The Perl Advent Calendar needs more articles for 2022. Submit your idea today!


Prima::Drawable::Glyphs - helper routines for bi-directional text input and complex scripts output


   use Prima;
   $::application-> begin_paint;
   ‭$::application-> text_shape_out('אפס123', 0,0);



The class implements an abstraction over a set of glyphs that can be rendered to represent text strings. Objects of the class are created and returned from Prima::Drawable::text_shape calls, see more in "text_shape" in Prima::Drawable. A Prima::Drawable::Glyphs object is a blessed array reference that can contain either two, four, or five packed arrays with 16-bit integers, representing, correspondingly, a set of glyph indexes, a set of character indexes, a set of glyph advances, a set of glyph position offsets per glyph, and a font index. Additionally, the class implements several sets of helper routines that aim to address common tasks when displaying glyph-based strings.


Each sub-array is an instance of Prima::array, an effective plain memory structure that provides standard perl interface over a string scalar filled with fixed-width integers.

The following methods provide read-only access to these arrays:


Contains a set of unsigned 16-bit integers where each is a glyph number corresponding to the font that was used for shaping the text. These glyph numbers are only applicable to that font. Zero is usually treated as a default glyph in vector fonts, when shaping cannot map a character; in bitmap fonts this number is usually same as defaultChar.

This array is recognized as a special case when is sent to text_out or get_text_width, that can process it without other arrays. In this case, no special advances and glyph positions are taken into the account though.

Each glyph is not necessarily mapped to a character, and quite often is not, even in english left-to-right texts. F ex character combinations like "ff", "fi", "fl" may be mapped to single ligature glyphs. When right-to-left, RTL, text direction is taken into the account, the glyph positions may change, too. See indexes below that addresses mapping of glyphs to characters.


Contains a set of unsigned 16-bit integers where each is a text offset corresponding to the text was used in shaping. Each glyph position thus points to a first character in the text that maps to the glyph.

There can be more than one character per glyph, such as the above example with a "ff" ligature. There can also be cases with more than one character per more than one glyph, f ex in indic scripts. In these cases it is easier to operate neither by character offsets nor by glyph offsets, but rather by clusters, where each cluster is an individual syntax unit that contains one or more characters per one or more glyphs.

In addition to the text offset, each index value can be flagged with a to::RTL bit, signifying that the character in question has RTL direction. This is not necessarily semitic characters from RTL languages that only have that attribute set; spaces in these languages are normally attributed the RTL bit too, sometimes also numbers. Use of explicit direction control characters from U+20XX block can result in any character being assigned or not assigned the RTL bit.

The array has an extra item added to its end, the length of the text that was used for the shaping. This helps for easy calculation of cluster length in characters, especially of the last one, where the difference between indexes is, basically, the cluster length.

The array is not used for text drawing or calculation, but only for conversion between character, glyph, and cluster coordinates (see Coordinates below).


Contains a set of unsigned 16-bit integers where each is a pixel distance of how much space the corresponding glyph occupies. Where the advances array is not present, or was force-filled by advances options in text_shape, a glyph advance value is basically a sum of a, b, and c widths of the corresponding glyph. However there are cases when depending on shaping input, these values can differ.

One of those cases is the combining graphemes, where the text consisting of two characters, "A" and combining grave accent U+300 should be drawn as a single "À" symbol, and where the font doesn't have that single glyph but rather two individual glyphs "A" and "`". There, where the grave glyph has its own advance for standalone usage, in this case it should be ignored though, and that is achieved by the shaper setting the advance of the "`" to zero.

The array content is respected by text_out and get_text_width, and its content can be changed at will to produce gaps in the text quite easily. F ex Prima::Edit uses that to display tab characters as spaces with 8x advance.


Contains a set of pairs of signed 16-bit integers where each is a X and Y pixel offset for each glyph. Like in the previous example with the "À" symbol, the grave glyph "`" may be positioned differently on the vertical axis in "À" and "à" graphemes, for example.

The array is respected by text_out (but not by get_text_width).


Contains a set of unsigned 16-bit integers where each is an index in the font substitution list (see "font_mapper" in Prima::Drawable). Zero means the current font.

The font substitution is applied by text_shape when polyfont options is set (it is by default), and when the shaper cannot match all fonts. If the current font contains all needed glyphs, this entry is not present at all.

The array is respected by text_out and get_text_width.


In addition to the natural character coordinates, where each index is a text offset that can be directly used in substr perl function, the Prima::Drawable::Glyphs class offers two additional coordinate systems that help abstract the object data for display and navigation.

The glyph coordinate system is a rather straighforward copy of the character coordinate system, where each number is an offset in the glyphs array. Similarly, these offsets can be used to address individual glyphs, indexes, advances, and positions. However these are not easy to use when one needs, for example, to select a grapheme with a mouse, or break set of glyphs in such a way so that a grapheme is not broken. These can be managed easier in the cluster coordinate system.

