This is correct, but maybe in a different way than you may have thought.

The shell parses the commandline in several distinct steps and one of these steps is the "expansion" of variables - that is, replacing them with their values. The problem is that all variables are expanded at the same time.

Therefore you need eval, because your construct implies a "second pass of expansion" to be applied onto your line. Unfortunately you cannot selectively restart a certain parsing step. You can only restart the whole process from the beginning. Therefore you perhaps need - to make your line work - to distinguish between the parts that are meant to be expanded in the first pass and the ones meant to be expanded in the second one. Notice that the shell "consumes" some characters and if you need to have them still there you need to escape them properly so that they are not consumed in the first pass:

Code:

$ echo "two words"
two words
$ echo \"two words\"
"two words"

$ eval echo "two words"
two words
$ eval echo \"two words\"
two words
$ eval echo \\"two words\\"
two words
$ eval echo \\\"two words\\\"
"two words"

To really understand all the differences between these lines pipe the echo-output into wc -w to count the words for every command variation.

A word about eval in general: stay away from it if you can! Most times you can find a solution where you do it differently and you don't need eval at all. This is (in almost every case) preferable, because eval is very dangerous. It is very easy to lose track of what really gets executed and why. Only if there is definitely no other way then use it as a last resort.

I hope this helps.

bakunin