Search - the worst example of anything, ever

[dern]

If you search for 'M3' on pistonheads you get "None Found". If you're more specific and search for 'M3' just in the titles of BMW/M-Power you get "None Found".

Don't you guys do any testing?

Cheers,

Mark

[mrmr96]

The search function is known to be crap. However the google indexing of PH is pretty good. So your best bet is to go to google and do a search like:

"M3 siteistonheads.com"

(without the quotemarks)

Obvisouly it doesn't allow you search subforums, or headings etc specifically. But you can find posts by a user by sticking there name in the googlebox too.

I really don't see why the built in PH search function is as crap as it is, but it has been crap for ages. There's standard BB software which most forums use which actually works.. so I'm really baffled as to why it's so broken on here.

Still, hope the google workaround helps you some.

[PetrolTed]

Yes, I did test it.

It's quite simple, the search is set up to index words of 3 characters or more. I'm sure it could be tweaked.

I believe the search facility now has dedicated hardware which should improve reliability greatly. The basic mechanism is pretty sound (IMO), although I grant you I did miss out a couple of features when I wrote it. That was many moooons ago now....

[Mr Will]

The search cannot cope with very short search terms as with the technology it is based on these generate an unacceptable level of load.

We are working to reduce this limit but in the mean time you'll either need to use the google search option or make the term longer.

[FunkyGibbon]

PetrolTed said:

Yes, I did test it.

It's quite simple, the search is set up to index words of 3 characters or more.

Indeed many database search engines index on triplets - quite common and not an error. However, many forums trap the fact that your search term is to short and throws an error - perhaps that's an easy quick fix.

Though to be honest I think the search function is fine if you put a reasonable query to it. If all else fails use the google search.

FG

[dern]

PetrolTed said:

It's quite simple, the search is set up to index words of 3 characters or more. I'm sure it could be tweaked.

Ok.

If you search for 'E36' or 'E46' in M-Power with just in titles on or off you still get no results. Same with 'wet'.

If you search for 'door' in the same forum you do get results but they're the same ones if you have the just in titles checkbox on or off. If you have it on you get results without the word 'door' in the titles.

I'm no software engineer (oh no, wait, I am) but if that's tested I'm a dutchman (for sure)

It must be indexing on 4 letter words or more. I can see why you'd do that but I think I'd add some exceptions to index on like 'M3', '911' or maybe 'TVR'.

Cheers,

Mark

[Mr Will]

We recently migrated the search from Windows to Linux, which unintentionally raised the minimum characters from 3 to 4. It will be set back to three (or maybe lower) in the very near future though.

The search in titles function has been broken for a while now, but unfortunately we haven't had time to fix it.

[dern]

Mr Will said:

We recently migrated the search from Windows to Linux, which unintentionally raised the minimum characters from 3 to 4. It will be set back to three (or maybe lower) in the very near future though.

If you left it at 4 and had a list of exceptional 2 or 3 character strings to index searches on you'd keep the performance benefits of not indexing on all of the 2 or 3 character strings but still have a search engine that usefully searched the site.

As well as the obvious ones such as 911, 928, tvr, s3, 205... you could capture those submitted in the search field and the automatically index on the popular ones.

Also, when a search returned nothing you could say so on the results page but then helpfully display the google results.

Just a thought.

[monthefish]

dern said:

PetrolTed said:

It's quite simple, the search is set up to index words of 3 characters or more. I'm sure it could be tweaked.

Ok.

If you search for 'E36' or 'E46' in M-Power with just in titles on or off you still get no results. Same with 'wet'.

If you search for 'door' in the same forum you do get results but they're the same ones if you have the just in titles checkbox on or off. If you have it on you get results without the word 'door' in the titles.

I'm no software engineer (oh no, wait, I am) but if that's tested I'm a dutchman (for sure)

It must be indexing on 4 letter words or more. I can see why you'd do that but I think I'd add some exceptions to index on like 'M3', '911' or maybe 'TVR'.

Cheers,

Mark

OK - an experiment.

