C++ Logo

sg16

Advanced search

Re: LWG ISSUE: Format's width estimation is too approximate and not forward compatible.

From: Corentin <corentin.jabot_at_[hidden]>
Date: Sun, 18 Sep 2022 17:55:38 +0200
Let's see
python3 -c 'print("|{}|\n|a|".format(chr(0x2E9A)))'

Konsole:
[image: image.png]
Windows terminal
[image: image.png]
Mac
[image: image.png]
xterm
[image: image.png]


On Sun, Sep 18, 2022 at 4:38 PM Victor Zverovich <victor.zverovich_at_[hidden]>
wrote:

> Yes and the proposed resolution changes handling of such cases in an
> incompatible way that goes against the original intent and makes width
> estimation less useful. This is a design change.
>
> - Victor
>
> On Sun, Sep 18, 2022 at 7:20 AM Corentin <corentin.jabot_at_[hidden]> wrote:
>
>>
>>
>> On Sun, Sep 18, 2022, 16:09 Victor Zverovich <victor.zverovich_at_[hidden]>
>> wrote:
>>
>>> Even a cursory look at the issue shows a number of undesirable changes
>>> that go against the intent of P1868R1. For example,
>>>
>>> U+2E9A
>>>
>>
>> This codepoint does not appear to be assigned.
>>
>>>
>>> has display width of 2 on common terminals while the proposed resolution
>>> changes it to 1.
>>>
>>> The part "it seems questionable to make an exception for those without
>>> input from Unicode" is itself a questionable design change.
>>>
>>> - Victor
>>>
>>> On Thu, Sep 15, 2022 at 10:12 AM Corentin via SG16 <
>>> sg16_at_[hidden]> wrote:
>>>
>>>>
>>>>
>>>> On Thu, Sep 15, 2022 at 5:59 PM Tom Honermann <tom_at_[hidden]>
>>>> wrote:
>>>>
>>>>> Thank you for looking into this, Corentin. It would be great to have
>>>>> some kind of principled approach to how estimated widths are assigned.
>>>>>
>>>>> I don't think this should be handled as an LWG issue though. This is a
>>>>> design change, so should go through LEWG (and probably SG16 as well).
>>>>>
>>>>> I think it would be helpful to have a paper that includes screenshots
>>>>> (of select characters, not all 8500+ characters!) to demonstrate how common
>>>>> terminals display them today. The goal being to ensure that they are
>>>>> handled consistently across platforms before we standardize them.
>>>>>
>>>>> A paper would also present the opportunity to attribute an estimated
>>>>> width of 0 (or a negative width!) to some characters. Perhaps it is worth
>>>>> asking whether an estimated width of 1 really makes sense for CR, LF, NL,
>>>>> VT, etc... It looks like these changes still don't cover edge cases like ﷽
>>>>> (U+FDFD ARABIC LIGATURE BISMILLAH AR-RAHMAN AR-RAHEEM).
>>>>>
>>>>
>>>> I'm not trying to change the design, nor am I trying to make case-by
>>>> case decisions, the original paper covered that.
>>>> I think these are questions worth asking, sure, at least for control
>>>> characters as, for an arbitrary non-cjk, non-emoji neither an information
>>>> nor a specification exist.
>>>> Note that however it's not straightforward
>>>> * There is no spec so it's handled manually
>>>> https://github.com/termux/wcwidth/blob/master/wcwidth.c#L497
