Bi-directional Display

[0001] This patent application is a continuation of U.S. patent application Ser. No. 09/838,377, docket number AUS920010277US1, filed on Apr. 19, 2001, by Steven Edward Atkin.

[0002] This invention was not developed in conjunction with any Federally sponsored contract.

[0003] Not applicable.

[0004] Not applicable.

[0005] 1. Field of the Invention

[0006] This patent application is a continuation of U.S. patent application Ser. No. 09/838,377, docket number AUS920010277US1, filed on Apr. 19, 2001, by Steven Edward Atkin, which is now under allowance. This invention relates to the technologies of computer displays and interpretation of file and data for display on a computer. This invention especially relates to the technologies of bi-directional display methods for displaying portions of data which require orientation from left-to-right and from right-to-left to support various international character sets and languages.

[0007] 2. Description of the Related Art

[0008] Prior to the introduction of rich encoding schemes such as Unicode and ISO 10646, most text streams consisted of characters originating from a single script. Traditionally an encoding was comprised of one national script plus a subset of the Latin script (ASCII 7) which fit within the confines of an 8 bit character type. In such an environment, presentation of text is a relatively trivial matter.

[0009] For the most part, the order in which a program stores its characters (logical order) is equivalent to the order in which they are visually presented (display order). Thus, there is a direct correlation between the logical order and display order. Exceptions to this rule include scripts which are written from right to left, such as Arabic, Hebrew, Farsi, Urdu, and Yiddish.

[0010] One existing method to solve this problem is to require computer users, such as computer programmers or web browser users, to enter characters in display order. This is no problem for users of left-to-right languages. However, for users of right-to-left languages, this requires the user to enter the characters and words in “reverse order”. For example, to create a text stream containing Arabic characters, the user must enter them backwards.

[0011] This solution is not elegant, and it becomes cumbersome when right-to-left and left-to-right scripts are intermixed, creating bi-directional scripts.

[0012] Another solution known in the art is to allow users to enter text in logical order, but to require them to use some explicit directional formatting codes within the script, for example, 0x202B and 0x202A in Unicode, for segments of text that run contrary to the base text direction. As this is acceptable in some instances, it has problems in practice, as well. First, it is undefined what a computer should do with the explicit control codes in tasks other than displaying the script. This may cause problems when these formatting codes are received by searching algorithms, or when they are interchanged between systems.

[0013] These explicit formatting codes require specific code points to be set-aside for them, as well. In some encodings, this may be unacceptable due to the fixed number of code points available and the number of code points required to represent the script itself.

[0014] Ideally, a system of encoding mixed direction scripts would maintain the flexibility of entering characters in logical order while still achieving the correct visual appearance and display order. Such algorithms do exist, and are called “implicit layout algorithms”.

[0015] Implicit layout algorithms require no explicit directional codes nor any higher order protocols. These algorithms can automatically determine the correct visual layout by simply examining the logical text stream. Yet in certain cases correct layout of a text stream may still remain ambiguous. Consider the following example in TABLE 1 in which Arabic letters are represented by upper case Latin characters.

TABLE 1
Ambiguous layout
	fred does not believe TAHT YAS SYAWLA I

[0016] In the absence of context, such as a base or paragraph direction, there are two possible ways to display the sentence. When displayed from left to right, it appears as “Fred does not believe I always say that”, and when displayed from right to left, it appears as “I always say that Fred does not believe”. As evident from this example, the two interpretations can represent completely different meanings, and may give no clue whatsoever that there has been an error in the display of the script.

[0017] The Unicode Bi-directional Algorithm rectifies such problems by providing a mechanism for unambiguously determining the visual representation of all raw streams of Unicode text. The algorithm is based upon existing implicit layout algorithms and is supplemented by the addition of explicit directional control codes.

[0018] Generally the Unicode implicit rules are sufficient for the layout of most text streams. However, there are cases in which the Unicode algorithm may give inappropriate or inaccurate display results. For example, a telephone number appearing in a stream of Arabic letters “MY NUMBER IS (321)713-0261.” This should not be rendered as a mathematical expression as show in TABLE 2. As demonstrated, without knowledge of the use of the numbers in this context, the correct display cannot correctly be determined.

TABLE 2
Rendering numbers
	Incorrect display: 0261-713(321) SI REBMUN YM
	Correct display: (321)713-0261 SI REBMUN YM

[0019] Various implementations of the Unicode Bi-directional Algorithm have been proposed in technical reports, such as Unicode Technical Report #9, including “Pretty Good Bidi Algorithm” (PGBA), “Free Implementation of the Bidi Algorithm” (FriBidi)], “IBM Classes for Unicode” (ICU), Java 1.2, Unicode Java Reference, and Unicode C Reference.

[0020] Currently, there exist two reference implementations of the Unicode Bidirectional algorithm, one in Java and the other in C, as well as printed textual descriptions contained in technical reports such as Unicode Technical Report #9.

[0021] Upon our testing of the reference implementations of the Unicode Bidirectional algorithm on a large number of concise and carefully crafted test cases of basic bidirectional text, several problems and ambiguous results are found.

