On 22/12/2017 00:38, Steven Green wrote:
> don't recall all of the testing that was done, but it was generally 
> concluded that it was skewed towards the first 5 characters of the 
> string, and was less predictable than most dot-dot options.. don't know 
> how you'd want to reproduce it, if you can't predict it

AFAIR, I NEVER used "like", because I never knew how it worked.

But, with that 5-char skew that Steve remembered, perhaps it was an 
implementation of (standard?) "Soundex"? If so, it would apply only to 
alphabetic (maybe alphanumeric?) strings:
   - always do an exact match on the very first char,
   - then match on the (english?) "sound" of the next 3/4/5/6/.. chars.

Eg:

If QBE has: Like paradox, this would translate to a Soundex code of P632

Then, DB records with Paradox, Poridex, Piridicks, etc (which also 
translate to P632) would be matches.

(I've not run the above example).

   - Mike


> OK Steve, I'll take the bait!
> 
> What is known of the traps/bugs/features/poo in 'LIKE'?
> 
> Anyone wanna do some tests?
> 
> I'll be quiet for a while trying to sensibly parse what we have so far,
> but will be lurking ...
> 
> - Bernie
> 
> ---
> This email has been checked for viruses by Avast antivirus software.
> https://www.avast.com/antivirus
>