Forums > MobyGames > The "s" bug

user avatar

Nélio (1976) on 5/2/2015 2:56 AM · Permalink · Report

I may have just solved it! If you know what I'm talking about, you know it's a big deal!

user avatar

Havoc Crow (29906) on 5/2/2015 5:17 AM · Permalink · Report

This one? (quoting Indra)

First (mis?)identified as the Auto-Assign bug, where the system autofills a developer name with whatever name has an additional s in it (though perhaps unrelated). For whatever reason, it also appears to pop up in other places. In short, we have no idea what this is and I have no idea why I put this in the planned section. :p Example: http://www.mobygames.com/images/shots/s/211251-get-out-amiga-screenshot-smashing-some-brickss.jpg (brickss)
user avatar

Nélio (1976) on 5/2/2015 6:59 AM · Permalink · Report

Yeap. It manifests itself in different ways.

user avatar

GTramp (81961) on 5/2/2015 7:35 AM · edited · Permalink · Report

It's great news indeed! I'll see it for myself when I'll be submitting some credits.

user avatar

Sciere (930960) on 5/2/2015 10:33 AM · Permalink · Report

I assume the fix isn't live yet, but it's already great you could identify the problem.

user avatar

Nélio (1976) on 5/2/2015 11:10 AM · Permalink · Report

Yes, not yet, but possibly as soon as tomorrow. We're doing an important update this weekend. :-)

user avatar

Terok Nor (42760) on 5/2/2015 12:14 PM · Permalink · Report

Great news. Any technical details to spare?

user avatar

Nélio (1976) on 5/2/2015 1:07 PM · Permalink · Report

Special character substitution. For searching and generating URLs, characters such as "à" and "á" get translated into "a", and "ç" into "c". The problem was with character "$" that was set to be translated into, you guessed it, "s". The issue is that when doing the replacing in Perl using a regex, the "$" character means end of string. So basically, all strings were being forced to end with an "s".

This already fixed the URL generation (which is what I was working on), but I know the search engine was relying on the same function, as well as many other components. I haven't tested the credits wizard nor the search engine, but I believe there will be improvements. Won't say they will work perfect, though.

user avatar

Terok Nor (42760) on 5/2/2015 9:16 PM · Permalink · Report

That old joke about regular expressions certainly comes to mind...

user avatar

Alex Z (1856) on 5/3/2015 2:50 PM · Permalink · Report

So that's why I sometimes have to remove the last letter when I search for something here!

Excellent work, Nélio!

user avatar

Daniel Saner (3503) on 5/3/2015 11:25 PM · Permalink · Report

[Q --start Alex Z wrote--]So that's why I sometimes have to remove the last letter when I search for something here!

Excellent work, Nélio! [/Q --end Alex Z wrote--]

First thing that came to my mind as well. I assume that this could fix a lot of the seemingly inexplicable search irregularities we've been seeing (not finding games when entering their exact title, etc.)

user avatar

Freeman (65146) on 5/2/2015 7:28 PM · Permalink · Report

Excellent news! Nice work, Nélio.

user avatar

Nélio (1976) on 5/3/2015 12:54 AM · Permalink · Report

Thanks. :-)

And the fix should go live within the next 24h!

user avatar

Freeman (65146) on 5/9/2015 9:09 PM · Permalink · Report

Some names that still have issues:

Entering 'Hiroyuki Ota' automatically links to Hiroyuki Otaka, not Hiroyuki Ōta (probably owing to the diacritical mark).

Entering Luis Mas automatically adds the credit to the Luis Mass entry.

When you enter Maria Antonia Rodriguez it automatically goes to the Antonia Rodriguez entry.

When you enter Jean-Francois Lavoie it automatically links the name to Jean-Francois Lajoie.

user avatar

Patrick Bregger (303331) on 5/16/2015 1:43 PM · Permalink · Report

Eric Zala gets matched to Eric Zalas.

user avatar

Patrick Bregger (303331) on 5/17/2015 7:46 AM · Permalink · Report

Guillaume Gagné-Gauthier --> Guillaume Gauthier

user avatar

Terok Nor (42760) on 5/17/2015 8:06 AM · edited · Permalink · Report

Given all the trouble it creates, I wonder if the best idea would be to disable non-exact credit matching altogether. It's always easier to merge than to split.

user avatar

Nélio (1976) on 5/17/2015 2:54 PM · Permalink · Report

I'm under the impression that my fix to the "s" bug isn't live at all. Also, it seems that an additional step is required even after my change is applied. So yeah, you're all getting the same old behavior we did before. I'm looking into it.

user avatar

Alex Z (1856) on 5/17/2015 4:54 PM · Permalink · Report

FWIW, I can only say that from my end your fix did work - I no longer have to remove the last letter of a game to search for it, and the searches give results as expected.

user avatar

Sciere (930960) on 5/20/2015 9:33 AM · Permalink · Report

Something strange seems to have happened since the fix. I'm working on a large listing that for instance has the name 'Alan Hsu'. I entered him already last week and he was found fine. If I enter him today, no developer at all is found and I need to make a new one. A lot more similar errors are listed here.

user avatar

Rola (8482) on 5/20/2015 10:32 AM · Permalink · Report

And I'd like to report that I was doing my favorite test every day and it's only today(?) that our search engine passed it: looking for "abrams" finally yelds all game titles, not just the one ending with that word.

user avatar

Patrick Bregger (303331) on 5/20/2015 3:11 PM · Permalink · Report

It often does not even work when I give the developer a fake AKA.

user avatar

Rola (8482) on 5/20/2015 3:41 PM · Permalink · Report

I do not know how the search script differs for each data set.

"Abram" now yelds all required game titles, but not the game group that has "Abrams"!

With people - searching "abrams" I get both "Abram" and "Abrams" surnames.
Searching for "abram" finds nothing!

Please analyze this.

user avatar

Nélio (1976) on 5/29/2015 4:16 PM · Permalink · Report

This is all way too complicated. It seems that each section of the site uses a different way to search for things. Using the main site search to lookup a developer works fine (apparently) but not within a credits submission. I think I'll have to treat every single thing as a different issue and fix one at a time.

Yay, fun.

user avatar

Rola (8482) on 5/29/2015 4:35 PM · Permalink · Report

Actually this division doesn't sound that bad.

It's OK to use "fuzzy search" for visitors, who often make typos or are quite clueless.

For approvers (maybe also when contributing?) it's good to assume they know what they're doing, so the results should be "exact match".

I hope that my findings were helpful.