Background

When I initially implemented the multi-threaded parsing support in Atom, I hacked in a feature called a "parsing operation limit" in order to avoid unnecessary threading overhead in the case of fast parsing. At the time, I avoided using actual time as the limiting "unit", because I wanted to avoid introducing any system call overhead during parsing.

Improvements

In this PR, I've replaced the ts_parser_set_operation_limit function with a new function called ts_parser_set_timeout_micros. This new fuction allows you to provide a timeout value in microseconds instead of an abstract "operation limit".

I've also added a --timeout flag to the tree-sitter parse command so that we can easily test the behavior of timeouts at the command line.

Implementation

The new timeout functionality relies on the clock(3) POSIX API and the CLOCKS_PER_SEC constant. I initially found that calling clock() on every parse action imposed a huge amount of overhead, so I added a counter to ensure that clock() is only called every 100 parse actions. This change eliminated any observable performance overhead.