There is unfortunately a large number of behaviors in sed which are not adequately standardized. It is hard to articulate a general answer which collects all of these.
For your specific question, a simple function wrapper like this should suffice.

sedi () {
    case $(uname -s) in
        *[Dd]arwin* | *BSD* ) sed -i '' "$@";;
        *) sed -i "$@";;
    esac
}

As you discovered, details of the c command are also poorly standardized. I would simply suggest

sedi "2s/.*/$a/" file.dat

where obviously use a different separator if $a could contain a slash; or if you are using Bash, use ${a//[\/]/\\/} to backslash all slashes in the value.
Other behaviors which differ between implementations include extended regex support ( -r or -E or not available), the ability to pass in multiple script fragments with -e (the portable solution is to separate commands with newlines), the ability to pass standard input as the argument to -f , and the general syntax of several commands (does the filename argument of the r and w commands extend up to the next whitespace, semicolon, or newline? Can you put a command immediately adjacent to a brace? What are the precise semantics of backslashed newlines and where are they mandatory? Where are backslashes before command arguments mandatory?) and obviously regex flags and features.