Python 3.8 (pre-release), 218 bytes

n=[(t:=[x for s in l.split('|')if(x:=s.strip())],max(map(len,t)))for l in open(0)] del n[1] j=''.join f=lambda c:j(f'|{l[c]:{w}}'for l,w in n) print(f(0),j('|'+w*'-'for _,w in n),*map(f,range(1,len(n[0][0]))),sep='\n') 

Try it online!

-12 bytes thanks to basic code golfing advice by @cairdcoinheringaahing :)

-25 bytes thanks to @ovs by using open(0) instead of sys.stdin!

K (ngn/k), 84 73 67 59 bytes

-6 bytes from @rak1507's trim implementation (2(|(&\^:)_)/)

-19 bytes from golfing

{,'/("|",'++?[;1;"-"]{(|/#'x)$x^,""}2(|(&\^:)_)/'"|"\)'x_1} 

Try it online!

Takes a list of strings, one for each row. Returns a list of strings, without leading/trailing spaces, and without trailing pipes.

• x_1 drop the row at index 1 (i.e., the one with all the -'s)
• (...)' run the code in parenthesis on each row of the input
  • "|"\ split each row on pipes (i.e. get the cells)
  • 2(|(&\^:)_)/' strip leading and trailing whitespace from each cell's contents
  • {...x^,""} remove empty cells
  • {(|/#'x)$x...} pad the cells (of each row) with spaces so they are all the same length
  • ?[;1;"-"] insert a cell containing the character "-" in the proper place
  • ++ transpose each row to generate the correct number of "-"s, then transpose back
  • "|",' prepend a "|" to each cell
• ,'/ concatenate the result by column

Ruby 2.7, 144 bytes

->a{(a=(a.map{_1.scan /\| ?\K[^| ]*[^| -]/}-[[]]).map{|l|l.map{_1.ljust l.map(&:size).max}}.transpose.map{?|+_1*?|}).insert 1,a[0].tr("^|",?-)} 

Try it online!

Takes in an array of lines, and outputs array of lines through the last output format. TIO uses an older version of Ruby, whereas in Ruby 2.7, we've numbered parameters, which saves six bytes!

JavaScript (ES8), 204 bytes

Expects and returns an array of lines. Uses the last output format.

s=>(m=[,0]).map((r,y)=>m[r&&y].map((s,x)=>(r?s:"").padEnd(w[x],"-"[r])).join`|`,s.map(w=(s,y)=>s.replace(/\| *(.+?) *(?=\||$)/g,(_,s)=>((v=s.length)<w[y]?0:w[y]=v,m[x+=x==1]=m[x++]||[])[y]=s,y+=y<2,x=0))) 

Try it online!

Commented

Step 1

Extracting the content of all cells and saving them into the matrix m[].

s.map(w = // w is an object used to store the max. width in each row (s, y) => // for each string s at position y in the input array: s.replace( // match in s all occurrences of: /\| *(.+?) *(?=\||$)/g, // "\|" a pipe // " *" followed by optional whitespace // "(.+?)" followed by a non-greedy string // (this is the payload) // " *" followed by optional whitespace // "(?=\||$)" followed by either a pipe or the end of // the line (not captured) (_, s) => // for each payload string s found in there: ( (v = s.length) // define v as the length of this string < w[y] ? // if it's smaller than w[y]: 0 // do nothing : // else: w[y] = v, // update w[y] to v m[x += x == 1] = // increment x if x = 1 m[x++] || [] // if undefined, initialize m[x] to an empty array // increment x afterwards )[y] = s, // save s in m[x][y] // initialization: y += y < 2, // increment y if y = 0 or y = 1 x = 0 // start with x = 0 ) // end of replace() ) // end of map() 

Step 2

Building the output.

