How to pretty print one column of piped input?

Question

bash 4.4.20 and jq-1.6 on RHEL8

I get this nice output from jq and column. Hard to read, though.

pgbackrest info --output=json \
    | jq -r  '.[] | .backup[] | "\(.type) \(.label) \(.info.delta)"' \
    | column -t
full  20250712-224418F                   57448997604
full  20250712-231256F                   57391883702
full  20250713-000402F                   57448997604
full  20250713-013005F                   57447727844
incr  20250713-013005F_20250714-000002I  4960846146
[snip]
incr  20250817-101503F_20250822-000002I  8954487106
incr  20250817-101503F_20250823-000002I  9482330434
full  20250824-014502F                   53490962042
incr  20250824-014502F_20250825-000002I  5594054978
incr  20250824-014502F_20250826-000002I  6828647855

Piping it through a while loop, reading into three separate variables and then printf to convert the third column using numfmt also works, but seems so brutal. Is there a cleverer way to do it (where "it" either converts to human-readable or adds thousands-separators)?

pgbackrest info --output=json \
    | jq -r  '.[] | .backup[] | "\(.type) \(.label) \(.info.delta)"' \
    | (while IFS= read -r line; do 
           IFS=" " read Type Label Size <<< "$line"
           printf "%s %s %s\n" $Type $Label $(numfmt --to=si --suffix=B $Size)
       done) | column -t
full  20250712-224418F                   58GB
full  20250712-231256F                   58GB
full  20250713-000402F                   58GB
full  20250713-013005F                   58GB
incr  20250713-013005F_20250714-000002I  5.0GB
[snip]
incr  20250817-101503F_20250822-000002I  9.0GB
incr  20250817-101503F_20250823-000002I  9.5GB
full  20250824-014502F                   54GB
incr  20250824-014502F_20250825-000002I  5.6GB
incr  20250824-014502F_20250826-000002I  6.9GB

Answer 1

Using the following to emulate OP's jq output in my environment:

$ cat jq.out
full  20250712-224418F                   57448997604
full  20250712-231256F                   57391883702
full  20250713-000402F                   57448997604
full  20250713-013005F                   57447727844
incr  20250713-013005F_20250714-000002I  4960846146
incr  20250817-101503F_20250822-000002I  8954487106
incr  20250817-101503F_20250823-000002I  9482330434
full  20250824-014502F                   53490962042
incr  20250824-014502F_20250825-000002I  5594054978
incr  20250824-014502F_20250826-000002I  6828647855

Using printf to replace column and add thousands separators:

cat jq.out |
while read -r type label size
do
    printf "%-6s %-35s %'15.0f\n" $type $label $size
done

If needing to type this regularly, and assuming using the same output format each time, we can wrap the guts in a function:

my_column() { 
    while read -r type label size
    do   
        printf "%-6s %-35s %'15.0f\n" $type $label $size
    done  
}

And then the main call becomes:

cat jq.out | my_column

Both of these generate:

full   20250712-224418F                     57,448,997,604
full   20250712-231256F                     57,391,883,702
full   20250713-000402F                     57,448,997,604
full   20250713-013005F                     57,447,727,844
incr   20250713-013005F_20250714-000002I     4,960,846,146
incr   20250817-101503F_20250822-000002I     8,954,487,106
incr   20250817-101503F_20250823-000002I     9,482,330,434
full   20250824-014502F                     53,490,962,042
incr   20250824-014502F_20250825-000002I     5,594,054,978
incr   20250824-014502F_20250826-000002I     6,828,647,855

NOTES:

• if this does not print the thousands separator then you likely need to play with your locale settings (eg, LC_ALL= printf ... or LC_ALL=en_US.TUF-8 printf ...)
• adjust the column widths by modifying the -6s, -35s and 15.0f formats
• the use of the (bash) printf builtin should mean this while/read loop is relatively fast and for smallish sets of data should be on par with the cost of spawning an additional subshell for a column call; for larger datasets the looping/formatting construct could be replaced with a comparable awk (perl, python, etc) call for better performance

OP's code would then become:

pgbackrest info --output=json |
    | jq -r  '.[] | .backup[] | "\(.type) \(.label) \(.info.delta)"' \
    | while read -r type label size; do printf "%-6s %-35s %'15.0f\n" $type $label $size; done

########
# or

pgbackrest info --output=json |
    | jq -r  '.[] | .backup[] | "\(.type) \(.label) \(.info.delta)"' \
    | my_column

---

In this case since custom formatting is only required for the one column we could fallback to OP's original numfmt idea while eliminating a) the while/read loop, b) the subshell for calling numfmt and c) the additional subshell invoked due to the parens; a couple ideas:

$ cat jq.out | numfmt --field 3 --grouping         | column -t -R 3
$ cat jq.out | numfmt --field 3 --format "%'15.0f" | column -t -R 3

NOTE: depends on locale

Wrapped in a function:

$ my_column() { numfmt --field 3 --grouping | column -t -R 3; }
$ cat jq.out | my_column

These also generate:

full  20250712-224418F                   57,448,997,604
full  20250712-231256F                   57,391,883,702
full  20250713-000402F                   57,448,997,604
full  20250713-013005F                   57,447,727,844
incr  20250713-013005F_20250714-000002I   4,960,846,146
incr  20250817-101503F_20250822-000002I   8,954,487,106
incr  20250817-101503F_20250823-000002I   9,482,330,434
full  20250824-014502F                   53,490,962,042
incr  20250824-014502F_20250825-000002I   5,594,054,978
incr  20250824-014502F_20250826-000002I   6,828,647,855

Answer 2

You could tell numfmt which field to format:

$ numfmt --help
[...]
  -d, --delimiter=X    use X instead of whitespace for field delimiter
      --field=FIELDS   replace the numbers in these input fields (default=1)
                         see FIELDS below
[...]

So, maybe something like:

pgbackrest info --output=json |
    jq -r  '.[] | .backup[] | "\(.type) \(.label) \(.info.delta)"' |
    numfmt --field=3 --to=si --suffix=B |
    column -t