choose is a tool for creating selection dialogs. It can also:

• grep using pcre2 syntax with arbitrary delimiters
• stream edit
• apply uniqueness and (limited) sorting

See benchmarks here comparing it to other tools.

sudo apt-get install cmake pkg-config libpcre2-dev libncursesw5-dev libtbb-dev
git clone https://github.com/jagprog5/choose.git && cd choose
make install
[ -f ~/.bashrc ] && source ~/.bashrc

make uninstall

choose --help

Dialogs can be used to select between tokens. By default, each token is delimited by a newline character. This command:

echo $'here❗\nis\neach\noption📋'\
  | choose --tui -p "Pick a word!"

Gives this interface:

┌───────────────────────┐  
│Pick a word!           │  
└───────────────────────┘  
> here❗  
  is  
  each  
  option📋  

Instead of a newline character, a literal sequence or regular expression can delimit the input.

echo -n "this 1 is 2 a 3 test"\
  | choose -r " [0-9] "

this  
is  
a  
test

The delimiter in the output can also be set.

echo -n "this test here"\
  | choose " " -o $'\n===\n'

this
===
test
===
here
===

Transformations can be done in a specified order. This command prints every other word by:

1. Suffixing each token with its arrival order.
2. Filtering for tokens that end with an even number.
3. Substituting to remove the arrival order.

echo -n 'every other word is printed here' | \
  choose -r ' ' --index=after            `# <-- 1` \
                -f '[02468]$'            `# <-- 2` \
                --sub '(.*) [0-9]+' '$1' `# <-- 3`

There's a difference between a typical shell pipeline like:

cat some_content | grep "test" | head -n 5

Compared to this:

cat some_content | choose -f "test" --head 5

The former is restricted to working with lines, whereas the latter works with tokens. Tokens are contiguous ranges and can contain newline characters, whereas lines can't. choose is line oriented by default, but doesn't have to be.

Rather than specifying how tokens are terminated, the tokens themselves can be matched for. A match and each match group form a token. This is like grep -o.

echo "aaabbbccc"\
  | choose --match "bbb(...)" -r

bbbccc
ccc

There's a few different ways that choose can edit a stream. This way is generally the best:

echo "this is a test" | choose -r --sed "\w+" --replace banana

In contrast, this implicitly separates the input into tokens each delimited by a newline. Then, on each token a global substitution is applied:

echo "this is a test" | choose -r --sub "\w+" banana

Lastly, this is a weird hack that leverages the input and output delimiters. It can be faster, but the replacement must be a literal string:

echo "this is a test" | choose -r "\w+" -o banana -d

choose uses PCRE2, which allows for lookarounds + various other regex features, compared to sed which allows only sed regex. This requires different logic for management of the match buffer, since lookbehind bytes must be properly retained as tokens are created. Meaning sed can't match expressions like this:

echo "banana test test" | choose -r --sed '(?<!banana )test' --replace hello

sed works per line of the input. choose doesn't assume the distinction of lines. suppose there is a file that consists of the newline character repeatedly. sed will apply the entire throughput of its logic on each empty line even though its not necessary to do so when apply the subsitution. choose is faster in these cases.

To emphasize a point, here is a tricky substitution which has a target that includes a newline and null character:

echo -e "this\n\0is\na\ntest" | choose -r --sed 'is\n\0is' --replace something

sed can't make a substitution if the target contains the delimiter (a newline character); the input is split into lines before substitution occurs, so the delimiter never makes it to the substitution logic. The way this is avoided is to use sed -z, which changes the delimiter from newline to null. But in this case, the target includes null too! So it can't process the input properly. One quick way of fixing this is to use tr to change the input before it gets to sed (by changing null to a different character), but this can lose information and lead to ambiguous cases (if the delimiter is changed to something found naturally in the input).

choose does an internal sort. This means it can run out of memory when working with large inputs. This is in contrast to gnu sort, which instead does an external sort; it writes sorted chunks to temporary files which are merged later. An internal sort makes sense for choose, since selection dialogs aren't meant to be very large (and in most cases the output from choose can be piped to gnu sort anyway).

Here is an example which sorts the input and leaves only unique entries:

echo -n "this is is test test "\
  | choose " " -us

is
test
this

Sorting and uniqueness take a comparison operator, each. The comparison can be done:

• lexicographically (the default alphabetical order)
• numerically (numbers like 1, 10, 11., -3.2, but NOT +3, positive leading sign is not allowed)
• general numerically (supports scientific notation)

For example, this sorts based on numeric value:

echo -e "10\n11\n5\n3" | choose -sn

3
5
10
11

A field can be specified, indicating which part of a token should be used when applying sorting and uniqueness.

echo "1,gamma,1
3,alpha,3
2,beta,2"\
  | choose -s --field '^[^,]*+.\K[^,]*+'

3,alpha,3
2,beta,2
1,gamma,1

Sorting is implemented to effectively leverage truncation. For example:

cat very_large_file | choose --sort --out=5

That command only stores the lowest 5 entries throughout its lifetime; the memory usage remains bounded appropriately, no matter the size of the input. The equivalent: sort | head -n5 does not do this and will be slower for large inputs. For clarity on when this occurs, see --is-bounded.

Uniqueness is applied upfront; unique tokens are remembered and compared against as new tokens arrive from the input. For contrast, GNU sort checks for consecutive unique elements (like the uniq command) just before the output.

ch_hist is a bash function installed with choose (only if bash is the shell being used). It allows a previous command to be re-run, like fzf.

  cd build
  rm -rf *
  cmake ..
  make install
> choose -h
┌────────────────────────────────────────────────────────────────────────────────┐
│Select a line to edit then run.                                                 │
└────────────────────────────────────────────────────────────────────────────────┘

ch_hist
ch_hist git
ch_hist hello there