[0022] To simulate Arabic and Hebrew input/output, a simple set of rules can be utilized. These rules make use of characters from the Latin-1 character set. The character mappings allow Latin-1 text to be used instead of real Unicode characters for Arabic, Hebrew, and control codes. This is an enormous convenience in writing, reading, running and printing the test cases. This form is the same as the one used by the Unicode Bidirectional Reference Java Implementation, as shown in TABLE 3.

[0023] Unfortunately not all the implementations adhere to these rules in their test cases. To compensate for this, changes were made to some of the implementations.

TABLE 3
Bidirectional character mappings
	Type	Arabic	Hebrew	Mixed	English
	L	a-z	a-z	a-z	a-z
	AL	A-Z		A-M
	R		A-Z	N-Z
	AN	0-9		5-9
	EN		0-9	0-4	0-9
	LRE	[	[	[	[
	LRO	{	{	{	{
	RLE	]	]	]	]
	RLO	}	}	}	}
	PDF	{circumflex over ( )}	{circumflex over ( )}	{circumflex over ( )}	{circumflex over ( )}
	NSM	˜	˜	˜	˜

[0024] In the Unicode C reference implementation, additional character mapping tables were added to match those of the Unicode Java Reference implementation. Also the bidirectional control codes were remapped from the control range 0x00-0x1F to the printable range 0x20-0x7E. This remapping allowed test results to be compared more easily.

[0025] In PGBA and FriBidi, the character attribute tables were modified to match the character mappings outlined in TABLE 3. However, the strategy we used for evaluation of ICU and Java was slightly different. In the ICU and Java test cases, the character types are used rather than a character mapping. So, in places where our test cases required a specific type, that type was simply used rather than a character mapping.

[0026] The test cases employed are presented in TABLES 4 through 7. The “source” column of each table shows the test case script input and a test case number, and the “expected” column sets forth what the correct display order output should have been.

TABLE 4
Arabic Charmap Tests
Source	Expected
1	car is THE CAR in arabic	car is RAC EHT in arabic
2	CAR IS the car IN ENGLISH	HSILGNE NI the car SI RAC
3	he said “IT IS 123, 456, OK”	he said “KO,456, 123 SI TI”
4	he said “IT IS (123, 456), OK”	he said “KO, (456, 123) SI TI”
5	he said “IT IS 123,456, OK”	he said “KO, 123,456 SI TI”
6	he said “IT IS (123,456), OK”	he said “KO, (123,456) SI TI”
7	HE SAID “it is 123, 456, ok”	“it is 123, 456, ok” DIAS EH
8	<H123>shalom</H123>	<123H/>shalom<123H>
9	HE SAID “it is a car!” AND	NAR DNA “!it is a car” DIAS
	RAN	EH
10	HE SAID “it is a car!x” AND	NAR DNA “it is a car!x” DIAS
	RAN	EH
11	−2 CELSIUS IS COLD	DLOC SI SUISLEC −2
12	SOLVE 1*5 1−5 1/5 1+5	5+1 5/1 5−1 5*1 EVLOS
13	THE RANGE IS 2.5..5	5..2.5 SI EGNAR EHT
14	IOU $10	10$ UOI
15	CHANGE −10%	%10− EGNAHC
16	−10% CHANGE	EGNAHC %10−
17	he said “IT IS A CAR!”	he said “RAC A SI TI!”
18	he said “IT IS A CAR!X”	he said “X!RAC A SI TI”
19	(TEST) abc	abc (TSET)
20	abc (TEST)	abc (TSET)
21	#@$ TEST	TSET $@#
22	TEST 23 ONCE abc	abc ECNO 23 TSET
23	he said “THE VALUES ARE 123,	he said “KO, 789, 456, 123
	456, 789, OK”	ERA SEULAV EHT”.
24	he said “IT IS A bmw 500, OK.”	he said “A SI TI bmw KO,
		500.”

[0027]

TABLE 5
Hebrew Charmap Tests
Source	Expected
1	HE SAID “it is 123, 456, ok”.	”it is 123, 456, ok” DIAS EH
2	<H123>shalom</H123>	<123H/>shalom<123H>
3	<h123>SAALAM</h123>	<h123>MALAAS</h123>
4	−2 CELSIUS IS COLD	DLOC SI SUISLEC −2
5	−10% CHANGE	EGNAHC −10%
6	TEST ˜˜˜23%%% ONCE abc	abc ECNO 23%%%˜˜˜ TSET
7	TEST abc ˜˜˜23%%% ONCE	abc ECNO abc ˜˜˜23%%% TSET
	abc
8	TEST abc@23@cde ONCE	ECNO abc@23@cde TSET
9	TEST abc 23 cde ONCE	ECNO abc 23 cde TSET
10	TEST abc 23 ONCE cde	cde ECNO abc 23 TSET
11	Xa 2 Z	Z a 2X

[0028]

