Section: New Results

Software engineering for infrastructure software

Our main work in this area has focused on driver porting. We aim at fully automating the backporting (or symmetrically forward porting) process: given any driver for one Linux kernel version, one would like to obtain a driver that has the same functionality for another kernel version. This requires identifying the changes that are needed, obtaining examples of how to carry these changes out, and inferring from these examples a change that is appropriate for the given driver code. We have carried out a preliminary study in this direction with David Lo of Singapore Management University; this work, published at ICSME 2016 [17], is limited to a port from one version to the next one, in the case where the amount of change required is limited to a single line of code.

More general automation of backporting requires more extensive search for relevant examples. This raises issues of scalability, because the Linux kernel code history is very large, and of expressivity, because we need to be able to express complex patterns to obtain change examples that are most relevant to a particular backporting problem. To this end, we have been adapted the notation used by Coccinelle, which describes how a change should be carried out, into a patch query language that allows describing patterns of changes that have been previously performed. The associated tool, Prequel, can find patches that match a particular pattern among several hundred thousand commits, often in tens of seconds [20]. This work is supported in part by OSADL, a consortium of companies, mostly in Germany, supporting the use and development of open source software in automation and other industries.

We will continue research in this direction over the next three years as part of the ANR PRCI ITrans project, awarded in 2016 and to be carried out in 2017-2020.