I profiled fzf and it turned out that it was spending significant amount
of time repeatedly converting character arrays into Unicode codepoints.
This commit greatly improves search performance after the initial scan
by memoizing the converted results.
This commit also addresses the problem of unbounded memory usage of fzf.
fzf is a short-lived process that usually processes small input, so it
was implemented to cache the intermediate results very aggressively with
no notion of cache expiration/eviction. I still think a proper
implementation of caching scheme is definitely an overkill. Instead this
commit introduces limits to the maximum size (or minimum selectivity) of
the intermediate results that can be cached.
When fzf works in filtering mode (--filter) and sorting is disabled
(--no-sort), there's no need to block until input is complete. This
commit makes fzf print the matches on-the-fly when the following
condition is met:
--filter FILTER --no-sort [--no-tac --no-sync]
or simply:
-f FILTER +s
This removes unnecessary delay in use cases like the following:
fzf -f xxx +s | head -5
However, in this case, fzf processes the input lines sequentially, so it
cannot utilize multiple cores, which makes it slightly slower than the
previous mode of execution where filtering is done in parallel after the
entire input is loaded. If the user is concerned about the performance
problem, one can add --sync option to re-enable buffering.