TABLE 6
Mixed Charmap Tests
Source	Expected
1	A˜˜	˜˜A
2	A˜a˜	a˜˜A
3	A1	1A
4	A1	1 A
5	A˜1	1˜A
6	1	1
7	a1	a 1
8	N1	1 N
9	A˜˜ 1	1 ˜˜A
10	A˜a1	a1˜A
11	N1	1N
12	a1	a1
13	A˜N1	1N˜A
14	NOa1	a1ON
15	½	½
16	1,2	1,2
17	5,6	5,6
18	A½	2/1A
19	A1,5	1,5A
20	A1,2	1,2A
21	1,.2	1,.2
22	1,A2	2A,1
23	A5,1	5,1A
24	+$1	+$1
25	1+$	1+$
26	5+1	5+1
27	A+$1	1$+A
28	A1+$	$+1A
29	1+/2	1+/2
30	5+	5+
31	+$	+$
32	N+$1	+$1N
33	+12$	+12$
34	a/1	a/1
35	1,5	1,5
36	+5	+5

[0029]

TABLE 7
Explicit Override Tests
Source	Expected
1	a}}}def	afed
2	a}}}DEF	aFED
3	a}}}defDEF	aFEDfed
4	a}}}DEFdef	afedFED
5	a{{{def	adef
6	a{{{DEF	aDEF
7	a{{{defDEF	adefDEF
8	a{{{DEFdef	aDEFdef
9	A}}}def	fedA
10	A}}}DEF	FEDA
11	A}}}defDEF	FEDfedA
12	A}}}DEFdef	fedFEDA
13	A{{{def	defA
14	A{{{DEF	DEFA
15	A{{{defDEF	defDEFA
16	A{{{DEFdef	DEFdefA
17	{circumflex over ( )}{circumflex over ( )}abc	abc
18	{circumflex over ( )}{circumflex over ( )}}abc	cba
19	}{circumflex over ( )}abc	abc
20	{circumflex over ( )}}{circumflex over ( )}abc	abc
21	}{circumflex over ( )}}abc	cba
22	}{circumflex over ( )}{abc	abc
23	}{circumflex over ( )}{circumflex over ( )}}abc	cba
24	}}abcDEF	FEDcba

[0030] All implementations were tested by using the test cases from TABLES 4 through 6. The implementations that support the Unicode directional control codes (LRO, LRE, RLO, RLE, and PDF) were further tested using the test cases from TABLE 7. At this time, the directional control codes are only supported by ICU, Java 1.2, Unicode Java reference, and Unicode C reference.

[0031] When the results of the test cases were compared, the placement of directional control codes and choice of mirrors was ignored. This is permitted as the final placement of control codes is arbitrary and mirroring may optionally be handled by a higher order protocol.

[0032] TABLES 8-10 detail the test result differences among the implementations with respect to the expected results. Only PGBA, FriBidi and the Unicode C implementations returned results that were different from the expected results; the Unicode Java reference, Java 1.2, and ICU passed all test cases.

	TABLE 8
	a. Arabic Test Differences for PGBA 2.4
	4 he said “KO ,)456 ,123( SI TI”
	6 he said “KO ,)123,456( SI TI”
	12 1+5 1/5 1−5 5*1 EVLOS
	14 $10 UOI
	15 %-10 EGNAHC
	16 EGNAHC %-10
	19 abc )TSET(
	24 he said “A SI TI bmw 500, KO.”
	b. Arabic Test Differences for FriBidi 1.12
	2 SI RAC the car NI ENGLISH
	7 “ok ,456 ,123 it is” DIAS EH
	8 <123H>shalom</123H>
	9 DIAS EH “it is a car!” DNA RAN
	10 DIAS EH “it is a car!x” DNA RAN
	11 -SI SUISLEC 2 COLD
	15 10-EGNAHC%
	16 -10% CHANGE
	19 (TSET) abc
	21 #@$ TEST
	22 ECNO 23 TSET abc
	c. Arabic Test Differences for Unicode C Reference
	7 “ok ,456 ,123 it is” DIAS EH
	11 DLOC SI SUISLEC 2-12

[0033]

TABLE 9
Hebrew Test Differences
	PGBA 2.4	FriBidi 1.12
5	EGNAHC % − 10
6	abc ECNO %%%23˜˜˜TSET
7	abc ECON %%%23˜˜˜abc TSET
11	Z 2 aX	a 2X

[0034]

TABLE 10
Mixed test differences
	PGBA	FriBidi 1.12
1		A˜˜
2	˜a˜A	˜Aa˜
10	1a˜A	˜Aa1
14	1a˜A
18	½A	½A
19		5.1A
21		2, 1
23		1, 5A
27		+$1A
28		1 + $A
32	15N
35		5,1

[0035] In the PGBA reference implementation, types AL and R are treated as being equivalent. This in itself does not present a problem as long as the data stream is free of AL and EN (European number). However, a problem arises when AL is followed by a EN. For example, test case 18 from TABLE 6. In this situation, the ENs should be treated as AN's (Arabic number) and not left as EN's.

