Shell Programming and Scripting

Files do not work that way, you can't insert in the middle. That's why we have text editors to do it for us.

If whatever's using this huge file can read from standard input, you can skip editing the file, just add it to the stream -- print it first, then the rest of the file.

Code:

(       echo "header line"
        exec cat hugefile ) | programwhichuseshugefile

This is implementation dependent.

Code:

$ strace ed file.txt 2>trace
q
$ grep "/tmp/" strace
open("/tmp/tmpfpHh7ZE", O_RDWR|O_CREAT|O_EXCL, 0600) = 3
unlink("/tmp/tmpfpHh7ZE")               = 0
$

Personally I would expect it to at least sometimes use a temp file -- it's effectively memory if the system is unloaded, and it's either that or crash on large files...

Yes that might be implementation dependent depending on the file size...big files might need an on disk scratchpad whereas small files could be loaded into one of the internal buffers...

In some implementations they use a temp file, in others they use memory if the file will fit in memory and a temp file if it won't, and in others they use memory and die on an ENOSPC error if the file won't fit into memory.

Even in cases where ed and ex edit a file entirely in memory, the entire 20GB file being discussed in this thread has to be read into memory and the updated file has to be written back to the file. There just plain is absolutely no way to add text to the start of a file without reading the entire file at least once and writing the entire contents of the updated file at least once with current BSD, UNIX, Linux, or Windows filesystems.