The cluster coordinates represent a virtually superimposed set of offsets where each corresponds to a set of one or more characters displayed by a one or more glyphs. Most useful functions below operate in this system.


Practically, most useful coordinates that can be used for implementing selection is either character or cluster, but not glyphs. The charater-based selections makes trivial extraction or replacement of the selected text, while the cluster-based makes it easier to manipulate (f ex with Shift- arrow keys) the selection itself.

The class supports both, by operating on selection maps or selection chunks, where each represent same information but in different ways. For example, consider embedded number in a bidi text. For the sake of clarity I'll use latin characters here. Let's have a text scalar containing these characters:


where ABC is right-to-left text, and which, when rendered on screen, should be displayed as


(and index array is (3,4,5,2,1,0) ).

Next, the user clicks the mouse between A and B (in text offset 1), drags the mouse then to the left, and finally stops between characters 2 and 3 (text offset 4). The resulting selection then should not be, as one might naively expect, this:


but this instead:


because the next character after C is 1, and the range of the selected sub-text is from characters 1 to 4.

The class offers to encode such information in a map, i.e. array of integers 1,1,0,1,1,0, where each entry is either 0 or 1 depending on whether the cluster is or is not selected. Alternatively, the same information can be encoded in chunks, or RLE sets, as array 0,2,1,2,1, where the first integer signifies number of non-selected clusters to display, the second - number of selected clusters, the third the non-selected again, etc. If the first character belongs to the selected chunk, the first integer in the result is set to 0.

Bidi input

When sending input to a widget in order to type in text, the otherwise trivial case of figuring out at which position the text should be inserted (or removed, for that matter), becomes interesting when there are characters with mixed direction.

F ex it is indeed trivial, when the latin text is AB, and the cursor is positioned between A and B, to figure out that whenever the user types C, the result should become ACB. Likewise, when the text is RTL and both text and input is arabic, the result is the same. However when f.ex. the text is A1, that is displayed as 1A because of RTL shaping, and the cursor is positioned between 1 (LTR) and A (RTL), it is not clear whether that means the new input should be appended after 1 and become A1C, or after A, and become, correspondingly, AC1.

There is no easy solution for this problem, and different programs approach this differently, and some go as far as to provide two cursors for both directions. The class offers its own solution that uses some primitive heuristics to detect whether cursor belongs to the left or to the right glyph. This is the area that can be enhanced, and any help from native users of RTL languages can be greatly appreciated.



Returns a, b, c metrics from the glyph $INDEX


Read-only accessor to the advances array, see Structure above.


Clones the object

cluster2glyph $FROM, $LENGTH

Maps a range of clusters starting with $FROM with size $LENGTH into the corresponding range of glyphs. Undefined $LENGTH calculates the range from $FROM till the object end.

cluster2index $CLUSTER

Returns character offset of the first character in cluster $CLUSTER.

Note: result may contain to::RTL flag.

cluster2range $CLUSTER

Returns character offset of the first character in cluster $CLUSTER and how many characters are there in the cluster.


Returns array of integers where each is a first character offsets per cluster.


Given a cursor positioned next to the cluster $AT_CLUSTER, runs simple heuristics to see what character offset it corresponds to. $PREFERRED_RTL is used when object data are not enough.

See "Bidi input" above.


Returns d, e, f metrics from the glyph $INDEX


Read-only accessor to the font indexes, see Structure above.

get_box $CANVAS

Return box metrics of the glyph object.

See "get_text_box" in Prima::Drawable.

get_sub $FROM, $LENGTH

Extracts and clones a new object that constains data from cluster offset $FROM, with cluster length $LENGTH.

get_sub_box $CANVAS, $FROM, $LENGTH

Calculate box metrics of a glyph string from the cluster $FROM with size $LENGTH.

get_sub_width $CANVAS, $FROM, $LENGTH

Calculate pixel width of a glyph string from the cluster $FROM with size $LENGTH.


Return width of the glyph objects, with overhangs if requested.

glyph2cluster $GLYPH

Return the cluster that contains $GLYPH.


Read-only accessor to the glyph indexes, see Structure above.


Returns array where each glyph position is set to a number showing how many glyphs the cluster occupies at this position

index2cluster $INDEX

Returns the cluster that contains the character offset $INDEX.


Read-only accessor to the indexes, see Structure above.


Returns array where each glyph position is set to a number showing how many characters the cluster occupies at this position


Umbrella call for justify_interspace if $OPTIONS{letter} or $OPTIONS{word} if set; for justify_arabic if $OPTIONS{kashida} is set; and for justify_tabs if $OPTIONS{tabs} is set.