(m = [, 0]) // initialize m[] with m[1] = 0 (reserved for the separator lines) // (this is actually done before step 1) .map((r, y) => // for each row r[] at position y in m[]: m[r && y] // use m[0] if r = 0, or m[y] otherwise .map((s, x) => // for each string s at position x in this row: (r ? s : "") // use an empty string if r = 0, or s otherwise .padEnd( // pad it with: w[x], // w[x] characters "-"[r] // using hyphens if r = 0, or spaces otherwise ) // end of padEnd() ).join`|` // end of inner map(); join with pipes ) // end of outer map() 

PHP, 429 420 bytes

$a=preg_split("/ (\W* )?/",$argv[1]);foreach($a as&$v)$v=array_values(array_filter(explode('|',$v)));for($i=-1;++$i<max(count($a),count($a[0]));)for($j=-1;++$j<=$i;){$t=trim($a[$i][$j]);$u=$a[$i][$j]=trim($a[$j][$i]);$a[$j][$i]=$t;$m[$i]=max(strlen($t),$m[$i]);$m[$j]=max(strlen($u),$m[$j]);}for($k=-1;$b=$a[++$k];print$k?" ":" $s ")for($l=-1;$b[++$l]!=null;){printf("|%-$m[$l]s",$b[$l]);$s.='|'.str_repeat('-',$m[$l]);} 

Try it online!

I'm pretty sure it could be golfed a bit more, when I have time I'll try to max it

EDIT: saved 9 bytes by using an alias for $a[$k] and using double quotes for the printf. Still work to do but 420 is a good number ;)

ungolfed explanation (the outputted newlines also replaced by \n for readability):

$a=preg_split("/\n(\W*\n)?/",$argv[1]); //split input by line removing the dashes under the headers foreach($a as&$v)$v=array_values(array_filter(explode('|',$v))); //split each line at the pipes, remove empty cells and rearrange indexes if needed for($i=-1;++$i<max(count($a),count($a[0]));) //looping on largest array dimension for($j=-1;++$j<=$i;){ //swapping array values, trimming strings and we keep the largest length for each column $t=trim($a[$i][$j]); $u=$a[$i][$j]=trim($a[$j][$i]); $a[$j][$i]=$t; $m[$i]=max(strlen($t),$m[$i]); $m[$j]=max(strlen($u),$m[$j]); } for($k=-1;$b=$a[++$k];print$k?"\n":"\n$s\n") //output of the result with the line for($l=-1;$b[++$l]!=null;){ printf("|%-$m[$l]s",$b[$l]); $s.='|'.str_repeat('-',$m[$l]); } 

Charcoal, 79 71 61 bytes

ＷＳ¿¬⁼ιη«≔⟦⟧υＦΦ⪪ι|κ«≔⁻Ｅκμ⌕Ａκ ζ≔✂κ⌊ζ⊕⌈ζ¹κＰ⁺|κＭ⊕¬υ↓⊞υＬκ»ＭＬυ↑|⌈υ↑ 

Try it online! Link is to verbose version of code. Explanation:

ＷＳ¿¬⁼ιη« 

Loop over the input rows, skipping the separator row.

≔⟦⟧υ 

Start collecting the output column cell widths.

ＦΦ⪪ι|κ« 

Split the row on |s and loop over the non-empty cells.

≔⁻Ｅκμ⌕Ａκ ζ 

Remove the positions of the spaces from the list of all character indices.

≔✂κ⌊ζ⊕⌈ζ¹κ 

Get the trimmed cell text.

Ｐ⁺|κ 

Print the vertical line and text without moving the cursor.

Ｍ⊕¬υ↓ 

Move down one or two lines depending on whether this is the first cell.

⊞υＬκ 

Push the width of the trimmed text to the list.

»ＭＬυ↑ 

Move up to the separator row.

|⌈υ 

Print the separator with its vertical line.

↑ 

Move to the first row.

Perl 5, 269 228 bytes

sub f{@t=map[split/\s*\|\s*/],pop()=~/^.\s?(.*?)\s*$/mg;@T=map{$c=$_-1;[map$$_[$c],@t]}1..@{splice@t,1,1};@w=map{max map y///c,@$_}@t;splice@T,1,0,[map'-'x$_,@w];join'',map{(map{sprintf'|%*s',-$w[$_],$$r[$_]}0..$#{$r=$_}),$/}@T} 

Try it online!

