text_widget_internals.txt 14.1 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338
This file documents how GtkTextView works, at least partially.  You
probably want to read the text widget overview in the reference manual
to get an application programmer overview of the public API before
reading this. The overview in the reference manual documents
GtkTextBuffer, GtkTextView, GtkTextMark, etc. from a public API

The BTree

The heart of the text widget is a data structure called GtkTextBTree,
which implements all the hard work of the public GtkTextBuffer object.
The purpose of the btree is to make most operations at least O(log N),
so application programmers can just use whatever API is convenient
without worrying about O(N) performance pitfalls.

The BTree is a tree of paragraphs (newline-terminated lines).  The
leaves of the tree are paragraphs, represented by a GtkTextLine. The
nodes of the tree above the leaves are represented by
GtkTextBTreeNode. The nodes are used to store aggregate data counts,
so we can for example skip 100 paragraphs or 100 characters, without
having to traverse 100 nodes in a list.

You might guess from this that many operations are O(N) where N is the
number of bytes in a paragraph, and you would be right. The text
widget is efficient for huge numbers of paragraphs, but will choke on
extremely long blocks of text without intervening newlines.

("newline" is a slight lie, we also honor \r, \r\n, and some funky
Unicode characters for paragraph breaks. So this means annoyingly that
the paragraph break char may be more than one byte.)

The idea of the btree is something like:

               ------ Node (lines = 6)
              /          Line 0
             /           Line 1
            /            Line 2
           /             Line 3
          /              Line 4
         /               Line 5
 Node (lines = 12)       
          \---------- Node (lines = 6)
                         Line 6
                         Line 7
                         Line 8
                         Line 9
                         Line 10
                         Line 11

In addition to keeping aggregate line counts at each node, we count
characters, and information about the tag toggles appearing below each

Structure of a GtkTextLine

A GtkTextLine contains a single paragraph of text. It should probably
be renamed GtkTextPara someday but ah well.  GtkTextLine is used for 
the leaf nodes of the BTree.

A line is a list of GtkTextLineSegment. Line segments contain the
actual data found in the text buffer.
Here are the types of line segment (see gtktextsegment.h,
gtktextchild.h, etc.):

  Character:         contains a block of UTF-8 text. 

  Mark:              marks a position in the buffer, such as a cursor.

  Tag toggle:        indicates that a tag is toggled on or toggled off at
                     this point. when you apply a tag to a range of
                     text, we add a toggle on at the start of the
                     range, and a toggle off at the end.  (and do any
                     necessary merging with existing toggles, so we
                     always have the minimum number possible)
  Child widget:      stores a child widget that behaves as a single 
                     Unicode character from an editing perspective.
                     (well, stores a list of child widgets, one per 
                     GtkTextView displaying the buffer)

  Image:             stores a GdkPixbuf that behaves as a single 
                     character from an editing perspective.

Each line segment has a "class" which identifies its type, and also
provides some virtual functions for handling that segment.
The functions in the class are:

 - SplitFunc, divides the segment so another segment can be inserted.

 - DeleteFunc, finalizes the segment

 - CleanupFunc, after modifying a line by adding/removing segments, 
   this function is used to try merging segments that can be merged, 
   e.g. two adjacent character segments with no marks or toggles 
   in between.

 - LineChangeFunc, called when a segment moves to a different line;
   according to comments in the code this function may not be needed
 - SegCheckFunc, does sanity-checking when debugging is enabled. 
   Basically equivalent to assert(segment is not broken).

The segment class also contains two data fields:
 - the name of the segment type, used for debugging

 - a boolean flag for whether the segment has right or left 
   gravity. A segment with right gravity ends up on the right of a
   newly-inserted segment that's placed at the same character offset,
   and a segment with left gravity ends up on the left of a
   newly-inserted segment. For example the insertion cursor 
   has right gravity, because as you type new text is inserted, 
   and the cursor ends up on the right.

