Improve prompt generating performance by caching prompt parts(%m, %M)

[tompng]

Fixes #1126

In prompt calculation, main.to_s was called on every keystroke and every line in multiline input.
Cache prompt parts(%m and %M) so that main.to_s is only called once per read-eval cycle.

Also fixes this problem.

irb(main):001> def self.to_s = 'a' * rand(1..10)
=> :to_s
# before
irb(aaa):002* def f 
irb(aaaaaaa):003*   puts 'Hello'
irb(aa):004*   puts 'World'
irb(aaaa):005> end

# after
irb(aaaa):002* def f
irb(aaaa):003*   puts 'Hello'
irb(aaaa):004*   puts 'World'
irb(aaaa):005> end

[st0012]

Since format_prompt is only called in generate_prompt, is it possible to be called outside of readmultiline context? (perhaps in the dyamic_prompt callback?)

If it can be, let's also wrap the caller inside the wrapper mentioned above to provide caching?

[tompng]

Currently, generate_prompt without cache is only used in test code testing format_prompt.
I'd like to make cache mechanism optional because:

• Testing format_prompt is easier.
• Always requiring wrapper is simple, but also introduces some sort of (small) complexity in call chain.

(But I'm not really sure, maybe wrong)

[st0012]

I see. Then let's at least add comments to explain why it could be nil, like

Suggested change

	part_cache = @prompt_part_cache || {}
	# @prompt_part_cache could be nil in unit tests
	part_cache = @prompt_part_cache || {}

[st0012]

I think we can create a wrapper to ensure wherever we can generate prompt we have this set and cleaned. Like

with_prompt_part_cached do
  ...
end

[tompng]

Nice catch! updated

[st0012]

I see. Then let's at least add comments to explain why it could be nil, like

Suggested change

	part_cache = @prompt_part_cache || {}
	# @prompt_part_cache could be nil in unit tests
	part_cache = @prompt_part_cache || {}