It seems grc versions are not updated on the server

[Trondtr]

For fao grc, the following sentence:

Hann skal læra at virða tær skyldur og tey rættindi, sum fólkaræði byggir á, har ímillum teir sáttmálar, ið Sameindu Tjóðir hava samtykt í sambandi við barnarættindi og mannarættindi annars.

gets a correction samtykt to samtykja, and the error message "Det ser ut til at verbet ditt ikkje skulle ende på ð.", corresponding to the tag &msyn-sup-inf. The problem is that this (false) alarm cannot be reproduced on the command line. The offending rule was removed from the grammarchecker.cg3 file on february 6th.

The conclusion is thus that MS Word accesses an online grammar version dating from before February 6th. I do not know whether this holds for fao only, or whether it goes for all the online grammars, but it should be looked into. The fao grc is in use by a whole language community (the largest of the ones we serve) and the issue should thus get some priority. If the error carries over to the other grc versions the issue is even worse. Due to bug #70, I have only tested on MS Word.

[snomos]

This bug is really about how and when to update the grammar checker *.zcheck files on the production server. We have a standardised process for rolling out new spellers, but nothing for grammar checkers.

[Trondtr]

While waiting, we should document how to inspect the dates for the grammarcheckers.

[Trondtr]

After update on the server (today), I now check as describet above: I add the following sentence

Hann skal læra at virða tær skyldur og tey rættindi, sum fólkaræði byggir á, har ímillum teir sáttmálar, ið Sameindu Tjóðir hava samtykt í sambandi við barnarættindi og mannarættindi annars.

and send it to checkng. Now (contrary to yesterday) the setence comes out as true negative. The update thus is done, and the new version works.

The bug should not be closed, though: What happened today was that 3 languages were manually updated (fo.zcheck, nb, smn). The other languages were not updated, and the automatic update procedure is also not in place. Also, the documentation for how to check for dates (either via easter eggs or via inspecting the dates on pahkat.uit.no) is not in place.

The good news is that 3 languages are updated, so that the last 9 months of work is available for the language communities. Thanks!

[Trondtr]

This bug is really about how and when to update the grammar checker *.zcheck files on the production server. We have a standardised process for rolling out new spellers, but nothing for grammar checkers.

So, let us partition the problem: if I understand correctly, we need two things:

1. a operational process for ensuring regression.
2. add the *.zcheck file to the list of files to be copied automatically to the server

There should be no reason for waiting months for the second point, it seems like a five-minute-fix, at most. The issue thus seems to have stranded at the first point.

I do not whether speller update is linked to the make check procedure for transducers, but if it is, it should be straightforward to to the same for grammar checkers. We do have a test procedure, and it can be used for this purpose as well.

[unhammer]

I guess giellalt/lang-smj#109 (comment) is the same issue.

The version running on the server is from before September 2019 ( giellalt/lang-smj@9e9a73a removed "msyn thingy" from errors.xml), that's over 5 years ago! @bbqsrc what's the procedure for getting these packages updated?

[snomos]

The version running on the server is from before September 2019

That is very strange, and points to other issues. The .zcheck file for SMJ was sent to Gothenburg and uploaded to the server on December 17 (I think that included SMJ - two files were sent, but I am unable to tell what languages where updated then). And even if it was not at that time, I know for sure that the SMJ grammar checker has been updated the last time in 2024, not in 2019.

That is, there is something strange going on.

[snomos]

About the procedure: the present procedure goes as follows:

• compile a .zcheck file locally
• send the file to a person with access to the API server
• manually copy the file to the API server

@bbqsrc is working on an update to the whole Páhkat ecosystem that will include automatic releases of grammar checkers, the same way we do speller releases now (a version tag in a specific format). This update will also cover the api server, and thus make the api server update automatically upon stable/tagged releases.

[unhammer]

There is also a possibility that the errors.xml is of an older date than the rest of the checker (that the rest of the checker is from December, and it's just the errors.xml that's five years old). Even so, it would be good to have the latest changes in.

[snomos]

See my comments in giellalt/lang-smj#109 - I am not able to replicate the issue with errors.xml. I don't understand why you get what you get.

[unhammer]

errors.xml was #98 so not relevant for this.

compile a .zcheck file locally

any reason not to just use the one built by Tino's packaging?

[snomos]

any reason not to just use the one built by Tino's packaging?

No, not really, I just find it more convenient to build locally (=how do I get the one built by Tino?).

[unhammer]

well, any ubuntu machine doing apt update && apt dist-upgrade will have new ones in /usr/share/voikko/4, but I guess compiling is easier from a mac :)

@TinoDidriksen is there a nightly-data script for osx? Feel like I've heard of one once.

[TinoDidriksen]

There is https://apertium.projectjj.com/osx/install-nightly-data.sh invoked as e.g. ./install-nightly-data.sh giella-smj-speller which installs /usr/local/share/voikko/4/smj.zcheck and friends.

Rather simple script that downloads the .deb file from https://apertium.projectjj.com/osx/nightly/data.php and unpacks it, with some path adjusting for .mode files.