Why parser generator tools are (mostly) useless in static analysis

The author presents a decent overview of some of the issues related to source code static analysis from a parsing perspective, but makes a pretty weak argument that parser generators are useless for the task.

Static analysis is a broad term, compilers generally do static analysis as part of many optimizations. Here I think the author is referring to static analysis used in areas such as program verification (ie security).

Usually if one is implementing an application to provide static analysis services, one would like to support as many languages as possible, broadening the accessible market.

Ultimately, such static analysis needs to end up with an abstract syntax tree (AST) that represents the program when machine code is emitted. This can be a very complex task for some languages:

1. C/C++ the source code should be preprocessed, with all includes resolved identically to the system being analyzed. Processing alters the source file from what the user sees, making relating error message to source location much more involved.

2. Templated source files ie php, asp, jsp, ... the templated text has no relavance to the meaning of the program.

There are two problems the author states with parser generators that I agree with:

1. Lack of facilities to relate parsing error messages to relevant offsets in the source file. Users expect relevant error messages.

2. The curiosity that the most current language parsers are implemented in that language (C# Rosalyn, Java comiler api). Antlr is the exception among parser generators in that there is a large corpus of grammars for a variety of languages. Usually they're out of date with the latest language spec (C#). I think lack of grammars is the criticism here.