Cache symbol completion result

[tompng]

Make symbol completion faster

• Cache Symbol.all_symbols.sort and reuse it.
• Restrict the number of symbol candidates to reduce completion/rendering cost in Reline.
• Use bsearch to perform O(logN) search from cached sorted symbol list.

irb -f
irb(main):001> s='a';symbols = 100000.times.map{s=s.succ; [s.to_sym, ('a'+s).to_sym]};
irb(main):002> :aaabq
               :aaabk 
               :aaabl 
               :aaabm 
               :aaabn 
               :aaabo 
               :aaabp 
               :aaabq▄
               :azyd █
               :azye ▀
               :azyf  
               :azyg  
               :azyh  
               :azyi  
               :azyj  
               :azyk  

Completion candidates of symbol contains first 50 + last 50 symbols matched to the completion target.

[tompng]

I think it is relatively safe because it checks the value strictly, but it's a bit hacky...

[st0012]

Tbh I'd prefer updating repl_type_completor gem to have a better interface for this. And we can print warnings to ask users to upgrade to that version.

[st0012]

I feel sorry to say this because I was the one who brought up this idea 😅 But are we sure that the performance improvement is worth the complexity? Especially the binary search part.

[st0012]

Would it be more efficient if we convert the prefix to symbol rather than calling to_s on all iterated symbols?

[tompng]

Sorry symbol_names was already a string. I deleted to_s and add missing return [] unless start_index.
We need to search from an array of string symbols.map(&:inspect) to exclude symbols like :"a b".

[st0012]

Tbh I'd prefer updating repl_type_completor gem to have a better interface for this. And we can print warnings to ask users to upgrade to that version.

[tompng]

I think there is room for discussion whether to keep symbol completion or to drop it.

If we're going to keep it, we need to do this bsearch because symbol filtering (main bottleneck in IRB side) is few times slower than Symbol.all_symbols.

[tompng]

In the following situation, Symbol.all_symbol.sort takes about a second. I think it is not acceptable.

Symbol.all_symbols.size is 100000 with average symbol length=15 (About the number in Rails app)
Running in docker (host and image platform mismatch)