>>>> * C1 control characters may be mapped differently on EBCDIC
>>>> platforms
>>>> * Combining characters not preceded by a grapheme starter are in
>>>> fact rendered
>>>> * We already ignore non leading codepoints in graphemes
>>>> * Negative width would be novel
>>>>
>>>> I do not intend to explore these further but I'd be happy to review
>>>> such work :)
>>>>
>>>> However, the original paper intended for CJK characters and emojis to
>>>> be covered, Unicode 14 and 15 added a number of such characters, I
>>>> considered this a bug.
>>>> But mostly, I consider specifying anything Unicode related in terms of
>>>> partially-assigned ranges to be a bug reminiscent of the identifiers of old.
>>>>
>>>> Now, I expect SG16 will deal with this issue, but all the information
>>>> is in the issue.
>>>> The only design-ish question is whether we want to grandfather U+3248
>>>> CIRCLED NUMBER TEN ON BLACK SQUARE..U+324F CIRCLED NUMBER EIGHTY ON BLACK
>>>> SQUARE
>>>> for backward compat reasons.
>>>>
>>>> Codepoints which would go from 1 to 2
>>>> https://gist.github.com/cor3ntin/e5731f77574b146d806e39283e8c7cb7
>>>> Screenshot of my terminal
>>>> [image: image.png]
>>>>
>>>> Tom.
>>>>> On 9/15/22 5:02 AM, Corentin via SG16 wrote:
>>>>>
>>>>>
>>>>> Affected clause [format.string.std]
>>>>>
>>>>> Issue:
>>>>> For the purpose of width estimation, format considers ranges of
>>>>> codepoints initially derived from an implementation of wcwidth with
>>>>> modifications.
>>>>> (See https://wg21.link/p1868r1)
>>>>>
>>>>> This however present a number of challenges:
>>>>>
>>>>> - From a reading of the spec, it is not clear how these ranges
>>>>> were selected.
>>>>> - Poor forward compatibility with future Unicode versions. The
>>>>> list will become less and less meaningful overtime or require active
>>>>> maintenance at each Unicode release (which we have not done for Unicode 14
>>>>> already)
>>>>> - Some of these codepoints are unassigned or otherwise reserved,
>>>>> which is another forward compatibility concern.
>>>>>
>>>>> Instead, we propose to
>>>>>
>>>>> - Rely on UAX-11 https://www.unicode.org/reports/tr11/ for most of
>>>>> the codepoints)
>>>>> - Grand-father specific and fully assigned, blocks of codepoints
>>>>> to support additional pictograms per the original intent of the paper and
>>>>> existing practices. We add the name of these blocks in the wording for
>>>>> clarity.
>>>>>
>>>>> Note that per UAX-11
>>>>>
>>>>> - Most emojis are considered East_Asian_Width="W"
>>>>> - By design, East_Asian_Width="W" includes specific unassigned
>>>>> ranges, which should always be treated as Wide.
>>>>>
>>>>> This change:
>>>>>
>>>>> - Considers 8477 extra codepoints as having a width 2 (as of
>>>>> Unicode 15) (mostly Tangut Ideographs)
>>>>> - Change the width of 85 unassigned code points from 2 to 1
>>>>> - Change the width of 8 codepoints (in the range U+3248 CIRCLED
>>>>> NUMBER TEN ON BLACK SQUARE ..U+324F CIRCLED NUMBER EIGHTY ON BLACK SQUARE)
