function __fish_print_help --description "Print help message for the specified fish function or builtin" --argument item error_message switch $item case . set item source case : set item true case '[' set item test end # Do nothing if the file does not exist if not test -e "$__fish_data_dir/man/man1/$item.1" -o -e "$__fish_data_dir/man/man1/$item.1.gz" return end # Render help output, save output into the variable 'help' set -l help set -l format set -l cols if test -n "$COLUMNS" set cols (math $COLUMNS - 4) # leave a bit of space on the right end # Pick which command we are using to render output or fail if none if command -qs nroff set format nroff -c -man -t if test -e $__fish_data_dir/groff/fish.tmac set -a format -M$__fish_data_dir/groff -mfish end if test -n "$cols" set -a format -rLL={$cols}n end else if command -qs mandoc set format mandoc -c if test -n "$cols" set -a format -O width=$cols end else echo fish: (_ "Cannot format help; no parser found") return 1 end if test -e "$__fish_data_dir/man/man1/$item.1" set help ($format "$__fish_data_dir/man/man1/$item.1" 2>/dev/null) else if test -e "$__fish_data_dir/man/man1/$item.1.gz" set help (gunzip -c "$__fish_data_dir/man/man1/$item.1.gz" 2>/dev/null | $format 2>/dev/null) end # The original implementation trimmed off the top 5 lines and bottom 3 lines # from the nroff output. Perhaps that's reliable, but the magic numbers make # me extremely nervous. Instead, let's just strip out any lines that start # in the first column. "normal" manpages put all section headers in the first # column, but fish manpages only leave NAME like that, which we want to trim # away anyway. # # While we're at it, let's compress sequences of blank lines down to a single # blank line, to duplicate the default behavior of `man`, or more accurately, # the `-s` flag to `less` that `man` passes. set -l state blank set -l have_name begin string join \n $error_message for line in $help # categorize the line set -l line_type switch $line case ' *' \t\* # starts with whitespace, check if it has non-whitespace printf "%s\n" $line | read -l word __ if test -n $word set line_type normal else # lines with just spaces probably shouldn't happen # but let's consider them to be blank set line_type blank end case '' set line_type blank case '*' # Remove man's bolding set -l name (string replace -ra '(.)'\b'.' '$1' -- $line) # We start after we have the name contains -- $name NAME; and set have_name 1; and continue # We ignore the SYNOPSIS header contains -- $name SYNOPSIS; and continue # Everything after COPYRIGHT is useless contains -- $name COPYRIGHT; and break # not leading space, and not empty, so must contain a non-space # in the first column. That makes it a header/footer. set line_type meta end set -q have_name[1]; or continue switch $state case normal switch $line_type case normal meta printf "%s\n" $line case blank set state blank end case blank switch $line_type case normal meta echo # print the blank line printf "%s\n" $line set state normal case blank meta # skip it end end end end | string replace -ra '^ ' '' | ul | # post-process with `ul`, to interpret the old-style grotty escapes begin set -l pager less set -q PAGER and echo $PAGER | read -at pager not isatty stdout and set pager cat # cannot use a builtin here # similar to man, but add -F to quit paging when the help output is brief (#6227) set -xl LESS isrFX # less options: # -i (--ignore-case) search case-insensitively, like man # -s (--squeeze-blank-lines) not strictly necessary since we already do that above # -r (--raw-control-chars) to display bold, underline and colors # -F (--quit-if-one-screen) to maintain the non-paging behavior for small outputs # -X (--no-init) not sure if this is needed but git uses it $pager end end