[0036] The handling of NSM is also different in PGBA. PGBA treats NSM as being equal to ON (other neutral). This delays the handling of NSM until the neutral type resolution phase rather than in the weak type resolution phase. By delaying their handling, the wrong set of rules are used to resolve the NSM type. For example, in test case 2 from TABLE 6 the last NSM should be treated as type L instead of type R.

[0037] There are a few problems with the FriBidi implementation, as well. Specifically, when an AL is followed by a EN the EN is not being changed to type AN. See test case 18 in TABLE 6. This is the same symptom as was found in PGBA, but the root cause is different. In FriBidi, step W2 (weak processing phase rule two) the wrong type is being examined it should be type EN instead of type N. Additionally, there is a problem in determining the first strong directional character. The only types that are recognized as having a strong direction are types R and L. Type AL should also be recognized as a strong directional character. For example, when test case 1 from TABLE 6 is examined FriBidi incorrectly determines that there are no strong directional characters present. It then proceeds to default the base direction to type L when it should actually be of type R. This problem also causes test cases 2, 9, and 11 from TABLE 4 to fail.

[0038] The greatest hindrance to the creation of a method for converting logical data streams to display streams lies in the problem description. The problem of bidirectional layout is ill defined with respect to the input(s) and output(s).

[0039] Certainly the most obvious input is the data stream itself. Several situations require additional input in order to correctly determine the output stream. For example, in Farsi mathematical expressions are written left to right while in Arabic they are written right to left. This may require a special sub input (directional control code) to appear within stream for proper handling to occur. If it becomes necessary to use control codes for obtaining the desired results the purpose of an algorithm becomes unclear.

[0040] The situation becomes even more cloudy when one considers other possible inputs (paragraph levels, line breaks, shaping, directional overrides, numeric overrides, etc.) Are to be treated as separate inputs? If they are treated as being distinct, when, where and how should they be used? Determining the output(s) is not simple either. The correct output(s) is largely based on the context in which an algorithm be used. If an algorithm is used to render text, then appropriate outputs might be a glyph vector and a set of screen positions. On the other hand, if an algorithm is simply being used determine character reordering, then an acceptable output might just be a reordered character stream.

[0041] The Unicode Bidirectional algorithm has gone through several iterations over the years. The current textual reference been greatly refined. Nevertheless, we believe that there is room for improvement. Implementing a bidirectional layout algorithm is not a trivial matter even when one restricts an implementation to just reordering. Part of the difficulty can be attributed to the textual description of the algorithm. Additionally there are areas that require further clarification.

[0042] As an example consider step L2 of the Unicode Bidirectional Reference Algorithm. It states the following, “From the highest level found in the text to the lowest odd level on each reverse any contiguous sequence of characters that are at level or higher.” This has more than one possible interpretation. It could mean that once the highest level has been found and processed the next level for processing should one less than the current level. It could also be interpreted meaning that the next level to be processed is the next lowest level actually present in the text, which may be greater one less than the current level. It was only through an examination of Unicode's Java implementation that we were to determine the answer.

[0043] There are also problems concerning the bounds of the Uni-code Bidirectional Algorithm. In the absence of higher order protocols it is not always possible to perform all the steps of Unicode Bidirectional Algorithm. In particular, step L4 requires mirrored characters to be depicted by mirrored glyphs their resolved directionality is R. However, glyph selection requires knowledge of fonts and glyph substitution tables. One possible mechanism for avoiding glyph substitutions is to perform mirroring via character substitutions. In this approach mirrored characters are replaced by their corresponding character mirrors. In most situations this approach yields the same results. The only drawback occurs when a mirrored character does not have its corresponding mirror encoded in Unicode. For example, the square root character (U221A) does not have its corresponding mirror encoded.

[0044] Such situations have placed developers in a quandary. One solution is to use the implementations (Java and C) as a reference. But these implementations don't agree in every case. Furthermore the implementations have different goals. The Java implementation follows the textual reference closely while the C implementation offers performance improvements.

[0045] However, if computer source code is to be used as a reference design, then source code that is more attuned to describing these types of methods and algorithms is required. The flexibility, extensibility, and understandability of the imperative language references causes these references to be inadequate.

[0046] For example, using the imperative language reference, it matters what character encoding one uses (UCS4, UCS2, or UTF8). In “C”, the size of types are not guaranteed to be portable, making C unsuitable as a reference. In the Java, reference implementation the ramifications of moving to UCS4 are unclear.

[0047] Therefore, there is a need in the art for a new reference method for bidirectional text script interpretation for display, which avoids the errors in interpretation of the existing references, as well as provides a framework upon which future, improved models may be realized. Preferably, the new method should separate details that are not directly related to the method such that text and character reordering is completely independent from character encoding.

[0048] The following detailed description when taken in conjunction with the figures presented herein provide a complete disclosure of the invention.

[0049] FIG. 1 shows the arrangement of components of a computer system which is capable of executing Haskell programs.

[0050] FIG. 2 illustrates the internal manipulation of Unicode as sequences of 32-bit integers.

[0051] FIG. 3 shows the five phases of the method in the form of a data flow diagram.

[0052] In a first aspect of the present invention, a bidirectional text display method embodied in a functional programming language, rather than an imperative programming language, is provided to solve the problems of the currently available bidirectional display methods. According to the preferred embodiment, the functional language Haskell is used to provide the encoding of the process of the invention. However, it will be recognized by those skilled in the art that alternate functional languages, such as Standard ML (SML), Miranda, Lisp, Scheme, or Erlang, may also be employed to encode the process of the invention.

[0053] In the first step of the method, bidirectional attributes are looked up and assigned to a logical character stream. The attributes are preferably obtained from an online character database.

[0054] Next, through explicit processing, level numbers are assigned, honoring any directional overrides present in the logical character stream. Subsequent weak and neutral type processing potentially causes attribute types to change based upon surrounding attribute types. Then, implicit processing assigns final level numbers to the stream which control reordering. Finally, reordering processing produces a sequence of characters in display order.

[0055] By separating the facets of layout dealing with reordering from those that are concerned with rendering, such as line breaking, glyph selection, and shaping, the Haskell-based method is more discernible and comprehendable, thereby allowing it to be more useful as a model upon which others may base bidirectional implementations.

[0056] The invention is realized in part by a computing platform, such as an IBM-compatible personal computer, Apple MacIntosh™, or other computer hardware platform, running a common operating system such as Linux, UNIX, Microsoft's Windows™, IBM's AIX™ or OS/2™. According to the preferred embodiment, the method is encoded in the functional programming language Haskell, which can be executed by many computing platforms suitably equipped with one of several widely-available Haskell interpreters, or compiled from Haskell to machine-specific executable code.

[0057] Turning to FIG. 1, a generalized organization of such a computer platform (10) is shown. The computer platform (10) has a central processing unit (CPU) (14), a set of device drivers and a basic input/output system (BIOS) (18), and typically an operating system (103), such as those mentioned previously. Most computer platforms, such as a personal computer, are also equipped with disk interfaces (15) and disks; user device I/O (16) to interface to keyboards, pointing devices, and a display; and a network interface card or device (17) allowing communications to a computer network, wireless network, or the Internet. Some computer platforms, such as personal digital assistants, web-enabled telephones, and Internet appliances may not be provided with all of these components, but in general, the functionality of these components is present in some form.

[0058] The computer platform (10) is also typically provided with one or more non-portable, machine-specific application programs (102).

[0059] According to the preferred embodiment, the computer platform is provided with a Haskell interpreter (101), preferably the Hugs 98 interpreter which is freely available from the “HugsOnline” web site for a variety of operating systems and computer platform.

[0060] The remaining disclosure of the invention is presented relative to the computer program implementation of the method for displaying bidirectional text scripts, referred to as Haskell Bidi (HaBi).

[0061] One might ask why implement the Unicode Bidirectional algorithm in a purely functional language, such as Haskell, when so many other implementations already exist? It is the authors contention that a greater understanding of the algorithm is best obtained by a clear functional description of its operations. Without a clear description, implementers may encounter ambiguities that ultimately lead to divergent implementations, contrary the primary goal of the Unicode Bidirectional Algorithm.

[0062] Currently available bidirectional text script display methods (BiDi) are implemented in imperative languages, such as C and Java, instead of a functional language, such as Haskell. The imperative nature of these languages leaves the possibility of special cases and circumstances not being properly handled by the final code, as demonstrated by the testing described in the BACKGROUND OF THE INVENTION.

[0063] Thus, in a first aspect of the present invention, a method of script-to-display interpretation for bidirectional text scripts is implemented in a functional language, preferably Haskell. More specifically, the preferred embodiment uses the Hugs 98 version of Haskell 98 as it is widely available (Linux, Windows, and Macintosh) and easily configurable.

[0064] Since the dominant concern in HaBi is comprehension and readability, the implementation closely follows the textual description as published in the Unicode Technical Report #9, as shown in the data flow diagram of FIG. 3. HaBi is comprised of five phases:

• • [0065] (a) resolution of explicit directional controls (32 and 33);
  • [0066] (b) resolution of weak types (34);
  • [0067] (c) resolution of neutral types (35);
  • [0068] (d) resolution of implicit levels (36 and 37); and
  • [0069] (e) reordering of levels (38).

[0070] Currently, there is no direct support for Unicode in the 98 implementation of Haskell 98. As such, the method treats Unicode lists of 16- or 32-bit integers. The method is divided into two Haskell 98 modules for Unicode manipulation.

[0071] The first module is used to create Unicode (UCS4, UCS2, and UTF-8) strings. The second module determines character types. Additional utility functions convert Haskell strings with optional Unicode character escapes to 16- or 32-bit integer lists.

[0072] A Unicode escape takes the form \uhhhh, analogous to the Java reference implementation. This escape sequence is used for representing code points outside the range 0-00-0x7f. This format was chosen so as to permit easy comparison of results to other implementations.

[0073] Internally, HaBi manipulates Unicode as sequences of 32-bit integers, as shown in FIG. 2. HaBi is prepared to handle surrogates as soon as Unicode assigns them in the future; the only change HaBi would require is an updated character attribute table. It would be more elegant to use the polymorphism of Haskell since the algorithm does not really care about the type of a character only its attribute.

[0074] Each Unicode character has an associated Bidirectional attribute and level number. Again, FIG. 3 shows the general relationship of this information throughout the steps of the method.

[0075] The first step in our implementation is to lookup and assign bidirectional attributes to the logical character stream. The attributes are preferably obtained from the online character database as published in Unicode 3.0.

[0076] At this point, explicit processing assigns level numbers as well as honoring any directional overrides. Weak and neutral processing potentially causes attribute types to change based upon surrounding attribute types. Implicit processing assigns final level numbers to the stream which control reordering. Reordering then produces a sequence of Unicode characters in display order.

[0077] HaBi uses the following three internal types:

• • [0078] (a) type Attributed=(Ucs4, Bidi);
  • [0079] (b) type Level=(Int, Ucs4, Bidi); and
  • [0080] (c) data Run=LL[Level]|LR[Level]|RR[Level]|RL[Level]

[0081] Wherever possible, HaBi treats characters collectively as sequential runs rather than as individual characters. By using one of data type Run's four possible type constructors, characters can then be grouped by level. These four constructors signify the possible combinations of starting and ending run directions. For example, the LL constructor signifies that the start of a run and the end of a run are both left to right. Therefore, runs of LL followed by RL are not created.

[0082] Before the details of the disclosed source code are discussed, it is important to make note of the following concerning HaBi:

• • [0083] (a) the logical text stream is assumed to have already been separated into paragraphs and lines;
  • [0084] (b) directional control codes are removed once processed;
  • [0085] (c) no limit is imposed on the number of allowable embeddings; and
  • [0086] (d) mirroring is accomplished by performing character replacement.

[0087] By separating those facets of layout dealing with reordering from those that are concerned with rendering (line breaking, glyph selection, and shaping), comphrension of the Haskell implementation is more discernible.

[0088] In the Haskell source code provided in TABLE 11, functions are named in such a way so as to correspond to the appropriate section in the Unicode Bidirectional textual reference. For example, the function named “weak” refers to overall weak type resolution. While the function named “w1_—7”, lines 45-71 of TABLE 11, specifically refers to Unicode steps 1 through 7 in weak resolution.

[0089] The function “logicalToDisplay”, lines 150-158 in TABLE 11, is used to convert a stream in logical order to one in display order. First, calls to the functions “explicit” (TABLE 11 lines 37-41), “weak” (lines 73-78), “neutral” (lines 94-99) and “implicit” (lines 114-119) form runs of fully resolved characters.

[0090] Calls to “reorder” (lines 134-140) and “mirror” (lines 142-148) are then applied to the fully resolved runs, which in turn yield a stream in display order. This is discussed in greater detail in the next few paragraphs.

[0091] The function “explicit” breaks the logical text stream into logical runs via calls to “p2_—3” (lines 1-8), “x2_—9” (lines 10-27), and “x10” (lines 29-35). The reference description suggests the use of stacks for keeping track of levels, overrides, and embeddings. In our implementation, stacks are used as well, but they are implicit rather than explicit (function “x2_—9” arguments two, three, and four). The functions “weak”, “neutral”, and “implicit” are then mapped onto each individual run.

[0092] In “weak” steps 1 though 7 (lines 45-71), two pieces of information are carried forward (the second and third arguments of function “w1_—7”) the current directional state and the last character's type. There are cases in the method where a character's direction gets changed but the character's intrinsic type remains unchanged. For example, if a stream contained an AL followed by a EN, the AL would change to type R (step three in weak types resolution). However the last character would need to remain AL so as to cause the EN to change to AN (step two in resolution of weak types). The functions “n1_—2” (lines 80-92) and “i1_—2” (lines 102-112) resolve the neutral and implicit character types respectively.

[0093] Further details of these functions are fairly straight forward. At this point, runs are fully resolved and ready for reordering (function reorder). Reordering occurs in two stages. In the first stage, shown as function “reverse Run” (lines 121-126), a run is either completely reversed or left as is. This decision is based upon whether a run's level is even or odd. If it is odd (right to left), then it is reversed. In the second stage, shown as function “reverse Levels” (lines 128-132), the list of runs are reordered. At first it may not be obvious that the list being folded is not the list of runs, but is the list of levels, highest level to the lowest odd level in the stream. Once reordering is finished, the list of runs are collapsed into a single list of characters in display order.

[0094] All of the test cases discussed previously yield the expected results for the implementation given in TABLE 11, thereby avoiding the problems and inaccuracies of the other tested reference designs.

[0095] In summary, by using a functional language as the basis upon which we provide our bidirectional text display method, we are able to separate details that are not directly related to the algorithm. As such, reordering is completely independent from character encoding.

[0096] It does not matter what character encoding one uses (UCS4, UCS2, or UTF8). The Haskell type system and HaBi character attribute function allows the character encoding to change while not impacting the reordering algorithm, as opposed to other implementations which may find this level of separation difficult to achieve. HaBi presents the steps as simple, easy to understand, functions without side effects. This allows implementers to comprehend the true meaning of each step in the algorithm independently of the others while yet remaining free from language implementation details. Additionally, the creation of test cases is thus more systematic.

[0097] It will be recognized by those skilled in the art that many variations and substitutions may be made to the embodiment described herein without departing from the spirit and scope of the invention. For example, other functional programming methodologies may be adopted, such as use of a specific macro language, or use of alternate suitable operating systems and computer platforms. As such, the scope of this invention should be limited only by the language of the following claims.

TABLE 11
Haskell Source Code for HaBi
	1 -- Rule P2, P3 determine base level of text from the first strong
	2 -- directional character
	3 p2_3 :: [Attributed] -> Int
	4 p2_3 [ ] = 0
	5 p2_3 ((_,L):xs) = 0
	6 p2_3 ((_,AL):xs) = 1
	7 p2_3 ((_,R):xs) = 1
	8 p2_3 (_:xs) = p2_3(xs)
	9
	10 -- Rules X2 - X9
	11 x2_9 :: [Int] -> [Bidi] -> [Bidi] -> [Attributed] -> [Level]
	12 x2_9 _[ ]=[ ]
	13 x2_9 (l:ls) os es ((x,RLE):xs)
	14 = x2_9 ((add l R):l:ls) (N:os) (RLE:es) xs
	15 x2_9 (l:ls) os es ((x,LRE):xs)
	16 = x2_9 ((add l L):l:ls) (N:os) (LRE:es) xs
	17 x2_9 (l:ls) os es ((x,RLO):xs)
	18 = x2_9 ((add l R):l:ls) (R:os) (RLO:es) xs
	19 x2_9 (l:ls) os es ((x,LRO):xs)
	20 = x2_9 ((add l L):l:ls) (L:os) (LRO:es) xs
	21 x2_9 ls os (e:es) ((x,PDF):xs)
	22 | elem e [RLE,LRE,RLO,LRO] = x2_9 (tail ls) (tail os) es xs
	23 x2_9 ls os es ((x,PDF):xs)
	24 = x2_9 ls os es xs
	25 x2_9 ls os es ((x,y):xs)
	26 | (head os) == N = ((head ls),x,y) : x2_9 ls os es xs
	27 | otherwise = ((head ls),x,(head os)) : x2_9 ls os es xs
	28
	29 -- Rule X10 group characters by level
	30 x10 :: (Int, Int) -> [Level] -> Run
	31 x10 (sor,eor) xs
	32 | even sor && even eor = LL xs
	33 | even sor && odd eor = LR xs
	34 | odd sor && even eor = RL xs
	35 | otherwise = RR xs
	36
	37 -- Process explicit characters X1 - X10
	38 explicit :: Int -> [Attributed] -> [Run]
	39 explicit l xs = zipWith x10 (runList levels l l) groups
	40 where levels = (map (\x -> level (head x)) groups)
	41 groups = groupBy levelEql (x2_9 [l][N][ ] xs)
	42
	43
	44
	45 -- Rules W1 - W7
	46 w1_7 :: [Level]->Bidi ->Bidi ->[Level]
	47 w1_7 [ ] — — = [ ]
	48 w1_7 ((x,y,L):xs) — — = (x,y,L):(w1_7 xs L L)
	49 w1_7 ((x,y,R):xs) — — = (x,y,R):(w1_7 xs R R)
	50 w1_7 ((x,y,AL):xs) — — = (x,y,R):(w1_7 xs AL R)
	51 w1_7 ((x,y,AN):xs) dir — = (x,y,AN):(w1_7 xs dir AN)
	52 w1_7 ((x,y,EN):xs) AL — = (x,y,AN):(w1_7 xs AL AN)
	53 w1_7 ((x,y,EN):xs) L — = (x,y,L):(w1_7 xs L EN)
	54 w1_7 ((x,y,EN):xs) dir — = (x,y,EN):(w1_7 xs dir EN)
	55 w1_7 ((x,y,NSM):xs) L N = (x,y,L):(w1_7 xs L L)
	56 w1_7 ((x,y,NSM):xs) R N = (x,y,R):(w1_7 xs R R)
	57 w1_7 ((x,y,NSM):xs) dir last = (x,y,last):(w1_7 xs dir last)
	58 w1_7 ((a,b,ES):(x,y,EN):xs) dir EN =
	59 (a,b,EN):(x,y,EN):(w1_7 xs dir EN)
	60 w1_7 ((a,b,CS):(x,y,EN):xs) dir EN =
	61 (a,b,EN):(x,y,EN):(w1_7 xs dir EN)
	62 w1_7 ((a,b,CS):(x,y,EN):xs) AL AN =
	63 (a,b,AN):(x,y,AN):(w1_7 xs AL AN)
	64 w1_7 ((a,b,CS):(x,y,AN):xs) dir AN =
	65 (a,b,AN):(x,y,AN):(w1_7 xs dir AN)
	66 w1_7 ((x,y,ET):xs) dir EN = (x,y,EN):(w1_7 xs dir EN)
	67 w1_7 ((x,y,z):xs) dir last
	68 | z==ET && findEnd xs ET == EN && dir /= AL
	69 = (x,y,EN):(w1_7 xs dir EN)
	70 | elem z [CS,ES,ET] = (x,y,ON):(w1_7 xs dir ON)
	71 | otherwise = (x,y,z):(w1_7 xs dir z)
	72
	73 -- Process a run of weak characters W1 - W7
	74 weak :: Run -> Run
	75 weak (LL xs) = LL (w1_7 xs L N)
	76 weak (LR xs) = LR (w1_7 xs L N)
	77 weak (RL xs) = RL (w1_7 xs R N)
	78 weak (RR xs) = RR (w1_7 xs R N)
	79
	80 -- Rules N1 - N2
	81 n1_2 :: [[Level]] -> Bidi -> Bidi -> Bidi -> [Level]
	82 n1_2 [ ] — — base = [ ]
	83 n1_2 (x:xs) sor eor base
	84 | isLeft x = x ++ (n1_2 xs L eor base)
	85 | isRight x = x ++ (n1_2 xs R eor base)
	86 | isNeutral x && sor == R && (dir xs eor) == R
	87 = (map (newBidi R) x) ++ (n1_2 xs R eor base)
	88 | isNeutral x && sor == L && (dir xs eor) == L
	89 = (map (newBidi L) x) ++ (n1_2 xs L eor base)
	90 | isNeutral x =
	91 (map (newBidi base) x) ++ (n1_2 xs sor eor base)
	92 | otherwise = x ++ (n1_2 xs sor eor base)
	93
	94 -- Process a run of neutral characters N1 - N2
	95 neutral :: Run -> Run
	96 neutral (LL xs) = LL (n1_2 (groupBy neutralEql xs) L L L)
	97 neutral (LR xs) = LR (n1_2 (groupBy neutralEql xs) L R L)
	98 neutral (RL xs) = RL (n1_2 (groupBy neutralEql xs) R L R)
	99 neutral (RR xs) = RR (n1_2 (groupBy neutralEql xs) R R R)
	100
	101
	102 -- Rule I1, I2
	103 i1_2 :: [[Level]] -> Bidi -> [Level]
	104 i1_2 [ ] — = [ ]
	105 i1_2 ((x:xs):ys) dir
	106 |attrib x ==R && dir ==L
	107 = (map (newLevel 1) (x:xs)) ++ (i1_2 ys L)
	108 | elem (attrib x) [AN,EN] && dir == L
	109 = (map (newLevel 2) (x:xs)) ++ (i1_2 ys L)
	110 | elem (attrib x) [L,AN,EN] && dir == R
	111 = (map (newLevel 1) (x:xs)) ++ (i1_2 ys R)
	112 i1_2 (x:xs) dir = x ++ (i1_2 xs dir)
	113
	114 -- Process a run of implicit characters I1 - I2
	115 implicit :: Run -> Run
	116 implicit (LL xs) = LL (i1_2 (groupBy bidiEql xs) L)
	117 implicit (LR xs) = LR (i1_2 (groupBy bidiEql xs) L)
	118 implicit (RL xs) = RL (i1_2 (groupBy bidiEql xs) R)
	119 implicit (RR xs) = RR (i1_2 (groupBy bidiEql xs) R)
	120
	121 -- If a run is odd (L) then reverse the characters
	122 reverseRun :: [Level] -> [Level]
	123 reverseRun [ ] = [ ]
	124 reverseRun (x:xs)
	125 | even (level x) = x:xs
	126 | otherwise = reverse (x:xs)
	127
	128 reverseLevels :: [[Level]] -> [[Level]] -> Int -> [[Level]]
	129 reverseLevels w [ ] — = w
	130 reverseLevels w (x:xs) a = if (level (head x)) >= a
	131 then reverseLevels (x:w) xs a
	132 else w ++ [x] ++ (reverseLevels [ ] xs a)
	133
	134 -- Rule L2 Reorder
	135 reorder:: [Run] -> Bidi -> [[Level]]
	136 reorder xs base = fold1 (reverseLevels [ ]) runs levels
	137 where
	138 flat = concat (map toLevel xs)
	139 runs = map reverseRun (groupBy levelEql flat)
	140 levels = getLevels runs
	141
	142 -- Rule L4 Mirrors
	143 mirror:: [Level] -> [Level]
	144 mirror [ ] = [ ]
	145 mirror ((x,y,R):xs) = case getMirror y of
	146 Nothing -> (x,y,R):(mirror xs)
	147 Just a -> (x,a,R):(mirror xs)
	148 mirror (x:xs) = x:(mirror xs)
	149
	150 logicalToDisplay :: [Attributed] -> [Ucs4]
	151 logicalToDisplay attribs
	152 =let baseLevel = p2_3 attribs in
	153 let baseDir = (if odd baseLevel then R else L) in
	154 let x = explicit baseLevel attribs in
	155 let w = map weak x in
	156 let n = map neutral w in
	157 let i = map implicit n in
	158 map character (mirror (concat (reorder i baseDir)))