Returns a boolean flag whether the glyph object was changed or not.

justify_arabic CANVAS, TEXT, WIDTH, %OPTIONS

Performs justifications of arabic TEXT with kashida to the given WIDTH, returns either success flag, or new text with explicit tatweel characters inserted.

   my $text = "\x{6a9}\x{634}\x{6cc}\x{62f}\x{647}";
   my $g = $canvas->text_shape($text) or return;
   $canvas->text_out($g, 10, 50);
   $g->justify_arabic($canvas, $text, 200) or return;
   $canvas->text_out($g, 10, 10);

Inserts tatweels only between arabic letters that did not form any ligatures in the glyph object, max one tatweel set per word (if any). Does not apply the justification if the letters in the word are rendered as LTR due to embedding or explcit shaping options; only does justification on RTL letters. If for some reason newly inserted tatweels do not form a monotonically increasing series after shaping, skips the justifications in that word.

Note: Does not use JSTF font table, on Windows results may be different from native rendering.


If justification is found to be needed, eventual ligatures with newly inserted tatweel glyphs are resolved via a call to text_shape(%OPTIONS) - so any needed shaping options, such as language, may be passed there.

as_text BOOL = 0

If set, returns new text with inserted tatweels, or undef if no justification is possible.

If unset, runs inplace justification on the caller glyph object, and returns the boolean success flag.

min_kashida INTEGER = 0

Specifies minimal width of a kashida strike to be inserted.

kashida_width INTEGER

During the calculation a width of a tatweel glyph is needed - unless supplied by this option, it is calculated dynamically. Also, when called in list context, and succeeded, returns 1, kashida_width that can be reused in subsequent calls.

justify_interspace CANVAS, TEXT, WIDTH, %OPTIONS

Performs inplace inter-letter and/or inter-word justifications of TEXT to the given WIDTH. Returns either a boolean flag whether there were any change made, or, new text with explicit space characters inserted.


as_text BOOL = 0

If set, returns new text with inserted spaces, or undef if no justification is possible.

If unset, runs inplace justification on the caller glyph object, and returns the boolean success flag.

letter BOOL = 1

If set, runs an inter-letter spacing on all glyphs.

max_interletter FLOAT = 1.05

When the inter-letter spacing is applied, it is applied first, and can take up to $OPTIONS{max_interletter} * glyph_width space.

Inter-word spacing does not have such limit, and in worst case, can produce two words moved to the left and to the right edges of the enclosing 0 - WIDTH-1 rectangle.

space_width INTEGER

as_text mode: during the calculation the width of space glyph may be needed - unless supplied by $OPTIONS{space_width}, it is calculated dynamically. Also, when called in list context, and succeeded, returns 1, space_width that can be reused in subsequent calls.

word BOOL = 1

If set, runs an inter-word spacing by extending advances on all space glyphs.

min_text_to_space_ratio FLOAT = 0.75

If word set, does not run inter-word justification if text to space ratio is too small (i e don't spread text too thin )

justify_tabs CANVAS, TEXT, %OPTIONS

Expands tabs as $OPTIONS{tabs} (default:8) spaces.

Needs glyph and the advance of the space glyph to replace the tab glyph. If no $OPTIONS{glyph} and $OPTIONS{width} are specified, calculates them.

Returns a boolean flag whether there were any change made. On success, if called in the list context, returns also space glyph ID and space glyph width for eventual use on the later calls.


First integer from the overhangs result.


Returns a map of integers where each character position corresponds to a glyph position. The name is a rudiment from pure fribidi shaping, where log2vis and vis2log were mapper functions with the same functionality.


Calculates how many clusters the object contains.


Create new object. Not used directly, but rather from inside text_shape calls.

new_array NAME

Creates an array suitable for the object for direct insertion, if manual construction of the object is needed. F ex one may set missing fonts array like this:

   $obj->[ Prima::Drawable::Glyphs::FONTS() ] = $obj->new_array('fonts');
   $obj->fonts->[0] = 1;

The newly created array is filled with zeros.


Creates a new empty object.


Calculates two pixel widths for overhangs in the beginning and in the end of the glyph string. This is used in emulation of a get_text_width call with the to::AddOverhangs flag.


Read-only accessor to the positions array, see Structure above.

reorder_text TEXT

Returns a visual representation of TEXT assuming it was the input of the text_shape call that created the object.


Creates a new object that has all arrays reversed. User for calculation of pixel offset from the right end of a glyph string.


Second integer from the overhangs result.