>>>>> from 2 to 1, because it seems questionable to make an exception for those
>>>>> without input from Unicode
>>>>>
>>>>>
>>>>> Proposed wording:
>>>>>
>>>>> Modify [format.string.std]/p12
>>>>>
>>>>> For a string in a Unicode encoding, implementations should estimate
>>>>> the width of a string as the sum of estimated widths of the first code
>>>>> points in its extended grapheme clusters. The extended grapheme clusters of
>>>>> a string are defined by UAX #29. The estimated width of the following code
>>>>> points is 2:
>>>>>
>>>>> - U+1100 – U+115F
>>>>> - U+2329 – U+232A
>>>>> - U+2E80 – U+303E
>>>>> - U+3040 – U+A4CF
>>>>> - U+AC00 – U+D7A3
>>>>> - U+F900 – U+FAFF
>>>>> - U+FE10 – U+FE19
>>>>> - U+FE30 – U+FE6F
>>>>> - U+FF00 – U+FF60
>>>>> - U+FFE0 – U+FFE6
>>>>> - U+1F300 – U+1F64F
>>>>> - U+1F900 – U+1F9FF
>>>>> - U+20000 – U+2FFFD
>>>>> - U+30000 – U+3FFFD
>>>>> - *Any codepoint with the East_Asian_Width="W"
>>>>> or East_Asian_Width="F" Derived Extracted Property as described by UAX #44*
>>>>> -
>>>>> *U+4DC0 - U+4DFF (Yijing Hexagram Symbols) *
>>>>> - *U+1F300 - U+1F5FF (Miscellaneous Symbols and Pictographs)*
>>>>> - *U+1F900 - U+1F9FF (Supplemental Symbols and Pictographs)*
>>>>>
>>>>> The estimated width of other code points is 1.
>>>>>
>>>>>
>>>>> ---- END OF WORDING ----
>>>>>
>>>>> Annex: Differences introduced by this change
>>>>>
>>>>> // Missing from the standard (count: 8477)
>>>>> // Used to be of width 1, becomes 2
>>>>>
>>>>> U+231A WATCH ..U+231B HOURGLASS
>>>>> U+23E9 BLACK RIGHT-POINTING DOUBLE TRIANGLE ..U+23EC BLACK
>>>>> DOWN-POINTING DOUBLE TRIANGLE
>>>>> U+23F0 ALARM CLOCK
>>>>> U+23F3 HOURGLASS WITH FLOWING SAND
>>>>> U+25FD WHITE MEDIUM SMALL SQUARE ..U+25FE BLACK MEDIUM SMALL SQUARE
>>>>> U+2614 UMBRELLA WITH RAIN DROPS ..U+2615 HOT BEVERAGE
>>>>> U+2648 ARIES ..U+2653 PISCES
>>>>> U+267F WHEELCHAIR SYMBOL
>>>>> U+2693 ANCHOR
>>>>> U+26A1 HIGH VOLTAGE SIGN
>>>>> U+26AA MEDIUM WHITE CIRCLE ..U+26AB MEDIUM BLACK CIRCLE
>>>>> U+26BD SOCCER BALL ..U+26BE BASEBALL
>>>>> U+26C4 SNOWMAN WITHOUT SNOW ..U+26C5 SUN BEHIND CLOUD
>>>>> U+26CE OPHIUCHUS
>>>>> U+26D4 NO ENTRY
>>>>> U+26EA CHURCH
>>>>> U+26F2 FOUNTAIN ..U+26F3 FLAG IN HOLE
>>>>> U+26F5 SAILBOAT
>>>>> U+26FA TENT
>>>>> U+26FD FUEL PUMP
>>>>> U+2705 WHITE HEAVY CHECK MARK
>>>>> U+270A RAISED FIST ..U+270B RAISED HAND
>>>>> U+2728 SPARKLES
>>>>> U+274C CROSS MARK
>>>>> U+274E NEGATIVE SQUARED CROSS MARK
>>>>> U+2753 BLACK QUESTION MARK ORNAMENT ..U+2755 WHITE EXCLAMATION MARK
>>>>> ORNAMENT
>>>>> U+2757 HEAVY EXCLAMATION MARK SYMBOL
>>>>> U+2795 HEAVY PLUS SIGN ..U+2797 HEAVY DIVISION SIGN
>>>>> U+27B0 CURLY LOOP
>>>>> U+27BF DOUBLE CURLY LOOP
>>>>> U+2B1B BLACK LARGE SQUARE ..U+2B1C WHITE LARGE SQUARE
>>>>> U+2B50 WHITE MEDIUM STAR
>>>>> U+2B55 HEAVY LARGE CIRCLE
>>>>> U+A960 HANGUL CHOSEONG TIKEUT-MIEUM ..U+A97C HANGUL CHOSEONG
>>>>> SSANGYEORINHIEUH
>>>>> U+16FE0 TANGUT ITERATION MARK ..U+16FE4 KHITAN SMALL SCRIPT FILLER
>>>>> U+16FF0 VIETNAMESE ALTERNATE READING MARK CA ..U+16FF1 VIETNAMESE