sub f{ @t=map[split/\s*\|\s*/], #split at |, trim away spaces before and after pop()=~/^.\s?(.*?)\s*$/mg; #pick part of each line that matter from input @T=map{$c=$_-1;[map$$_[$c],@t]}#transpose input table @t 1..@{splice@t,1,1}; #cols @w=map{max map y|||c,@$_}@t; #find max width of cols in transposed table splice@T,1,0,[map'-'x$_,@w]; #insert second line with --- of each width join'', #join pieces of output into single string map{ #for each line in transposed table @T ( map { sprintf '|%*s', #string right padded... -$w[$_], #...to width of this column $$r[$_] #info in column number $_ } 0..$#{$r=$_} #0..last index of current line in @T ), $/ #newline \n (default) } @T #for each line in transposed table @T } 

05AB1E, 45 bytes

ā2ÊÏε'|¡õÚðδÚ}©øćU®ε€gà}©'-ךXšεεR®NèjR„| ì]» 

Input as a list of lines. Outputs with spaces, and without trailing |.

Try it online (the header | is to convert the input into a list of lines).

Explanation:

ā # Push a list in the range [1, (implicit) input-length] 2Ê # Check for each that they're NOT equal to 2 Ï # And only leave the lines in the input at truthy indices # (so we've removed the second line with the dashes) ε # Map each remaining line to: '|¡ '# Split it on "|" õÚ # Remove leading/trailing empty strings, if the line started and/or ended # with a "|" δ # Map over each part: ð Ú # Remove leading/trailing spaces }© # After the map: store the result in variable `®` (without popping) ø # Zip/transpose, swapping rows/columns ć # Pop and push remainder-list and first item separated to the stack U # Pop this first item, and store it in variable `X` ® # Push the list of variable `®` again ε # Map over each list: € # Map over each part: g # Pop and get the length of the part à # Pop and only leave the maximum length }© # After the map: store this as new value `®` (without popping) '-× '# Transform each integer in the list into that many "-" as string š # Prepend this list to the remainder-list Xš # And then prepend the extracted head `X` back as well ε # Map over each list: ε # Map over each part inside the list: R # Reverse the current part ® # Push the list of integers `®` Nè # Pop and index the map-index into it j # Pad leading spaces up to that many characters to the reversed part R # Then reverse it back so the spaces are trailing „| ì # And prepend "| " to the part ] # Close the nested maps » # Join each inner list by spaces, and then each string by newlines # (after which the result is output implicitly) 

Julia 1.0, 146 144 bytes

x->["|"*join(rpad.(i,l),"|") for i=zip(insert!.((n=[strip.(split(i,"|";keepempty=0>1)) for i=x[[1;3:end]]];),2,"-".^(l=maximum.(length,n)))...)] 

input and output are arrays of strings

Try it online!
All tests!

Edit: -2 bytes by replacing false by 0>1

APL (Dyalog Extended), 40 bytes

⊃1(,/↑,2↓⊢){'|','-'⍪⍢(1∘⊖)↑⌂deb⍵⊆⍨⍵≠⊃⍵}¨ 

Try it online!

Adám's method.

64 bytes(old)

{↑,/⍉↑{'|',¨¯1⌽(m∘↑¨1⌽⍵),⍨⊂'-'/⍨m←⌈/≢¨⍵}¨¯1⌽1↓1⌽⌂deb¨'|'(≠⊆⊢)¨⍵} 

Try it online!

Takes the input as a list of lines, outputs a character matrix.

-9 bytes using dfns.deb.

Explanation

'|'(≠⊆⊢)¨⍵ split each line on pipes(|)

⌂deb¨ trim whitespace from both sides of each cell

¯1⌽1↓1⌽ drop the row with hyphens

{'|',¨¯1⌽(m∘↑¨1⌽⍵),⍨⊂'-'/⍨m←⌈/≢¨⍵}¨ do the following for each row:

m←⌈/≢¨⍵ set m to the maximum cell length

⊂'-'/⍨ create a cell of hyphens that long

,⍨ join with

m∘↑¨1⌽⍵ input shifted by 1, with each cell padded to length m.

¯1⌽ shift back to original position

'|',¨ prepend a pipe to each cell

Finally,

⍉↑ convert to matrix, and transpose

↑,/ join each row's cells, and convert to matrix again