giellalt / bugzilla-dummy

0 stars 0 forks source link

Problems changing external web analysis from vislcg to vislcg3. (Bugzilla Bug 664) #1332

Closed albbas closed 16 years ago

albbas commented 16 years ago

This issue was created automatically with bugzilla2github

Bugzilla Bug 664

Date: 2008-03-05T11:20:00+01:00 From: Trond Trosterud <> To: Saara Huhmarniemi <> CC: sjur.n.moshagen, tomi.k.pieski

Last updated: 2008-03-05T14:09:40+01:00

albbas commented 16 years ago

Comment 2621

Date: 2008-03-05 11:20:00 +0100 From: Trond Trosterud <>

We need to use vislcg3, as our rulefiles cannot be run by vislcg.

Now, it seems I have changed all relevant instances of vislcg to vislcg3.

But the result of all this is that no analysis works, at all :-(

In addition to checking in changes, we have also run the scripts, albeit with various results:

988 ./cgi-export 990 sudo mv /opt/sami/www/cgi-bin/smi/conf.pl /opt/sami/www/cgi-bin/smi/conflict_conf.pl 991 ./cgi-export 1019 ./copy_bin.sh 1031 Copying requires sudo rights 1032 Making and copying ccat.. 1033 cd /usr/tmp/gt/script/samiXMLParser && make 1034 g++ -c -g main.cpp 1035 Assembler messages: 1036 Fatal error: can't create main.o: Permission denied make: *** [main.o] Error 1 cp /usr/tmp/gt/script/samiXMLParser/ccat /usr/local/bin Password: cp: cannot stat `/usr/tmp/gt/script/samiXMLParser/ccat': No such file or directory 1037 Copying scripts.. 1038 cp /usr/tmp/gt/script/preprocess /usr/local/bin 1039 cp /usr/tmp/gt/script/lookup2cg /usr/local/bin 1040 cp /usr/tmp/gt/script/text_cat /usr/local/bin 1041 cp /usr/tmp/gt/script/cgi-export /usr/local/bin 1042 Copying Perl modules.. 1043 cp -r /usr/tmp/gt/script/samiChar /usr/lib/perl5/5.8.8 1044 cp -r /usr/tmp/gt/script/langTools /usr/lib/perl5/5.8.8

Our other actions can be read out of the cvs log.

Now, before the last commands, all lgs except fao worked, since fao has a vislcg3 dis-file. The other languages have two versions of the dis-file in /opt/smi/$lang/bin, one cg2 version (dis.rle) and one cg3 version (dis.bin). Until now, vislcg have accessed the dis.rle ones, and the dis.bin ones ahve been unused.

Evidently, the latest actions have cut that link off also, since now no analyser gives any answer, also not e.g. the kal converter (just to name one example).

Note the differences between vislcg and vislcg3:

vislcg: uses $lang-dis.rle files written for vislcg2 only uses --minimal as a flag for tracing rules uses --quiet as a non-output mode uses --grammar=$lang-dis.rle as flag

vislcg3 uses $lang-dis.rle files written for vislcg3, but can also read files written for vislcg2. Some of our dis-files are called $lang-3disl.rle or $lang-dis3.rle to distinguish, but they will be changed to $lang-dis.rle as soon as this hassle is over. uses --trace as a flag for tracing rules uses 2>/dev/null (or whatever) to suppress error messages. uses -g $lang-dis.rle as flag Can also run binary files: -g $lang-dis.bin All the disambiguators now generate $lang-dis.bin files to their bin catalogues.

albbas commented 16 years ago

Comment 2625

Date: 2008-03-05 13:28:00 +0100 From: Trond Trosterud <>

Now, most problems are solved (thanks to a telephone conf with Saara).

There is one spurious problem left:

It works with sme and all sámi letters, but for fao, it only works with ASCII. If I add non-ASCII letter, then fao gives no output.

To replicate: Go to Faroese analysis, and glue in e.g. Vitja okkara forsíðu fyri høvundar og síggj, hvussu tú kanst skriva eina grein.

Do the same for e.g. Vitja okkara grein.

The first sentence works for "all analyses", but not for "disambiguate". The second (ASCII-only) works for both.

I have added explicit utf8 requirement, but it does not help, and it should not be needed, since vislcg3 taks codepage from the environment, and since it worked for sme also before I added codepage info.

albbas commented 16 years ago

Comment 2626

Date: 2008-03-05 14:09:40 +0100 From: Trond Trosterud <>

Forget the last bug. My son borrowed my Mac last night, I suspect he had left the browser at Latin 1, thereby causing wrong input. With the browser set to UTF-8, I got the web input right, as well as the text I typed in myself.

Bug closed.