The segment itself contains contains a header, plus some
variable-length data that depends on the type of the segment. 
The header contains the length of the segment in characters and in
bytes. Some segments have a length of zero. Segments with nonzero
length are referred to as "indexable" and would generally be
user-visible; indexable segments include text, images, and widgets. 
Segments with zero length occupy positions between characters, and
include marks and tag toggles.

The GtkText*Body structs are the type-specific portions of 

Character segments have the actual character data allocated in the
same malloc() block as the GtkTextSegment, to save both malloc()
overhead and the overhead of a pointer to the character data.

Storing and tracking tags in the BTree

A GtkTextTag is an object representing some text attributes.  A tag
can affect zero attributes (for example one used only for internal
application bookkeeping), a single attribute such as "bold", or any
number of attributes (such as large and bold and centered for a
"header" tag).

The tags that can be applied to a given buffer are stored in the
GtkTextTagTable for that buffer. The tag table is just a collection of

The real work of applying/removing tags happens in the function
_gtk_text_btree_tag(). Essentially we remove all tag toggle segments
that affect the tag being applied or removed from the given range;
then we add a toggle-on and a toggle-off segment at either end of the
range; then for any lines we modified, we call the CleanupFunc
routines for the segments, to merge segments that can be merged.

This is complicated somewhat because we keep information about the tag
toggles in the btree, allowing us to locate tagged regions or
add/remove tags in O(log N) instead of O(N) time. Tag information is
stored in "struct Summary" (that's a bad name, it could probably use
renaming). Each BTreeNode has a list of Summary hanging off of it, one
for each tag that's toggled somewhere below the node. The Summary
simply contains a count of tag toggle segments found below the node.

Views of the BTree (GtkTextLayout)

Each BTree has one or more views that display the tree.  Originally
there was some idea that a view could be any object, so there are some
"gpointer view_id" left in the code. However, at some point we decided
that all views had to be a GtkTextLayout and so the btree does assume
that from time to time.

The BTree maintains some per-line and per-node data that is specific 
to each view. The per-line data is in GtkTextLineData and the per-node
data is in another badly-named struct called NodeData (should be
PerViewNodeData or something). The purpose of these is to store:

 - aggregate height, so we can calculate the Y position of each
   paragraph in O(log N) time, and can get the full height 
   of the buffer in O(1) time. The height is per-view since 
   each GtkTextView may have a different size allocation.

 - maximum width (the longest line), so we can calculate the width of
   the entire buffer in O(1) time in order to properly set up the
   horizontal scrollbar.

 - a flag for whether the line is "valid" - valid lines have not been
   modified since we last computed their width and height. Invalid
   lines need to have their width and height recomputed.

At all times, we have a width and height for each view that can be
used. This starts out as 0x0. Lines can be incrementally revalidated,
which causes the width and height of the buffer to grow. So if you
open a new text widget with a lot of text in it, you can watch the
scrollbar adjust as the height is computed in an idle handler.  Lines
whose height has never been computed are taken to have a height of 0.

Iterators (GtkTextIter)

Iterators are fairly complex in order to avoid re-traversing the btree
or a line in the btree each time the iterator is used. That is, they
save a bunch of pointers - to the current segment, the current line,

Two "validity stamps" are kept in the btree that are used to detect
and handle possibly-invalid pointers in iterators. The
"chars_changed_stamp" is incremented whenever a segment with
char_count > 0 (an indexable segment) is added or removed. It is an
application bug if the application uses an iterator with a
chars_changed_stamp different from the current stamp of the BTree.
That is, you can't use an iterator after adding/removing characters.

The "segments_changed_stamp" is incremented any time we change any
segments, and tells outstanding iterators that any pointers to 
GtkTextSegment that they may be holding are now invalid. For example, 
if you are iterating over a character segment, and insert a mark in
the middle of the segment, the character segment will be split in half
and the original segment will be freed. This increments
segments_changed_stamp, causing your iterator to drop its current
segment pointer and count from the beginning of the line again to find 
the new segment.

