myTech.Today

Automatic Syntax Error Recovery

November 17 2020
e-mail: laurie@tratt.net   twitter: laurencetratt

Programming is the best antidote to arrogance I've come across — I make so many errors that I am continually reminded of my own fallibility. Broadly speaking, I think of errors as severe or minor. Severe errors are where I have fundamentally misunderstood something about the system I am creating. Each severe error is a bespoke problem, often requiring custom tooling to understand and fix it. Minor errors, in contrast, are repetitive and quickly fixed. However, they’re also much more numerous than severe errors: even shaving a couple of seconds off of the time it takes a programmer to fix a class of minor errors is worthwhile when you consider how often they occur.

The most minor of minor errors, and also I suspect the most frequent, are syntax errors. They occur for three main reasons: mental sloppiness; inaccurate typing [1]; or an incomplete understanding of the language’s syntax. The latter case is generally part of a brief-ish learning phase we go through and I’m not sure what a good solution for it might look like. The former two cases, however, are extremely common. When I’ve made a small typo, what I want is the parser in my compiler or IDE to pinpoint the location of the syntax error accurately and then recover from it and continue as if I hadn’t made an error at all. Since compilation is often far from instantaneous, and I often make multiple errors (not just syntax errors), good quality syntax error recovery improves my programming efficiency.

Unfortunately, LR parsers – of which I am particularly fond – have a poor reputation for syntax error recovery. I’m going to show in this article that this isn’t inevitable, and that it’s possible to do surprisingly good automatic syntax error recovery for any LR grammar. If you want to know more details, you might be interested in the paper Lukas Diekmann and I recently published called Don't Panic! Better, Fewer, Syntax Errors for LR Parsers. The paper also has a fairly brief accompanying talk, if you find that sort of thing helpful: