Monday, 20 November 2017
Computer Science departments should be teaching students how to build software for the world ie they should be internationalising their curricula. I am a long time practitioner of internationalised Computer Science teaching. I do though appear to be a solitary voice. Industry, on the other hand, is building software for the world and needs Computer Scientists with Global Skills.
Over the years, the Global Skills I have taught students include:
① Internationalisation and Localisation of software
② Character sets
③ Unicode and Unicode encodings
④ Internationalising websites and building Adaptive Internationalised websites
⑤ Usage of language tags
⑥ Fonts - glyph variants and relationship to Unicode
⑦ Keyboard Mappings and Input Methods
⑧ Internationalised Domain Names
⑨ Unicode Regular Expressions
⑩ Characteristics of English, Chinese, Japanese, Korean, Russian and Arabic languages/scripts
One misconception is that one needs to know multiple (human) languages in order to build software for the world. Not necessary. But one does need to have an understanding of the characteristics of (human) languages/scripts, for example:
• Chinese and Japanese do not have spaces separating words
• Arabic is written right to left
• Korean Hangeul letters, Jamo, are joined into syllabic blocks
If one wants to internationalise Computer Science Curricula the obvious thing to do is to teach students Global Skills, as above.
Let me give an example. If you look at speakerdeck.com/andre_schappo/unicode-regular-expressions you will see that up to and including slide 10, I am using ASCII regex patterns and strings. If you search the internet you will find thousands of regex examples using ASCII only. Now go to slide 11 and suddenly the regex world changes dramatically. In slide 11, I have Chinese and Emoji text. Fast forward to slide 32 and you will see Egyptian Hieroglyphs.
In addition to making students World Ready, internationalising Computer Science Curricula makes for a richer and more interesting subject.
Time for Computer Science Departments to embrace the world by teaching students Global Skills.
Friday, 10 November 2017
Subtitled: html tables with Chinese characteristics
Subtitled: applying Chinese and Korean font metrics to html tables
Using the template font-family: font name ,monospace;, let's try some more Chinese fonts.
Firefox + font PingFang SC produces a perfect square. Chrome + PingFang SC gives a vertical rectangle.
Firefox + Yuanti SC 圆体-简 produces a perfect square. Chrome + Yuanti SC 圆体-简 gives a vertical rectangle.
Firefox + Baoli SC 报隶-简 produces a perfect square. Chrome + Baoli SC 报隶-简 gives a vertical rectangle.
Firefox + Lantinghei SC 兰亭黑-简 produces a perfect square. Chrome + Lantinghei SC 兰亭黑-简 gives a vertical rectangle.
Using font Heiti SC, Chrome gives much better results than Firefox. Firefox is way out. Chrome + Heiti SC produces a near perfect square. Firefox + Heiti gives a horizontal rectangle.
My expectation is that all browsers, when Chinese fonts are specified, should give perfect squares for the table cells as the squared Chinese character is a fundamental characteristic of the Chinese language. From my experimentation, one can see that there are differences between browsers and a perfect square is not always produced.
One of the characteristics of the Korean language script Hangul (also romanised as Hangeul) 한글 is that the individual letters Jamo are formed into squared syllable blocks. Letʼs explore this characteristic with the Korean font Nanum Myeongjo 나눔 명조. Firefox + Nanum Myeongjo produces a near perfect square. Chrome produces a vertical rectangle.
Using font "Noto Sans Korean" results in really awful aspect ratios in both browsers. Chrome + "Noto Sans Korean" produces vertical rectangles. Firefox + "Noto Sans Korean" gives horizontal rectangles.
I reason that there are dependencies between browsers and font metrics. In the "Noto Sans Korean" case, it seems that the 2 browsers have a 90 degree difference in interpretation of this fontʼs metrics.
My favourite combinations, so far, are: Firefox + Hannotate SC 手札体-简, Firefox + PingFang SC 苹方-简, Firefox + Yuanti SC 圆体-简, Firefox + Baoli SC 报隶-简, Lantinghei SC 兰亭黑-简, Firefox + Nanum Myeongjo 나눔 명조 and Chrome +Heiti SC 黑体-简.
〖 thoughts — Should look at more fonts. Are there any other human language scripts that are squared?〗
Here is what my Sudoku board looks like with Firefox + Hannotate SC font. This font contains glyphs for English as well as Chinese. I do like the Hannotate SC font style used for English as well as the Hannotate SC font style used for Chinese. I had a bit of fun with the Emoji as I selected them by Chinese name. Actually, an Emoji Sudoku might be popular. Instead of the numbers 1 thru 9, an Emoji Sudoku would have 9 different Emoji.
Friday, 3 November 2017
Letʼs examine MySQL. Prior to version 5.5, MySQL could only handle a part of Unicode, specifically the BMP (Basic Multilingual Plane). The reason being that itʼs Unicode UTF8 encoding is a 3 byte encoding. In order to access the whole of Unicode, a 4 byte UTF8 encoding is required. MySQL version 5.5 and greater have a 4 byte UTF8 encoding called utf8mb4 and thus can address and store the whole of Unicode.
Unicode has 17 planes, each of which has FFFF codepoints. The 2 most commonly used planes are: Plane 0, the BMP (Basic Multilingual Plane) and Plane 1, the SMP (Supplementary Multilingual Plane). The BMP codepoints range is 0➜FFFF and the SMP range is 10000➜1FFFF. A 3 byte UTF8 encoding can only address the BMP codepoints range 0➜FFFF. Therefore MySQL versions < 5.5 cannot handle SMP characters. More seriously, with MySQL versions < 5.5, if any SMP characters are used then on storage in a database the first encountered SMP character and all subsequent characters are discarded. The discarded characters cannot be recovered. There are still many systems with this problem. Even those systems that have been upgraded to MySQL 5.5 or greater can still have the problem because of associated code and/or tables not yet updated for 4 byte UTF8 encoding.
So, basically, SMP characters break some MySQL setups. I will refer to these breakable MySQL setups as BMP only MySQL. There are solutions and work arounds that can be used to make BMP only MySQL handle the whole of Unicode.
Letʼs first look at Emoji as they are hugely popular. Most Emoji are in the SMP but there are a small number Emoji which have been in Unicode for many a year which are in the BMP. These can be safely used with BMP only MySQL. I have listed those I have found below. There may be more. I have appended to each of these Emoji the single BMP character variation selector FE0F. This variation selector directs a rendering agent to use, if available, an emojified glyph for the Emoji. What these Emoji look like on your device will depend on the fonts your device has and the browser you are using. On my OSX Mac, all these Emoji look really good and they all have emojified glyphs. With my Android phone, only about half of the below Emoji have emojified glyphs.
The SMP contains much more than Emoji. SMP characters include: Cuneiform, Egyptian Hieroglyphs, Byzantine Musical Symbols, Mahjong Tiles, Playing Cards and much more. All these SMP characters will break BMP only MySQL. WRT web technologies we can fix this SMP breaking problem. Rather than encoding these SMP characters as UTF8 we can instead encode them as NCRs (Numeric Character References). NCRs only use ASCII characters and so are BMP only MySQL safe. ASCII is a subset of Unicode and occupies the first 128 codepoints of the BMP. NCRs take the form &#xnumber;. The NCR for the SMP character 🀣 is 🀣, the NCR for the SMP character 𓀬 is 𓀬 and the NCR for the SMP character 🎋 is 🎋.
Wednesday, 1 November 2017
Apple: OSX and iOS
- m10lmac.blogspot.co.uk — Multilingual Mac
Computer Science/IT/ICT Curricula Internationalisation
- groups.google.com/forum/#!forum/computer-science-curriculum-internationalization — I am administrator for this forum
IDNs (Internationalised Domain Names and EAI (Email Address Internationalisation)
- uasg.tech — Universal Acceptance Steering Group - I am a member of their email discussion list
- idnforums.com — IDN Forums - a Domainers forum - I am a member of this forum.
- Get a free
- 电邮.在线 — Chinese 中文 email address or
- ডাটামেল্.ভারত — Bengali email address or
- ડાટામેલ.ભારત — Gujarati email address or
- डाटामेल.भारत — Hindi email address or
- डेटामेल.भारत — Marathi email address or
- ਡਾਟਾਮੇਲ.ਭਾਰਤ — Punjabi email address or
- இந.இந்தியா — Tamil email address or
- datamail.in — English email address
Regular Expressions (regex)
- regular-expressions.info/refunicode.html — Regular Expression Unicode Syntax Reference
- speakerdeck.com/andre_schappo/unicode-regular-expressions — one of my presentations
- twitter.com/r12a — Richard Ishida
- unicode.org — the definitive source for all things Unicode - I am a member of their email discussion list
- babelstone.co.uk/Unicode/unicode.html — unicode, the movie
- w3.org/International — W3C Internationalization (i18n) Activity
- validator.w3.org/i18n-checker/ — i18n checker
- developer.mozilla.org/en-US/docs/Web/CSS/list-style-type — CSS list-style-type property
Wednesday, 25 October 2017
Subtitle: "How I Discovered the Undiscoverable!"
Note: This is a work in progress and will probably take me several weeks to complete
"노팅엄".length returns 3 (3 encoding units)
...and, some non Emoji SMP (Supplementary Multilingual Plane) Unicode characters
"𓀌".length returns 2 (2 encoding units)
"🀤".length returns 2 (2 encoding units)
Before we proceed further I need to give you further information. I can write Chinese on a Computer and Emoji can be selected by Chinese name using OSX Sierra's Simplified Pinyin Input Method. See schappo.blogspot.co.uk/2016/01/emoji-by-name.html
When I want Emoji I sometimes use OSX's Emoji and Symbol Viewer and sometimes select by Chinese name.
Now we come to the random bit. I typed yu in the Simplified Pinyin Input Method and there were 6 different Emoji to choose from. I chose 🌧️ . I had no reason to type yu nor to chose 🌧️ , I was just experimenting. Now we come back to the length property.
"🌧️".length returns 3 (??????????!) [U+1F327 U+FE0F]
"🌦️".length returns 3 (??????????!) [U+1F326 U+FE0F]
It was most definitely not the answer I was expecting. After some 10 minutes investigation I discovered the reason for this unexpected answer. With these two Emoji the variation selector U+FE0F codepoints.net/U+FE0F is being appended thus giving a count of 3. We now have the answer to the length anomaly. But why do some Emoji have the variation selector appended and not others?
Peter Edberg gives this excellent explanation.
This is about characters U+1F327,U+1F326
The variation selector FE0F is *not* unnecessary with these. Looking at unicode.org/Public/emoji/5.0/emoji-data.txt those characters do *not* have the Emoji-Presentation property set, and they do have variation sequences defined.
From unicode.org/reports/tr51/#Emoji_Variation_Selector_Notes, such singleton emoji characters “should have emoji presentation selectors on base characters with Emoji_Presentation=No whenever an emoji presentation is desired”
I stated: I see that U+1F321➜1F32C do not have the Emoji_Presentation property set.
Peter Edberg responded: From unicode.org/emoji/charts-5.0/emoji-versions-sources.html you can see that these characters came into Unicode as a result of their being in the Webdings/Wingdings set, where they had a prior history of being non-emoji text characters. That is why they have Emoji_Presentation=No by default.
Letʼs now examine my bold claim "I discovered the Undiscoverable"
In order to make this discovery there is a set of required knowledge, skills and personality traits. These include:
- A good understanding of Unicode
- The ability to write Chinese using OSX Sierra's Pinyin Simplified Input Method
- Knowing that Emoji can be selected by Chinese name using Sierra's Pinyin Simplified Input Method
- A desire to experiment and explore
Window of opportunity for my discovery — I reason that the window of opportunity for the discovery started when 🌧️ was available. It was added to Unicode version 7.0 in 2014. It would probably have been another year before it became available and integrated into Apple's OSX. I made the discovery on Saturday 21st October 2017. twitter.com/andreschappo/status/921722952504238081 Given this reasoning the window of opportunity for this discovery is approximately 3 years.
Consequences: My first thought was that this anomaly would cause problems with Emoji Domains. Using mothereff.in/punycode, 🌧️.ws (with variation selector) gives the punycode address xn--v86c7044b.ws and 🌧.ws (without variation selector) gives the punycode address xn--kh8h.ws So, these are obviously different addresses. When 🌧️.ws is pasted into the Firefox address bar it needs to convert from the Unicode form 🌧️.ws to the punycode form. The punycode form it uses is xn--kh8h.ws, it is therefore evident that Firefox disregards the variation selector on conversion. Computers and Routers use the punycode form, the Unicode form is used for display to humans.
I realise that Emoji Domains are IDNA2008 disallowed, but, I figure they will be around for a good number of years yet to come.
〖 I will check out BMP and SMP Emoji with IDNA2003, UTS46 and IDNA2008 using unicode.org/cldr/utility/idna.jsp 〗
Why was this my first thought. I am a long time practitioner of internationalised Computer Science teaching and IDNs Internationalised Domain Names (Emoji Domains are a controversial subset of IDNs) are an important part of i18n. I am an active member of the UASG discussion email list uasg.tech. I am also an active member of IDN Forums idnforums.com. I have learned much from the Domainers on IDN Forums. Thanks guys 👋
〖 Cor Blimey - I am getting into Combinatorial Explosion or Infinite Regression or Both😜 〗
...much more to add but that is all for now...
〖 To Do List —
- Calculate Odds against the discovery been made — I will need help on this as my Maths is rubbish
- Check all Emoji available in Sierra to see if any others have a count of 3
- Consequences of 🌧️ counting anomaly
- Check OSX High Sierra to see if the 🌧️ counting anomaly still exists
- CJK variants domain names - need to do some research https://en.wikipedia.org/wiki/Variation_Selectors_(Unicode_block)
Monday, 10 July 2017
The codepoints and the UTF-8 are all written in hexadecimal(hex). The binary bits are an intermediate form for the purposes of encoding and decoding.
We start with the following form which is designed for encoding Unicode codepoints to UTF-8 and decoding UTF-8 to Unicode codepoints.
The first thing we can do is fill in the fixed bits. They are the fixed bits defined by the encoding scheme. I have entered the fixed bits in red to make them distinct from variable bits.
So, how do we determine where the codepoints go on the form. We need to look at the free bits to determine the range of values that can be accommodated.
- 1 byte row - 7 free variable bits giving a range of 0 ➔ 7F
- 2 byte row - 11 free variable bits giving a range of 80 ➔ 7FF
- 3 byte row - 16 free variable bits giving a range of 800 ➔ FFFF
- 4 byte row - 21 free variable bits giving a range of 10000 ➔ 1FFFFF (the actual maximum value of a codepoint is 10FFFF)
It is a Unicode convention, when writing codepoints, to use a minimum of four hex digits. So for codepoints <1000, one should left pad with zeroes. Hence my entries U+0444 and U+0057 rather than U+444 and U+57.
Sunday, 2 July 2017
So, you have just written some Cool Code which will search for and find occurrences of specified text strings. You have access to Big Data text eg all the text in all public webpages. You will,of course, want to test your Cool Code. Letʼs perform some, seemingly, very simple tests.
Letʼs search for the word 'Scorpion'. Your code works just fine and hence finds all occurrences of the word 'Scorpion'.
Now test with the following two words.
Your Cool Code works fine as all I have done is applied some CSS styling, thus giving each of the two words differing appearance.
Now test you Cool Code with the following two words.
If you have only programmed for ASCII text then your now not so Cool Code will fail. These two words have differing appearance because they are not made up of the ASCII characters you are familiar with. These words use characters from the Unicode Math Alphanumeric Symbols block, U+1D400-1D4FF.
Should the Math Alphanumeric Symbols Scorpion be treated the same as the ASCII Scorpion wrt the search results of your code? In this context I think "Yes", most definitely. A person reading this blog, for example, will just perceive the word Scorpion whatever characters are used to write the word. The reader may well also visualise the insect with a "sting in the tail"😱
What of current working practice?
With twitter, a user has no means of changing text style within a tweet. It has thus become common to use Unicode Math Alphanumeric Symbols to change appearance. I could, for example, use Unicode Math Alphanumeric Symbols to emphasise a word (eg Scorpion) or phrase within a tweet. The meaning of the tweet remains the same.
Google returns the same number of search results whichever of the above forms of Scorpion I use. At time of writing this is "About 144,000,000 results". I deduce Google is treating ASCII Scorpion and Unicode Math Alphanumeric Symbols 𝑆𝑐𝑜𝑟𝑝𝑖𝑜𝑛 & 𝐒𝐜𝐨𝐫𝐩𝐢𝐨𝐧 as equivalent.
Sogou 搜狗 is a Chinese search engine. Using Sogou: ASCII Scorpion returns 93,341 results, Math Alphanumeric Symbols 𝑆𝑐𝑜𝑟𝑝𝑖𝑜𝑛 returns 4,738, Math Alphanumeric Symbols 𝐒𝐜𝐨𝐫𝐩𝐢𝐨𝐧 returns 61. I think it evident that Sogou does not treat my three forms of Scorpion as equivalent.
I side with Google on this.
Here is a taster of what is happening in the behind the scenes technicalities of Unicode. Letʼs take just one of the Unicode Math Alphanumeric Symbols I have used, 𝐒 MATHEMATICAL BOLD CAPITAL S U+1D412. If you visit codepoints.net/U+1D412 you will see a wealth of information about this character. Of relevance to this blog is the Decomposition Mapping which is to the, oh so familiar, ASCII uppercase capital S. This Unicode information can be used to compute string equivalents which can then be used for search thus providing all relevant results.
The moral of this "Sting in the Tale" is: If you do not already know it, you must learn Unicode, it is essential.