of a single document instance, and could be overlaid on the existing paragraph structure. Accordingly RichEdit's table implementation is very efficient and fast, in fact, much faster than Word’s (although less general). Improvements have been made over the years, but the discussion that follows applies to RichEdit 4.0, which shipped with Office 2002, and RichEdit 4.1, which ships with Windows XP and Vista to this day. It also applies to later versions that ship with Office 2003 & 2007, which have additional features..

Specifically a cell containing a single line of text is represented only by that text, not by some larger structure. An empty cell consists of the single character, the cell mark U+0007. A cell containing multiple lines of text is expressed in terms of a structure that is substantially smaller than a complete edit instance, followed by the CELL mark. Tables can be nested up to 15 levels deep; higher nestings are represented by tab-delimited text. Cells can contain multiple paragraphs of any kind, e.g., bidirectional text, arbitrary tabs and alignments.

The Spring of 1999 was shortly after the Unicode Technical Committee added the U+FFF9..U+FFFB delimiter characters for describing ruby text in Japanese. These characters were available for more general use and seemed ideal for RichEdit’s internal table structure. This choice preceded the addition of the internal-use-only U+FDDO..U+FDEF characters that we use for mathematical structure characters, among other things.

In the (in-memory) backing store, a table row has the form

{CR...}CR

where { stands for the Unicode STARTGROUP character U+FFF9, and CR  is the ASCII Carriage Return character U+000D. The delimiter } stands for the Unicode ENDGROUP character U+FFFB and ... stands for a sequence of cells, each consisting of cell text terminated by the CELL mark U+0007. For example, a row with three empty cells has the plain text understructure U+FFF9 U+000D U+0007 U+0007 U+0007 U+FFFB U+000D. The start and end group character pairs are assigned identical PARAFORMAT2 information that describe the row and cell parameters.  If rows with different parameters are needed, they may follow one another with appropriate PARAFORMAT2 parameters. A horizontally or vertically merged cell has two characters: NOTACHAR (0xFFFF) followed by CELL (0x7). Any text that appears in a merged cell is stored in the first cell of the set of merged cells.

One way to insert tables is to copy/paste tables from Word. RichEdit reads and writes table RTF. For more programmatic purposes, RichEdit 4.0 introduced the message EM_INSERTTABLE (WM_USER + 232), which acts similarly to EM_REPLACESEL but inserts one or more table rows with empty cells instead of plain text. Specifically it deletes the text (if any) currently selected by the selection and then inserts empty table row(s) with the row and cell parameters given by wparam and lparam, respectively, as defined below. It leaves the selection pointing to the start of the first cell in the first row. The client can then populate the table cells by pointing the selection at the various cell end marks and inserting and formatting the desired text. Such text can include nested table rows, etc. Since wparam and lparam point at row and cell parameter structures, this API isn't compatible with Visual Basic and can't be easily added to RichEdit’s object model TOM, although TOM2 does have a general set of table interfaces.

The TABLEROWPARMS and TABLECELLPARMS structures are defined as

typedef struct _tableRowParms

{                           // EM_INSERTTABLE wparam is a (TABLEROWPARMS *)

BYTE    cbRow;          // Count of bytes in this structure

BYTE    cbCell;         // Count of bytes in TABLECELLPARMS

BYTE    cCell;          // Count of cells

BYTE    cRow;           // Count of rows

LONG    dxCellMargin;   // Cell left/right margin (\trgaph)

LONG    dxIndent;       // Row left (right if fRTL indent (similar to \trleft)

LONG    dyHeight;       // Row height (\trrh)

DWORD   nAlignment:3;   // Row alignment (like PARAFORMAT::bAlignment,

//  \trql, trqr, \trqc)

DWORD   fRTL:1;         // Display cells in RTL order (\rtlrow)

DWORD   fKeep:1;        // Keep row together (\trkeep}

DWORD   fKeepFollow:1;  // Keep row on same page as following row (\trkeepfollow)

DWORD   fWrap:1;        // Wrap text to right/left (depending on bAlignment)

// (see \tdfrmtxtLeftN, \tdfrmtxtRightN)

