The performance penalty of XML for program intermediate representations

Abstract

XML has been proposed as a file format for representing program intermediate forms such as abstract syntax trees (ASTs), and dependence graphs. Although XML offers many advantages over custom binary representations of the same information, there is an associated performance cost. This paper reports on the results of experiments aimed at quantifying this cost. Two representations are studied: the abstract syntax tree, and the system dependence graph (SDG). The performance penalty for using XML is found to be very significant for ASTs, and crippling for the SDG. The tradeoff between the performance and flexibility is discussed.