Meeting_2012-12-04

What is needed for a new release:

• a working installer
• working plx conversion

Variablar for installeraren:

• Variablar relevante for om ting funkar i det heile:
  • Office-versjon: 2003, 2007, 2010, 2013
  • Windows-versjon: XP, Vista, W7, W8
  • språkkode: ISO (se-NO) vs LCID (1083)
  • bit-versjon for Windows: 32b vs 64b
• variablar uavhengig av grunnleggjande funksjonalitet:
  • Office-utgåve: home/student vs pro/full
  • installeringsmodus: admin/rot/nett vs lokal brukar vs terminalserver

Verkty for debugging:

• dependencywalker

DONE:

• Børre har laga og dokumentert ein testrutine, og fylgjer oppskrifta der slavisk kvar gong

TODO:

• look at further combinations of Windows and Office (Børre)
  • working on it
• get a proper MS certificate, to remove the warning users get when installing (Børre)

POSTPONED:

• find an outsourcing option (Børre, Sjur)
• next week: try updating NSIS ourselves (Børre, Sjur)
  • using Tino's documentation & old docu's from PLX
• send a request to an NSIS developer e-mail list (Sjur)
• try to find an NSIS script for building MSI packages (Børre)
• add Windows 8 support to our installer (if possible)

Remaining PLX bugs:

First PRI

Words are accepted and suggested in their compounding-form

REGRESSION - ALPH + NOUN NOT ACCEPTED

REGRESSION - DOESN'T RECOGNIZE SOME ACRO/PROP+NOUN, seems random

Unsolvable. If we fix these, we will get back the clitic-within-compound bug. The question is what is worse, and we have decided to accept some of the bugs above.

The middle ground we settle on is to accept hyphen-final words for word types that always require hyphens, like proper nouns, abbreviations and acronyms, but NOT for other words.

SUB accepted

Second Last PRI

Section for Tags-not-working

Bugs below this line can be left out of the next release if we are short on time.

Compounds

guollibusse is accepted by the test speller (as it should according to the PLX codes), but not by the full speller.

Test speller entries:

guol-li	NAIE,NePE,UI	// guolli+N+Sg+Nom
guol-li	NABO	// guolli+N+SgNomCmp+RCmpnd
guol-le--	WIX	// guolli+N+SgNomCmp+RCmpnd-
guol-le	NABO	// guolli+N+SgNomCmp+RCmpnd

LEXICON GOAHTICMP
 +SgNomCmp@U.NeedsVowRed.ON@:X7@U.NeedsVowRed.ON@ R ;
 +SgNomCmp@U.NeedsVowRed.OFF@:@U.NeedsVowRed.OFF@ R ;

bus-se	NAIE,NePE,UI	// busse+N+Sg+Nom
bus-se	NAIE,NePAE,UI	// busse+N+SgGenCmp+RCmpnd
bus-se	NAIE,NePAE,UI	// busse+N+Sg+Gen
bus-se	NABO	// busse+N+SgNomCmp+RCmpnd
bus-se	GaBO,NePIE,NOX,UI	// busse+N+Sg+Gen

Full speller entries:

guol-li	NAIE,NePE,UI	// cf first PLX line above
guol-le	NABO	// cf "guolle" PLX line above

Why do we get the second guolli PLX line in the test speller, but not in the full speller? That's the crucial point.

The analyser says:

guolli
guolli guolli+Ani+N+Sg+Nom

Task:

• build both spellers anew, with exactly the same code base
• go through each fst node by node for guoll[ei], and check if there are additional paths in one of the fst's
• check the fst's where guolle and guolli are extracted

Last PRI

DONE:

• Tomi has solved even more PLX bugs

TODO:

• investigate the inconsistent PLX conversion bug, cf guollibusse ( Tomi)
• solve the remaining PLX bugs one by one (Tomi)

• December 1:
  • PLX bugs solved
  • Alpha installer (MSI package built using WIX)
• December 10: Beta release
• December 20: Divvun 3 release