DWORD   fIdentCells:1;  // lparam points at single struct valid for all cells

} TABLEROWPARMS;

typedef struct _tableCellParms

{                           // EM_INSERTTABLE lparam is a (TABLECELLPARMS *)

LONG    dxWidth;        // Cell width (\cellx)

WORD    nVertAlign:2;   // Vertical alignment (0/1/2 = top/center/bottom

//  \clvertalt (def), \clvertalc, \clvertalb)

WORD    fMergeTop:1;    // Top cell for vertical merge (\clvmgf)

WORD    fMergePrev:1;   // Merge with cell above (\clvmrg)

WORD    fVertical:1;    // Display text top to bottom, right to left (\cltxtbrlv)

SHORT   dxBrdrLeft;     // Left border width (\clbrdrl\brdrwN) (in twips)

SHORT   dyBrdrTop;      // Top border width  (\clbrdrt\brdrwN)

SHORT   dxBrdrRight;    // Right border width (\clbrdrr\brdrwN)

SHORT   dyBrdrBottom;   // Bottom border width (\clbrdrb\brdrwN)

COLORREF crBrdrLeft;    // Left border color (\clbrdrl\brdrcf)

COLORREF crBrdrTop;     // Top border color (\clbrdrt\brdrcf)

COLORREF crBrdrRight;   // Right border color (\clbrdrr\brdrcf)

COLORREF crBrdrBottom;  // Bottom border color (\clbrdrb\brdrcf)

COLORREF crBackPat;     // Background color (\clcbpat)

COLORREF crForePat;     // Foreground color (\clcfpat)

} TABLECELLPARMS;

Note that paragraph-format information containing the TABLEROWPARMS and TABLECELLPARMS information is attached to the table-row delimiters as set up by the EM_ INSERTTABLE message, so merely duplicating the plain-text table structure in the backing store isn't enough to insert a working table. In fact, methods like ITextRange::SetText() convert the special delimiters U+FFF9.U+FFFB to spaces (U+0020). Note also that this table structure is nestable.

The definition of EM_INSERTTABLE is extensible, since in the future we'll probably have to support more parameters for table rows and cells. The API also inserts a consistent table row all at once, so that no illegal table parts are present on return. Hence if the document is saved after such an insertion, valid Word-compatible RTF will be written. lparam points at the TABLECELLPARMS structure for the first cell in an array of TABLECELLPARMS structures.  It's important that cbCell = sizeof(TABLECELLPARMS).  That way RichEdit knows how much cell information the client is specifying.  In particular, in the future if more cell parameters are defined, older clients can get away with specifying less and the new RichEdit can assign default values for the new parameters.  Similarly cbRow says how many bytes are defined by the client for TABLEROWPARMS, in case RichEdit is revised to support more row parameters that the client doesn't know about.

To make simple tables easier to define, if fIdenticalCells = 1, lparam points at a single TABLECELLPARMS structure that is valid for all cells in the row.  Note that a nonzero cell border width is guaranteed to give at least a one-pixel border.

The colors are limited to the standard 16 colors defined by

RGB(  0,   0,   0),     // \red0\green0\blue0

RGB(  0,   0, 255),     // \red0\green0\blue255

RGB(  0, 255, 255),     // \red0\green255\blue255

RGB(  0, 255,   0),     // \red0\green255\blue0

RGB(255,   0, 255),     // \red255\green0\blue255

RGB(255,   0,   0),     // \red255\green0\blue0

RGB(255, 255,   0),     // \red255\green255\blue0

RGB(255, 255, 255),     // \red255\green255\blue255

RGB(  0,   0, 128),     // \red0\green0\blue128

RGB(  0, 128, 128),     // \red0\green128\blue128

RGB(  0, 128,   0),     // \red0\green128\blue0

RGB(128,   0, 128),     // \red128\green0\blue128

RGB(128,   0,   0),     // \red128\green0\blue0

RGB(128, 128,   0),     // \red128\green128\blue0

RGB(128, 128, 128),     // \red128\green128\blue128

RGB(192, 192, 192),     // \red192\green192\blue192

plus two custom colors. The border widths are limited to the range 0 to 255 twips.

If the color index is not in the range 1..18, then autocolor is used, which usually ends up being the system Text or Background colors.