Iterators also cache some random information such as the current line
number, just because it's free to do so.


If you think of GtkTextBTree as the backend for GtkTextBuffer,
GtkTextLayout is the backend for GtkTextView. GtkTextLayout was also
used for a canvas item at one point, which is why its methods are not
underscore-prefixed and the header gets installed. But GtkTextLayout
is really intended to be private.

The main task of GtkTextLayout is to validate lines (compute their
width and height) by converting the lines to a PangoLayout and using
Pango functions. GtkTextLayout is also used for visual iteration, and
mapping visual locations to logical buffer positions.

Validating a line involves creating the GtkTextLineDisplay for that 
line. To save memory, GtkTextLineDisplay objects are always created
transiently, we don't keep them around.

The layout has three signals:

 - "invalidated" means some line was changed, so GtkTextView 
   needs to install idle handlers to revalidate.

 - "changed" means some lines were validated, so the aggregate
   width/height of the BTree is now different.

 - "allocate_child" means we need to size allocate a 
   child widget

gtk_text_layout_get_line_display() is sort of the "heart" of
GtkTextLayout. This function validates a line. 

Line validation involves:

 - convert any GtkTextTag on the line to PangoAttrList
 - add the preedit string

 - keep track of "visible marks" (the cursor)

A given set of tags is composited to a GtkTextAttributes. (In the Tk
code this was called a "style" and there are still relics of this in
the code, such as "invalidate_cached_style()", that should be cleaned

There's a single-GtkTextAttributes cache, "layout->one_style_cache",
which is used to avoid recomputing the mapping from tags to attributes
for every segment. The one_style_cache is stored in the GtkTextLayout
instead of just a local variable in gtk_text_layout_get_line_display()
so we can use it across multiple lines. Any time we see a segment that
may change the current style (such as a tag toggle), the cache has to
be dropped.

To compute a GtkTextAttributes from the GtkTextTag that apply to a
given segment, the function is _gtk_text_attributes_fill_from_tags(). 
This "mashes" a list of tags into a single set of text attributes. 
If no tags affect a given attribute, a default set of attributes are
used. These defaults sometimes come from widget->style on the
GtkTextView, and sometimes come from a property of the GtkTextView
such as "pixels_above_lines"


Once you get GtkTextLayout and GtkTextBTree the actual GtkTextView 
widget is not that complicated.

The main complexity is the interaction between scrolling and line
validation, which is documented with a long comment in gtktextview.c.

The other thing to know about is just that the text view has "border
windows" on the sides, used to draw line numbers and such; these
scroll along with the main window.

Invisible text

Invisible text doesn't work yet. It is a property that can be set by a
GtkTextTag; so you determine whether text is invisible using the same
mechanism you would use to check whether the text is bold, or orange.

The intended behavior of invisible text is that it should vanish
completely, as if it did not exist. The use-case we were thinking of
was a code editor with function folding, where you can hide all
function bodies. That could be implemented by creating a
"function_body" GtkTextTag and toggling its "invisible" attribute to
hide/show the function bodies.

Lines are normally validated in an idle handler, but as an exception,
lines that are onscreen are always validated synchronously. Thus
invisible text raises the danger that we might have a huge number of
invisible lines "onscreen" - this needs to be handled efficiently.

At one point we were considering making "invisible" a per-paragraph
attribute (meaning the invisibility state of the first character in
the paragraph makes the whole paragraph visible or not
visible). Several existing tag attributes work this way, such as the
margin width. I don't remember why we were going to do this, but it
may have been due to some implementation difficulty that will become
clear if you try implementing invisible text. ;-)

To finish invisible text support, all the cursor navigation
etc. functions (the _display_lines() stuff) will need to skip
invisible text. Also, various functions with _visible in the name,
such as gtk_text_iter_get_visible_text(), have to be audited to be
sure they don't get invisible text. And user operations such as
cut-and-paste need to copy only visible text.