Search a multidimensional array

[Alex Tweedly]

On 20/04/2018 20:15, Richard Gaskin via use-livecode wrote:
> This is an interesting problem, Andrew.  Thanks for posting it.
>
Indeed!
> One modest performance gain available easily may be to change this:
>
>   repeat for each element tThisItem in tInventoryArray
>          if tThisItem["description"] contains tSearchQuery then
>             put tThisItem into tSortedInventoryArray[ \
>               (the number of elements of tSortedInventoryArray) + 1]
>          end if
>   end repeat
>
> ...to:
>
>   put 0 into i
>   repeat for each element tThisItem in tInventoryArray
>          if tThisItem["description"] contains tSearchQuery then
>             add i to i
>             put tThisItem into tSortedInventoryArray[i]
>          end if
>   end repeat
>
I'd go for something like
   put tMatchingKeys
   repeat for each line K in the keys of tInventoryArray
          if tInventoryArray [K]["description"] contains tSearchQuery then
             put K & CR after temp
          end if
   end repeat

( or use    repeat for each key K in tInventoryArray
I used that other form because I keep a variable with the set of keys - 
and that can be re-used to subsequently search subsets with the 'repeat 
for each line' style of repeat :-)

and then use the key list for subsequent access. Unless you are changing 
or destroying the 'extracted' elements, it will save you significant 
time to just use it in-place.

Also,  testing with an earlier version of LC (maybe 7.x ??) showed that 
"repeat for each key ..." was faster than "repeat for each element ..." 
- no idea why, and haven't retested in at least a couple of years.

I have a library (announced a couple of days ago - don't use that 
initial version !!) for doing almost exactly this. For a data table 
apparently similar to yours (100 columns, 50,000 rows) this kind of 
simple search takes around 180 millisecs (on an aging Macbook Pro).

NB - the latest version is *much* changed from the version on github - I 
will finish testing tonight and post the new version to Github and put a 
copy on my own website (and announce to this list).

-- Alex.
P.S. I agree with Richard's request for a copy of the data / current 
code if it is not sensitive. Feel free to email directly to me if that 
is possible.