clear
{{language.name}} No language found.
swap_horiz
{{language.name}} No language found.
search
deniko
2019-07-29 13:12
Why am I finding "Yikes!" searching for "perfect"?

https://i.imgur.com/EoSIHeH.png
hide replies
AlanF_US
2019-07-29 16:00
I see the same thing. Very odd. I don't see anything in the log on the "Yikes!" page that would explain it.
brauchinet
2019-07-29 16:32
This must be a temporary error in the search engine.
for example: right -> Yikes!
https://tatoeba.org/eng/sentenc...rom=eng&to=und
or darn ->
https://tatoeba.org/eng/sentenc...rom=eng&to=und
hide replies
deniko
2019-07-29 16:34 - 2019-07-29 16:39
Yeah, I can see it for "right", but I don't see "yikes" in the search results for darn.

In any case, that's bizarre.

> a temporary error

Classifying any error as "temporary" is quite an optimistic outlook :)
hide replies
brauchinet
2019-07-29 16:40 - 2019-07-29 16:48
Yeah, "darn" yields other unexpected results.
Funny, now I tried a lot of simple English words and didn't get false results (at least not on the first page). It seems I was lucky with the first two.

Edit: No, there are many false results.
gillux
2019-07-31 06:50
That’s very weird indeed. I am confident that this is caused by the recent update of the search engine. I’m working on it.
gillux
2019-07-31 07:22
Problem solved. Apparently English was the only language affected.
hide replies
deniko
2019-07-31 08:37
Thanks! That was fast, well done! @brauchinet's optimism was well founded then.

Is there a simple explanation of why that was happening, I wonder? It's so bizarre I'm really curious about the why.
hide replies
gillux
2019-07-31 14:47
I am not 100% sure about the why. The only thing I am sure is that the English index showed some errors when I checked it using the "indextool --check" command. So I recreated it.

When we moved from Manticore 2 to Manticore 3 [1] a few days ago, indexes were converted to a new format using a special tool [2]. That tool doesn’t look very stable yet because it caused some unexpected problems I had to work around at that time. So I suspect it also didn’t convert the English index correctly.

I remember "darn" eng→ita showed 4 results of which 2 were incorrect, and now it shows 4 results of which 4 are corrects. So I suspect the tool messed around with the list of sentences each keywords are associated to.

[1] https://github.com/Tatoeba/tatoeba2/issues/1929
[2] https://manticoresearch.com/201...-to-version-3/
hide replies
deniko
2019-07-31 15:27
Thank you elaborating. Index corruption seems like a plausible explanation as indexes are definitely used when you do a search.
Guybrush88
2019-08-01 07:25 - 2019-08-01 07:26
"Apparently English was the only language affected."

Actually, it seems that also other languages are having problems. I tried searching just now sentences in Italian and Esperanto contibuted in a span of time from two days ago to yesterday and they're not indexed yet
hide replies
TRANG
2019-08-01 10:31
It's a different problem.

In your case, the issue was that the sentences were not indexed. The scripts to update the indexes were not running because the cron jobs were disabled.

I ran the scripts manually and I think gillux re-enabled the cron jobs, so it should be fine now :)
hide replies
Guybrush88
2019-08-01 10:32
I understand, thanks to both you and gillux :)
gillux
2019-08-01 11:20
My bad, I forgot to re-enable reindexation /o\
Should be fine now.