Historical String Parsing

In some early Bourne-compatible shells, the second statement below does not do what you might initially suspect:

STRING1="This is a test"

STRING2=$STRING1

Most modern Bourne shells parse the right side of the assignment statement first (including any splitting on spaces), then expand the variable $STRING1, thus copying the complete value of STRING1 into STRING2.

Some older shells, however, may do the space splitting after expanding the variable. Such shells interpret the second statement as though you had typed the following:

STRING2=This is a test

as a two-part statement: an assignment statement (FIRST_ARGUMENT=This) followed by a command (is) with two arguments (a and test).

Because there is no semicolon between the assignment and the command, the shell treats this assignment statement as an attempt to modify the environment passed to the is command (a technique described in Overriding Environment Variables for Child Processes (Bourne Shell)). This is clearly not what you intended to do.

For maximum compatibility, you should always write such assignment statements like this:

STRING1="This is a test"

STRING2="$STRING1"

In any Bourne shell, this is interpreted correctly as:

STRING2="This is a test"

Similarly, in modern shells, quotation marks and other special characters are parsed before expansion. Thus, quotation marks inside a variable do not affect the splitting behavior. For example:

FOO="\"this is\" a test"

ls $FOO

is equivalent to:

ls \"this

ls is\"

ls a

ls test

In older Bourne shells, however, this may be misinterpreted as:

ls "this is"

ls a

ls test

In general, it is not worth the effort to support shells with this broken splitting behavior, and it is unlikely that you will encounter them; the modern splitting behavior has been common since the mid-1990s.