sed: getting at the current file name

[Ralf Wildenhues]

Hi Paolo, bug-gnu-utils,

for operating on several files, GNU sed provides --separate which is
great; but sed is missing a way to get at the file name of the currently
processed file.  This would be very convenient for things such as:

  # find some bug, transform the found places to a vim quickfix file
  for f in $files; do
    sed -n "/$pattern_or_complex_script/"' {
              =
              p
            }' $f | sed "N; s/^/$f:/; s/\n/:/"
  done > quickfix-file
  vim -c 'cf quickfix-file'

where the script is too complex to be emulated by grep, or git grep.

With --separate and a way to print the current file name we could
replace the shell loop with two sed invocations.  For example, if a
command `F' would print the file name of the currently open file or `-',
plus a newline, then one could use:

    sed -ns "/$pattern_or_complex_script/"' {
              F
              =
              p
            }' $files | sed "N; N; s/\n/:/g" > quickfix-file

saving lots of processes.

Alternatively, F could operate on pattern space, say, append the file
name.  Not sure if it should prepend a newline in this case.

Would you accept such an extension?  Or tell me to use awk?
Currently, the only way I see to do something like this efficiently in
sed is with real weird shell hackery that in-band-signals the file name
into the sed script and also on the command line.

FWIW, this would help for example
<http://thread.gmane.org/gmane.comp.lib.gnulib.bugs/19448/focus=19451>.

Thanks,
Ralf