seirl's webzone   \_o< ~ ♫♩

me_irl

I'm Antoine Pietri, a Ph.D student living in Paris. I have a software engineering degree from the computer science school EPITA. I have various interests like programming, algorithms, languages and language theory, networks, tooling, and good abstractions.

You can find my resume here .

research

Software Heritage

I'm a Ph.D student working on the Software Heritage project at Inria on large-scale analysis of the source code available in the Software heritage archive, under the supervision of Roberto di Cosmo and Stefano Zacchiroli.

Click here to see the full subject of my thesis.

Enabling Big Code analysis on a very large source code corpus

Context

Software Heritage is an ambitious digital preservation initiative that is amassing unprecedented amounts of software source code from a variety of origins, and keeps track of all their evolution history, as captured by version control systems. This is leading to the creation of one of the largest source code corpus and is building the corresponding development graph where the nodes are the contents of the source files, directories and commit points, and the edges carry the name of the files, and the links between commits and directories.

The availability of this universal software source code knowledge base provides unique opportunities for what is now known as “Big Code” research: querying, analysing and extracting knowledge both from the contents of the data and from the structure of the commit graph. We expect that having a way to leverage the data contained in very large source code archives that include development history, like the Software Heritage archive, will pave the way to learn more about the evolution of software at a very large scale than it has ever been possible before.

The exploitation of such an unprecedented very large source code collection poses significant challenges in terms of data representation, query languages, system architecture and specialised algorithms.

The Software Heritage Merkle DAG
The Software Heritage Merkle Directed Acyclic Graph

Objectives of the thesis

The main focus of the thesis will be to find efficient ways to perform large scale computations not only on the content of the archive, but also on useful data that could be derived from it, like code diffs, releases, branch/merge history, and reuse patterns.

The current data representation structure of Software Heritage is a Merkle DAG, which allows efficient deduplication, verification, and indexing of the content of the archive. While this format is cost-efficient for preservation purposes, it has not been designed with large scale analysis in mind. The current design uses data indirections extensively, which adds considerable latency to read-intensive operations. An important part of our goal will be to explore more efficient ways of storing and indexing all the software artifacts currently contained in the DAG, to allow for intensive computations to take place on the archive.

As part of this quest for the most efficient data representation for such an archive, we will look for heuristics to find and isolate repetitive code snippets, as it would not only allow to deduplicate and analyze the content at a more fine-grained level, but also give valuable insights to implement efficient storage, retrieval, and analysis of those snippets.

Another important aspect that will be looked into is the interface for performing computations on the archive. As the Merkle DAG structure is pretty unique, we will investigate ways of performing efficient and expressive queries by working on a query language that operates on this structure. This language could either be an adaptation or an extension of relational algebra implementations like SQL, or a completely new domain-specific language.

As we expect this project to become an important gateway for scientists working on Big Code analysis, the thesis will have to work with efficient distributed architectures as an integral part of the project’s design. For instance, it should be easy and cost-efficient to make the analysis infrastructure scale-out to accommodate for load increases. At the same time, the deployment should be easy, for example allowing instantiation on state-of-the-art public cloud offering.

Finally, the thesis will necessarily include practical applications and experimental validation of the analysis approach, in order to show meaningful results and usage examples, and as a way to test the general usability of the system.

Location

The PhD will take place at Inria Paris, and the research will be performed inside the Software Heritage group, under the supervision of Dr. Stefano Zacchiroli and Prof. Roberto Di Cosmo.

Vcsn

I was previously working as a Research Student on the C++ finite state machine framework Vcsn in the LRDE laboratory, under the supervision of Akim Demaille and with help from Jacques Sakarovitch, where I worked on:

prologin

I have been an organizer of the French national programming contest Prologin since 2011. I was the president of the organization in 2014, and I now mostly work on the technical aspects of the contest.

I worked on a lot of really interesting projects there, with my fellow organizers Alexandre Macabies, Pierre Bourdon, Rémi Audebert and countless others:

I wrote algorithmic exercises and written subjects for the contest, some of them with Jill-Jênn. You can find the most interesting ones here (in French):

I also created the tournament games of the finals in 2013, 2014, 2015 and 2016.

tech skillz

This is a non-exhaustive list of technologies and stuff I like to use.

random projects

Here are some of the ones worth mentioning (some of them are school projects that grew out of hand):

In real FOSS projects, I never found the time to actually do more work than hanging around on their IRC channels and fixing some bugs. You might stumble upon some of my minor contributions in some projects, like CPython, BitlBee or WeeChat, but that's about it.

contact & links