Releases/FeatureDictionary

Fix the dictionary proliferation problem

Summary

Fix the proliferation of dictionaries in the OS.

Owners

• Names: CaolanMcNamara, ChristopherAillon, JeremyKatz, BillNottingham

Current status

• Targeted release: Fedora 9
• Last modified: 2008-04-07
• Percentage of completion: 100%
• This is complete, all major applications and default GNOME/KDE spell checking now goes through hunspell. All that remains is to package dictionaries for the lesser used languages where there hasn't already been a sufficiently vibrant fedora-using language community that has taken up packaging a dictionary for their language.

Usage cases/rationale

We have separate dictionaries for each language for OpenOffice.org, Firefox, Thunderbird, and aspell (which gnome and KDE use). This is dumb.

Benefit to Fedora

We get code reuse, a smaller distribution, and a decreased memory footprint.

Scope

Requires changing the OpenOffice.org, thunderbird, firefox, and dictionary packages.

Test Plan

Test spell checking in all apps.

Dependencies

None.

Details

1. Split out hunspell from OpenOffice.org - rhbz#214764 complete
2. Make OpenOffice.org use it - rhbz#214764 complete
3. Split out the dictionaries into separate packages - rhbz#218769 (english) complete
4. Make OpenOffice.org use system dictionaries - complete
5. Make gedit/xchat use it, i.e. enchant. enchant by default already generally prefers using hunspell over aspell, just needs to be told where the dictionaries are - complete
6. Make evolution use it, i.e. gnome-spell. gnome-spell can be patched to use enchant to achieve this - rhbz#426347 complete
7. Make tomboy/pidgin use it, i.e. gtkspell. Same story as gnome-spell - rhbz#245888 complete
8. Make Firefox (and other gecko apps) use it - rhbz#218762 complete, upstream state is now resolved
9. Make KDE use enchant and/or hunspell - complete - KDE 4 already defaults to enchant in Sonnet. (For !K3Spell, see "legacy KSpell" below.) The aspell backend was dropped entirely in Rawhide. For kdelibs3:
   • The legacy KSpell uses command-line spellcheckers. Kevin Kofler wrote a patch to support hunspell, and kde-settings in Rawhide was changed to make it the default.
   • The newer KSpell2 API is plugin-based and uses libraries. It is what KDE 4's Sonnet is based on. Kevin Kofler backported Sonnet's enchant backend. The aspell and ispell backends were dropped in Rawhide.

   See the fedora-devel-list message.

10. Remove copy of hunspell from enchant - rhbz#426402 complete
11. Remove copy of hunspell from xulrunner complete
12. Split enchant to have a separate enchant-aspell rpm to enable optionally removing the aspell support - rhbz#426402 complete
13. Prefer hunspell over aspell as the default for install in comps. See table below for mis-match in language support. rhbz#439037 complete
14. Repackage/replace the aspell dictionaries with hunspell dictionaries 80% see table below for language support

Optional

1. Write an aspell compatibility layer so aspell apps can use the same dictionaries no volunteer -> deferred, is this neccessary at all ? All major desktop apps work now out of the box
2. Make vim use hunspell - rhbz#219777 patch available, not necessary if vim continues to not use any spell-checking, but preferred over introducing built-in vim spellchecker which has yet another format which hunspell dicts are converted to for use

