Mechanical verification of compiler correctness

A large investment is made in the development, testing, validation and veri cation of source code for critical applications. But there remains a semantic gap between the source code produced and the object code which is executed. Standards for the development of critical systems recognize this, and mandate either the use of a trusted compiler (one that has been proven to produce correct object code) or the demonstration that the object code is a correct re nement of the source code (a lengthy and complex process). This thesis examines: 1. the extent to which tool support is an essential ingredient in proofs of compiler correctness; 2. the extent to which these proofs may be automated; 3. the relationship between the complexity of the source language and the proof e ort required; 4. the scalability of a mechanical method of compiler veri cation. To do this, we present the development of the proof of correctness of a compiler for a small imperative language Tosca, targeted at an imaginary assembler Aida. The work presented here is an extension of the Z speci cation and hand proof presented in Stepney's book \High Integrity Compilation | A Case Study", recasting the speci cation and proofs within the framework of the PVS speci cation and veri cation system from SRI International. We also assess the lessons learnt in the translation of a hand treatment of the problem in a partial logic, to an automatic treatment in a total logic with a richly expressive type system; from hand proof to automatic checking, and the inter-relationship between the speci cation and the ability to automate the proof process.