Indeed it is best to keep the file original. Awk can be easily adjusted to work with the original file. For example an adjustment of Jotne's suggestion:

Code:

awk -F"[Tt][Rr]" '{gsub(/\n/,x); for (i=1;i<NF;i++) {p+=length($i); print ++a, p+1+(a-1)*2}}' RS=± file

Will maybe work with gawk and maybe mawk, since they have very good line limitations.

Also a perl solution like:

Code:

perl -0077 -ne 's/\n//g; print (++$c," ",(pos() +1 -2)."\n") while /tr/gi' file

But while it perhaps may be even less likely than awk to run into line length limitations, just like the awk approach it will read the entire file in memory, which with 200M records is at least a 200 MB footprint...

I came up with a similar approach to Don's, but it uses index() rather than match() and it works for variable length patterns:

Code:

awk -v pattern="tr" '
BEGIN {
  pat_width=length(pattern)
}

{
  curline=tolower($0)
  chunk=rest curline
  while (pos=index(chunk,pattern)) {
    relpos+=pos
    print ++count, basepos + relpos
    chunk=substr(chunk, pos+pat_width)
    relpos+=pat_width - 1
  } 
  relpos=1-pat_width
  rest=substr(curline, length(curline) - pat_width + 2)
  basepos+=length(curline)
}

' file

Also, with all the approaches so far, they will look for the next match AFTER last match.

This next approach will also find additional pattern that were already part of a previous match:

Code:

awk -v pattern="trt" '
BEGIN {
  pat_width=length(pattern)
}

{
  curline=tolower($0)
  chunk=rest curline
  while (pos=index(chunk,pattern)) {
    relpos+=pos
    print ++count, basepos + relpos
    chunk=substr(chunk, pos+1)
  } 
  rest=substr(curline, length(curline) - pat_width + 2)
  basepos+=length(curline)
  relpos=-pat_width+1
}

' file

If we take the last part of Don's example: trtRTrTR, when trying to match "try" it will find 3 matches, while the others find only two.

Output:

Code:

1 1
2 3
3 5
4 893
5 895
6 897

Whereas the previous (using the pattern "trt" ) will find:

Code:

1 1
2 5
3 893
4 897