>>>>> ALTERNATE READING MARK NHAY
>>>>> U+17000 TANGUT IDEOGRAPH-# ..U+187F7 TANGUT IDEOGRAPH-#
>>>>> U+18800 TANGUT COMPONENT-001 ..U+18CD5 KHITAN SMALL SCRIPT CHARACTER-#
>>>>> U+18D00 TANGUT IDEOGRAPH-# ..U+18D08 TANGUT IDEOGRAPH-#
>>>>> U+1AFF0 KATAKANA LETTER MINNAN TONE-2 ..U+1AFF3 KATAKANA LETTER MINNAN
>>>>> TONE-5
>>>>> U+1AFF5 KATAKANA LETTER MINNAN TONE-7 ..U+1AFFB KATAKANA LETTER MINNAN
>>>>> NASALIZED TONE-5
>>>>> U+1AFFD KATAKANA LETTER MINNAN NASALIZED TONE-7 ..U+1AFFE KATAKANA
>>>>> LETTER MINNAN NASALIZED TONE-8
>>>>> U+1B000 KATAKANA LETTER ARCHAIC E ..U+1B122 KATAKANA LETTER ARCHAIC WU
>>>>> U+1B132 HIRAGANA LETTER SMALL KO
>>>>> U+1B150 HIRAGANA LETTER SMALL WI ..U+1B152 HIRAGANA LETTER SMALL WO
>>>>> U+1B155 KATAKANA LETTER SMALL KO
>>>>> U+1B164 KATAKANA LETTER SMALL WI ..U+1B167 KATAKANA LETTER SMALL N
>>>>> U+1B170 NUSHU CHARACTER-# ..U+1B2FB NUSHU CHARACTER-#
>>>>> U+1F004 MAHJONG TILE RED DRAGON
>>>>> U+1F0CF PLAYING CARD BLACK JOKER
>>>>> U+1F18E NEGATIVE SQUARED AB
>>>>> U+1F191 SQUARED CL ..U+1F19A SQUARED VS
>>>>> U+1F200 SQUARE HIRAGANA HOKA ..U+1F202 SQUARED KATAKANA SA
>>>>> U+1F210 SQUARED CJK UNIFIED IDEOGRAPH-624B ..U+1F23B SQUARED CJK
>>>>> UNIFIED IDEOGRAPH-914D
>>>>> U+1F240 TORTOISE SHELL BRACKETED CJK UNIFIED IDEOGRAPH-672C ..U+1F248
>>>>> TORTOISE SHELL BRACKETED CJK UNIFIED IDEOGRAPH-6557
>>>>> U+1F250 CIRCLED IDEOGRAPH ADVANTAGE ..U+1F251 CIRCLED IDEOGRAPH ACCEPT
>>>>> U+1F260 ROUNDED SYMBOL FOR FU ..U+1F265 ROUNDED SYMBOL FOR CAI
>>>>> U+1F680 ROCKET ..U+1F6C5 LEFT LUGGAGE
>>>>> U+1F6CC SLEEPING ACCOMMODATION
>>>>> U+1F6D0 PLACE OF WORSHIP ..U+1F6D2 SHOPPING TROLLEY
>>>>> U+1F6D5 HINDU TEMPLE ..U+1F6D7 ELEVATOR
>>>>> U+1F6DC WIRELESS ..U+1F6DF RING BUOY
>>>>> U+1F6EB AIRPLANE DEPARTURE ..U+1F6EC AIRPLANE ARRIVING
>>>>> U+1F6F4 SCOOTER ..U+1F6FC ROLLER SKATE
>>>>> U+1F7E0 LARGE ORANGE CIRCLE ..U+1F7EB LARGE BROWN SQUARE
>>>>> U+1F7F0 HEAVY EQUALS SIGN
>>>>> U+1FA70 BALLET SHOES ..U+1FA7C CRUTCH
>>>>> U+1FA80 YO-YO ..U+1FA88 FLUTE
>>>>> U+1FA90 RINGED PLANET ..U+1FABD WING
>>>>> U+1FABF GOOSE ..U+1FAC5 PERSON WITH CROWN
>>>>> U+1FACE MOOSE ..U+1FADB PEA POD
>>>>> U+1FAE0 MELTING FACE ..U+1FAE8 SHAKING FACE
>>>>> U+1FAF0 HAND WITH INDEX FINGER AND THUMB CROSSED ..U+1FAF8 RIGHTWARDS
>>>>> PUSHING HAND
>>>>>
>>>>> // Reserved (count: 85) - Become used to be of width 2, becomes 1
>>>>>
>>>>> U+2E9A
>>>>> U+2EF4 ..U+2EFF
>>>>> U+2FD6 ..U+2FEF
>>>>> U+2FFC ..U+2FFF
>>>>> U+3040
>>>>> U+3097 ..U+3098
>>>>> U+3100 ..U+3104
>>>>> U+3130
>>>>> U+318F
>>>>> U+31E4 ..U+31EF
>>>>> U+321F
>>>>> U+A48D ..U+A48F
>>>>> U+A4C7 ..U+A4CF
>>>>> U+FE53
>>>>> U+FE67
>>>>> U+FE6C ..U+FE6F
>>>>> U+FF00
>>>>>
>>>>> // Used to be of width 2, becomes 1
>>>>> U+3248 CIRCLED NUMBER TEN ON BLACK SQUARE ..U+324F CIRCLED NUMBER
>>>>> EIGHTY ON BLACK SQUARE
>>>>>
>>>>>
>>>>>
>>>>> --
>>>> SG16 mailing list
>>>> SG16_at_[hidden]
>>>> https://lists.isocpp.org/mailman/listinfo.cgi/sg16
>>>>
>>>

Received on 2022-09-18 15:55:53