Bash – Delete content of file but keep name and hierarchy

Update: fixed a typo in the script: changed print $NF to print $3; also tidied things up, and added some comments.

Assuming file names do not contain \n, the following prints out a sorted list which breaks (as in: section control breaks) at unique file name, unique md5sum, and shows the corresponding group of file paths.

#!/bin/bash

# Choose which script to use for the final awk step 
out_script=out_all

# Print all duplicated file names, even when md5sum is the same 
out_all='{ if( p1 != $1 ) { print nl $1; print I $2 }
      else if( p2 != $2 ) { print I $2 }
      print I I $3; p1=$1; p2=$2; nl="\n" }
   END { printf nl}'

# Print only duplicated file names which have multiple md5sums.
out_only='{ if( p1 != $1 ) { if( multi ) { print pend }
                             multi=0; pend=$1 "\n" I $2 "\n" }
       else if( p2 != $2 ) { multi++; pend=pend I $2 "\n" } 
       pend=pend I I $3 "\n"; p1=$1; p2=$2 } 
   END { if( multi ) print pend }'

# The main pipeline 
find "${1:-.}" -type f -name '*' |  # awk for duplicate names
awk -F/ '{ if( name[$NF] ) { dname[$NF]++ }
           name[$NF]=name[$NF] $0 "\n" } 
     END { for( d in dname ) { printf name[d] } 
   }' |                             # standard md5sum output 
xargs -d'\n' md5sum |               # " "==text, "*"==binary
sed 's/ [ *]/\x00/' |               # prefix with file name  
awk -F/ '{ print $3 "\x00" $0 }' |  # sort by name. md5sum, path 
sort |                              # awk to print result
awk -F"\x00" -v"I=   " "${!out_script}"

Output showing only file names with multiple md5s

afile.html
   53232474d80cf50b606069a821374a0a
      ./test/afile.html
      ./test/dir.svn/afile.html
   6b1b4b5b7aa12cdbcc72a16215990417
      ./test/dir.svn/dir.show/afile.html

Output showing all files with the same name.

afile.html
   53232474d80cf50b606069a821374a0a
      ./test/afile.html
      ./test/dir.svn/afile.html
   6b1b4b5b7aa12cdbcc72a16215990417
      ./test/dir.svn/dir.show/afile.html

fi    le.html
   53232474d80cf50b606069a821374a0a
      ./test/dir.svn/dir.show/fi    le.html
      ./test/dir.svn/dir.svn/fi    le.html

file.html
   53232474d80cf50b606069a821374a0a
      ./test/dir.show/dir.show/file.html
      ./test/dir.show/dir.svn/file.html

file.svn
   53232474d80cf50b606069a821374a0a
      ./test/dir.show/dir.show/file.svn
      ./test/dir.show/dir.svn/file.svn
      ./test/dir.svn/dir.show/file.svn
      ./test/dir.svn/dir.svn/file.svn

file.txt
   53232474d80cf50b606069a821374a0a
      ./test/dir.show/dir.show/file.txt
      ./test/dir.show/dir.svn/file.txt
      ./test/dir.svn/dir.show/file.txt
      ./test/dir.svn/dir.svn/file.txt

The easy way is with zsh. You can use glob qualifiers to match files according to criteria such as their type and size. The wildcard pattern **/ matches any level of subdirectories. The history modifiers h and t are easy ways of extracting the directory and the base part of a filename. Call mkdir -p to create the directories when needed.

cd dirA
for x in **/*(.Lm+300); do
  mkdir -p ../dirB/$x:h &&
    mv -- $x ../dirB/$x
done

The portable way is with find. Use -exec to invoke a shell snippet for every file.

cd dirA
find . -type f -size +300000k -exec sh -c 'for x do
  mkdir -p "../dirB/${x%/*}"
  mv "$x" "../dirB/$x"
done' sh {} +

Parallelization is rarely useful for input/output: it lets you take advantage of multiple CPUs but the CPU is rarely a bottleneck in I/O.