Converts cluster selection range into text selection range

selection_chunks_clusters, selection_chunks_glyphs $START, $END

Calculates a set of chunks of texts, that, given a text selection from positions $START to $END, represent each either a set of selected and non-selected clusters/glyphs.

selection_diff $OLD, $NEW

Given set of two chunk lists, in format as returned by selection_chunks_clusters or selection_chunks_glyphs, calculates the list of chunks affected by the selection change. Can be used for efficient repaints when the user interactively changes text selection, to redraw only the changed regions.

selection_map_clusters, selection_map_glyphs $START, $END

Same as selection_chunks_XXX, but instead of RLE chunks returns full array for each cluster/glyph, where each entry is a boolean value corresponding to whether that cluster/glyph is to be displayed as selected, or not.

selection_walk $CHUNKS, $FROM, $TO = length, $SUB

Walks the selection chunks array, returned by selection_chunks, between $FROM and $TO clusters/glyphs, and for each chunk calls the provided $SUB->($offset, $length, $selected), where each call contains 2 integers to chunk offset and length, and a boolean flag whether the chunk is selected or not.

Can be also used on a result of selection_diff, in which case $selected flag is irrelevant.

sub_text_out $CANVAS, $FROM, $LENGTH, $X, $Y

Optimized version of $CANVAS->text_out( $self->get_sub($FROM, $LENGTH), $X, $Y ).

sub_text_wrap $CANVAS, $FROM, $LENGTH, $WIDTH, $OPT, $TABS

Optimized version of $CANVAS->text_wrap( $self->get_sub($FROM, $LENGTH), $WIDTH, $OPT, $TABS ). The result is also converted to chunks.


Returns the length of the text that was shaped and that produced the object.

x2cluster $CANVAS, $X, $FROM, $LENGTH

Given sub-cluster from $FROM with size $LENGTH, calculates how many clusters would fit in width $X.


Dumps glyph object content in a readable format.


This section is only there to test proper rendering


Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.

   Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat.
Latin combining

D̍üi̔s͙ a̸u̵t͏eͬ ịr͡u̍r͜e̥ d͎ǒl̋o̻rͫ i̮n̓ r͐e̔p͊rͨe̾h̍e͐n̔ḋe͠r̕i̾t̅ ịn̷ vͅo̖lͦuͦpͧt̪ątͅe̪

   v̰e̷l̳i̯t̽ e̵s̼s̈e̮ ċi̵l͟l͙u͆m͂ d̿o̙lͭo͕r̀e̯ ḛu̅ fͩuͧg̦iͩa̓ť n̜u̼lͩl͠a̒ p̏a̽r̗i͆a͆t̳űr̀

Lorem Ipsum используют потому, что тот обеспечивает более или менее стандартное заполнение шаблона.

   а также реальное распределение букв и пробелов в абзацах

זוהי עובדה מבוססת שדעתו של הקורא תהיה מוסחת על ידי טקטס קריא כאשר הוא יביט בפריסתו.

  המטרה בשימוש ב-Lorem Ipsum הוא שיש לו פחות או יותר תפוצה של אותיות, בניגוד למלל

العديد من برامح النشر المكتبي وبرامح تحرير صفحات الويب تستخدم لوريم إيبسوم بشكل إفتراضي

  كنموذج عن النص، وإذا قمت بإدخال "lorem ipsum" في أي محرك بحث ستظهر العديد من

Lorem Ipsum के अंश कई रूप में उपलब्ध हैं, लेकिन बहुमत को किसी अन्य रूप में परिवर्तन का सामना करना पड़ा है, हास्य डालना या क्रमरहित शब्द ,

  जो तनिक भी विश्वसनीय नहीं लग रहे हो. यदि आप Lorem Ipsum के एक अनुच्छेद का उपयोग करने जा रहे हैं, तो आप को यकीन दिला दें कि पाठ के मध्य में वहाँ कुछ भी शर्मनाक छिपा हुआ नहीं है.


  Lorem Ipsum的目的就是为了保持字母多多少少标准及平

มีหลักฐานที่เป็นข้อเท็จจริงยืนยันมานานแล้ว ว่าเนื้อหาที่อ่านรู้เรื่องนั้นจะไปกวนสมาธิของคนอ่านให้เขวไปจากส่วนที้เป็น Layout เรานำ Lorem Ipsum มาใช้เพราะความที่มันมีการกระจายของตัวอักษรธรรมดาๆ แบบพอประมาณ ซึ่งเอามาใช้แทนการเขียนว่า ‘ตรงนี้เป็นเนื้อหา, ตรงนี้เป็นเนื้อหา' ได้ และยังทำให้มองดูเหมือนกับภาษาอังกฤษที่อ่านได้ปกติ ปัจจุบันมีแพ็กเกจของซอฟท์แวร์การทำสื่อสิ่งพิมพ์ และซอฟท์แวร์การสร้างเว็บเพจ

   กวนสมาธิของคนอ่านให้เขวไปจากส่วนที้เป็น Layout เรานำ Lorem Ipsum

(Note: libthai is required for text wrapping by the word boundary)

Largest well-known grapheme cluster in Unicode



Dmitry Karasik, <>.