TL;DR: --totals --file=/dev/null | grep 'Total bytes written'
Since tar doesn't do any compression, it's possible to know exact size of tar archive without reading the files - just by looking at their sizes. It's almost like arithmetical sum of all their sizes, but tar-specific - since tar also uses some blocks to encode file metadata and directory structure.
This can be useful, for example, if you're making a website with some files and want to let users download whole folders as archives without actually storing these archives on disk.
In that case you would probably use a script like this:
#!/bin/sh
echo "HTTP/1.0 200 OK"
echo
tar --create "path"
Which makes a basic HTTP response and pipes output of tar command.
But when downloading a file produced with such script you will notice absence of progress bar in your browser - because it doesn't know the size of the file it's downloading. To fix it, you want to supply Content-Length header, with the size of the file. And it must be exact size - otherwise browser will download incomplete file or report an error.
Turns out, it's possible to find the size of the tar archive before making the actual archive, and answer is in this email from 2007:
you just need to add --totals --file=/dev/null to the arguments of your tar command and look at the line containing Total bytes written text.
Like this:
echo -n "Content-Length: "
tar --create "path" --totals --file=/dev/null * 2>&1 | sed '/Total bytes written/!d;s/.*: \([0-9]*\) (.*/\1/'
As it's said in the email, it will tell you the size of resulting archive without reading actual files - just by looking at their sizes. It still needs to read all directories, however.
To make your download script produce files with nice names, you can use Content-Disposition header:
echo "Content-Disposition: filename=\"all.tar\""
And if you want to have better control of tree structure inside your archive, you might want to cd to target directory first:
tar --create -C "path" '*'
or even
tar --create -C "path/.." "$(basename "$path")"
so your archive contained only one item in its root, matching the name of the dir you're archiving.
Note that you probably must change both tar lines - the one which report the archive size and the one which makes the actual archive.
If the size advertised in Content-Length header doesn't match the actual archive size - it's better not to have it at all!