Falsehoods Programmers Believe

github.com

8 points·manx··3 comments

A couple years ago, a user of one of my software told me "I can't find some customers in the software, when I type their name nothing happen. It seems it only happens with the new Chinese students!".

Those Chinese customers all had 2 letters last name (mostly 'Xi' I think) but when I developed the software I thought any last name would be at least 3 letters long and to avoid taxing the database, I didn't start autocompletion before 3 letters where entered in the textbox.

Now the search starts at 2 letters, but I fully expect to find a one letter last name customer before retirement ;)

Depending on how you encode names and what you define as a "letter", many Korean last names are single characters (Lee: 이, Park: 박, Kim: 김), though they're usually transliterated into multiple letters (Oh, not O). Every Korean "character" has multiple "letters", but they treated as one entity most of the time (like Chinese characters).

Personally, I wouldn't risk it and would just make single character lookups reasonably fast. I doubt the number of relevant records exceed 100k, so the savings shouldn't be significantly enough to bother with. However, I've seen worse things done with poorer justifications.

All the names are transliterated in Latin alphabet, when the passports are delivered I guess.

The system is already quite slow as is (the DB is distant and on an underpowered server) so I don't want to tax it even more. What I may do next change is to cache locally all names at the start of the client and hit the DB only when it didn't find the name in the cache.

Reply on news.ycombinator.com