I just searched for the word dutchman, which (1) can clearly be seen in the post above, (2) is not a common and (3) is greater than 3 characters.




The search results.... (sorted for most recent first)

[dern]

My guess is that this is because when you press the search button it doesn't go in to the store of all the posts looking for your search terms (i.e. if the posts are stored in a database it doesn't do a select there and then). I imagine it goes in to some indexed store of the posts made on ph up to the point the last indexing run was done. If the indexing run is done nightly and you run the same search tomorrow you'd probably get the result your were after.

I guess that although searching the db directly would give many benefits (it would work for a start ) performance and visibility to external search engines such as google would not number amongst them.

[PetrolTed]

Search always used to run off an overnight backup, so yes it wouldn't be live. It's possible to update it with a live feed, but first Will needs to be replicated so he can do more work.

[Mr Will]

PetrolTed said:

Search always used to run off an overnight backup, so yes it wouldn't be live. It's possible to update it with a live feed, but first Will needs to be replicated so he can do more work.

Now that's a good idea!

[Mr Will]

PetrolTed said:

Search always used to run off an overnight backup, so yes it wouldn't be live. It's possible to update it with a live feed, but first Will needs to be replicated so he can do more work.

Now that's a good idea!

[anonymous-user]

I find the google search works well but it has gone all PH on me



So if one were a software tester, is it Haymarket that advertise any jobs at PH or is there a secret PH place?

[Slow M]

Ted,

Searching TVR>Classics for bump steer gives me nothing even though that topic is on today's page 1!

Searching TVR>Classics for bump within one year gives me three hits but today's isn't even one of them.

Searching TVR>Classics for bump within eight years gives me the same three.

I can't seem too look back further than one year (i think).

Also, I did screen shots but wasn't able to upload them all to show you.
Error message was

?

There is an invaluable resource in being able to search these posts that is no longer accessible.
This started not too long ago, I think it was still OK last year.
Do you think it will be repaired before too long?

Thanks,
B.

[PetrolTed]

As far as I know the search database is still built overnight, so you won't get today's data until tomorrow.

I don't know how far back the data goes now.

Mr Will looks after it all now - I've not had access to the system internals for over 2 years now

[Slow M]

MrWill,

Searching TVR>Classics for bump steer gives me nothing even though that topic is on today's page 1!

Searching TVR>Classics for bump within one year gives me three hits but today's isn't even one of them.

Searching TVR>Classics for bump within eight years gives me the same three.

I can't seem too look back further than one year (i think).

Also, I did screen shots but wasn't able to upload them all to show you.
Error message was

?

There is an invaluable resource in being able to search these posts that is no longer accessible.
This started not too long ago, I think it was still OK last year.
Do you think it will be repaired before too long?

Thanks,
B.

[Mr Will]

Slow M said:

MrWill,

Searching TVR>Classics for bump steer gives me nothing even though that topic is on today's page 1!

Searching TVR>Classics for bump within one year gives me three hits but today's isn't even one of them.

Searching TVR>Classics for bump within eight years gives me the same three.

I can't seem too look back further than one year (i think).

Also, I did screen shots but wasn't able to upload them all to show you.
Error message was

?

There is an invaluable resource in being able to search these posts that is no longer accessible.
This started not too long ago, I think it was still OK last year.
Do you think it will be repaired before too long?

Thanks,
B.

The search only goes back ~1 year at the moment, so anything before that time will not be picked up. It also only updates every night, so anything from the current day will not be picked up either.

We will be working to reduce these limitations in the near future, but we are a little short of time at the moment so I'm afraid it is taking longer than we hoped.

[Slow M]

Mr Will, thanks for the reply.

FYI, it was not catching an Item that was about 48 hours old.

B

[Mr Will]

Slow M said:

Mr Will, thanks for the reply.

FYI, it was not catching an Item that was about 48 hours old.

B

It is based on the date of each individual post, not on the topic start or last posted date. None of the recent posts on that thread mentioned the phrase "bump steer" which is why it wasn't picking it up.