Dictionaries

1. Language Support Matrix

Language Code	Language	aspell	hunspell	notes
aa	Afar
af	Afrikaans	aspell-af	hunspell-af
am	Amharic		available	And one for non-commercial use
an	Aragonese
ar	Arabic	aspell-ar	hunspell-ar
as	Assamese
ast	Asturian
az	Azeri		available
be	Belarusian		available
ber	Amazigh
bg	Bulgarian	aspell-bg	hunspell-bg
bn	Bengali	aspell-bn	hunspell-bn
bo	Tibetan
br	Breton	aspell-br
bs	Bosnian
byn	Blin
ca	Catalan	aspell-ca	hunspell-ca
crh	Crimean
cs	Czech	aspell-cs	hunspell-cs
csb	Kashubian		available
cy	Welsh	aspell-cy	hunspell-cy
da	Danish	aspell-da	hunspell-da
de	German	aspell-de	hunspell-de
dz	Bhutanese
el	Greek	aspell-el	hunspell-el
en	English	aspell-en	hunspell-en
es	Spanish	aspell-es	hunspell-es
et	Estonian		hunspell-ee
eu	Basque		hunspell-eu
fa	Farsi		available
fi	Finnish			Finnish Community has a parallel voikko solution. With an enchant backend, an OpenOffice.org extension, and a Firefox extension)
fil	Filipino			Filipino is the official Tagalog dialect, see tl entry for Tagalog dictionary
fo	Faeroese	aspell-fo	hunspell-fo
fr	French	aspell-fr	hunspell-fr
fur	Furlan		available
fy	Frisian		available
ga	Irish	aspell-ga	hunspell-ga
gd	Scots Gaelic	aspell-gd	hunspell-gd
gez	Ge'ez
gl	Galician	aspell-gl	hunspell-gl
gu	Gujarati	aspell-gu	hunspell-gu
gv	Manx
ha	Hausa
he	Hebrew	aspell-he	hunspell-he
hi	Hindi	aspell-hi	hunspell-hi
hr	Croatian	aspell-hr	hunspell-hr
hsb	Upper Sorbian
hu	Hungarian		hunspell-hu
hy	Armenian		available
id	Indonesian	aspell-id	hunspell-id
ig	Igbo
ik	Inupiaq
is	Icelandic	aspell-is	hunspell-is
it	Italian	aspell-it	hunspell-it
iu	Inuktitut
ja	Japanese
ka	Georgian
kk	Kazakh
kl	Greenlandic
km	Khmer		available
kn	Kannada
ko	Korean
ku	Kurdish		available
kw	Cornish			crubadan corpus building
ky	Kyrgyz
lg	Luganda			A general translation effort.
li	Limburgish			crubadan corpus building
lo	Lao
lt	Lithuanian		hunspell-lt
lv	Latvian		available
mai	Maithili
mg	Malagasy		available
mi	Maori		available
mk	Macedonian		available
ml	Malayalam	aspell-ml	hunspell-ml
mn	Mongolian		available
mr	Marathi	aspell-mr	hunspell-mr
ms	Malay		hunspell-ms
mt	Maltese			Seems that there was one but now it has disappeared, maybe it was a conversion of the same words as in aspell-mt ?
nb	Bokmaal	aspell-no	hunspell-nb
nds	Lowlands Saxon		available
ne	Nepali		hunspell-ne
nl	Dutch	aspell-nl	hunspell-nl
nn	Nynorsk	aspell-no	hunspell-nn
nr	Ndebele (Southern)		hunspell-nr
nso	Sotho (Northern)		hunspell-nso
oc	Occitan		hunspell-oc
om	Oromo			crubadan corpus building. Oromo wiki entry
or	Oriya	aspell-or	hunspell-or
pa	Punjabi	aspell-pa	hunspell-pa
pap	Papiamento			crubadan corpus building
pl	Polish	aspell-pl	hunspell-pl
pt	Portuguese	aspell-pt	hunspell-pt
ro	Romanian		hunspell-ro
ru	Russian	aspell-ru	hunspell-ru
rw	Kinyarwanda		available
sa	Sanskrit			An apparent effort to create a Sanskrit hunspell dictionary
sc	Sardinian
se	Saami		available	A colossal 50Megs
shs	Secwepemctsin
si	Sinhala
sid	Sidama
sk	Slovak	aspell-sk	hunspell-sk
sl	Slovenian	aspell-sl	hunspell-sl
so	Somali			An apparent effort to create a Somali hunspell dictionary
sq	Albanian		hunspell-sq
sr	Serbian	aspell-sr	hunspell-sr
ss	Swati		hunspell-ss
st	Sotho (Southern)		hunspell-st
sv	Swedish	aspell-sv	hunspell-sv
ta	Tamil	aspell-ta	hunspell-ta
te	Telugu	aspell-te
tg	Tajik
th	Thai		hunspell-th
ti	Tigrigna			non-commercial use
tig	Tigre			crubadan corpus building
tk	Turkmen
tl	Tagalog		available
tn	Tswana		hunspell-tn
tr	Turkish		available	But like Finnish through voikko the typical solution for Turkish has been the Zemberek library, and to have an enchant backend, an Openoffice.org Extension, and a Firefox extension)
ts	Tsonga		hunspell-ts
tt	Tatar		available	Hard to see where this came from originally, and what license it is exactly, GPLv2+ (?). Perhaps it is an original work of ALT Linux and that actually is the canonical upstream ?
ug	Uyghur
uk	Ukrainian		hunspell-uk
ur	Urdu		hunspell-ur
uz	Uzbek		hunspell-uz
ve	Venda		hunspell-ve
vi	Vietnamese		hunspell-vi
wa	Walloon		hunspell-wa
wo	Wolof			www.alfanet.anafa.org make Wolof localizations of Firefox and Abiword.
xh	Xhosa		hunspell-xh
yi	Yiddish
yo	Yoruba			An apparent effort to create a Yoruba hunspell dictionary
zh	Chinese
zu	Zulu		hunspell-zu

User experience

Should not affect user experience.

Contingency plan

Continue to ship older dictionaries.

Documentation

[1]

Release Notes

There is a new default spell checking back-end, hunspell, for both the GNOME and KDE desktops, as well as applications such as OpenOffice.org, Firefox, and other XULRunner-based applications. This common back-end includes a set of shared, multi-lingual dictionaries for use with hunspell. This feature uses a single set of common dictionaries regardless of the application, which gives consistent suggestions for misspelled words and uses less diskpace by eliminating duplicate dictionaries.

Comments

Note that JDS is going down this route as well

The OpenOffice.org hunspell dictionary list of working dictionaries

The mozilla hunspell dictionary list of tri-licensed dictionaries

How to build a dictionary

How to convert an ispell affix to hunspell .aff

Language Codes

A somewhat related issue .

Will help on adding Indic hunspell dictionaries in Fedora - paragn.

php5 and bluefish still link to aspell at least - kmaraas. (It's not practical for me to port everything, just the core default installed components and the default spell-checking solutions for the main desktop environments and applications - caolanm)

Ubuntu is now following the Fedora practice as well.

---