| .. | ||
| algo | ||
| curses | ||
| fzf | ||
| util | ||
| cache_test.go | ||
| cache.go | ||
| chunklist_test.go | ||
| chunklist.go | ||
| constants.go | ||
| core.go | ||
| Dockerfile.arch | ||
| Dockerfile.centos | ||
| Dockerfile.ubuntu | ||
| item_test.go | ||
| item.go | ||
| LICENSE | ||
| Makefile | ||
| matcher.go | ||
| merger_test.go | ||
| merger.go | ||
| options_test.go | ||
| options.go | ||
| pattern_test.go | ||
| pattern.go | ||
| reader_test.go | ||
| reader.go | ||
| README.md | ||
| terminal.go | ||
| tokenizer_test.go | ||
| tokenizer.go | ||
fzf in Go
This directory contains the source code for the new fzf implementation in Go. The new version has the following benefits over the previous Ruby version.
Motivation
No Ruby dependency
There have always been complaints about fzf being a Ruby script. To make matters worse, Ruby 2.1 dropped ncurses support from its standard libary. Because of the change, users running Ruby 2.1 or above were forced to build C extensions of curses gem to meet the requirement of fzf. The new Go version will be distributed as an executable binary so it will be much more accessible and easier to setup.
Performance
With the presence of GIL, Ruby cannot utilize multiple CPU cores. Even though the Ruby version of fzf was pretty responsive even for 100k+ lines, which is well above the size of the usual input, it was obvious that we could do better. Now with the Go version, GIL is gone, and the search performance scales proportional to the number of cores. On my Macbook Pro (Mid 2012), it was shown to be an order of magnitude faster on certain cases. It also starts much faster than before though the difference shouldn't be really noticeable.
Differences with Ruby version
The Go version is designed to be perfectly compatible with the previous Ruby
version. The only behavioral difference is that the new version ignores the
numeric argument to --sort=N option and always sorts the result regardless
of the number of matches. The value was introduced to limit the response time
of the query, but the Go version is blazingly fast (almost instant response
even for 1M+ items) so I decided that it's no longer required.
System requirements
Currently prebuilt binaries are provided only for OS X and Linux. The install script will fall back to the legacy Ruby version on the other systems, but if you have Go 1.4 installed, you can try building it yourself.
However, as pointed out in golang.org/doc/install, the Go version may not run on CentOS/RHEL 5.x and thus the install script will choose the Ruby version instead.
The Go version depends on ncurses and some Unix system calls, so it shouldn't run natively on Windows at the moment. But it should be not impossible to support Windows by falling back to a cross-platform alternative such as termbox only on Windows. If you're interested in making fzf work on Windows, please let me know.
Build
# Build fzf executables and tarballs
make
# Install the executable to ../bin directory
make install
# Build executables and tarballs for Linux using Docker
make linux
Third-party libraries used
- ncurses
- mattn/go-runewidth
- Licensed under MIT
- mattn/go-shellwords
- Licensed under MIT
Contribution
For the time being, I will not add or accept any new features until we can be sure that the implementation is stable and we have a sufficient number of test cases. However, fixes for obvious bugs and new test cases are welcome.
I also care much about the performance of the implementation (that's the reason I rewrote the whole thing in Go, right?), so please make sure that your change does not result in performance regression. Please be minded that we still don't have a quantitative measure of the performance.