MIRAI is an abstract interpreter for the Rust compiler's mid-level intermediate representation (MIR). It is intended to become a widely used static analysis tool for Rust.
MIRAI can be used as a linter that finds panics that may be unintentional or are not the best way to terminate a program. This use case generally requires no annotations and is best realized by integrating MIRAI into a CI pipeline.
MIRAI can also be used to verify correctness properties. Such properties need to be encoded into annotations of the source program.
A related use is to better document an API via explicit precondition annotations and then use MIRAI to check that the annotations match the code.
Finally, MIRAI can be used to look for security bugs via taint analysis (information leaks, code injection bugs, etc.) and constant time analysis (information leaks via side channels). Unintentional (or ill-considered) panics can also become security problems (denial of service, undefined behavior).
You'll need to install MIRAI as described here.
Then use cargo mirai
to run MIRAI over your current package. This works much like cargo check
but uses MIRAI rather
than rustc to analyze the targets of your current package.
cargo mirai
does a top-down full-program path sensitive analysis of the
entry points of your
package. To analyze test functions instead, use cargo mirai --tests
.
This will likely produce some warnings. Some of these will be real issues (true positives) that you'll fix by changing the offending code. Other warnings will be due to limitations of MIRAI and you can silence them by adding annotations declared in this crate.
Once MIRAI gives your code a clean bill of health, your code will be better documented and more readable. Perhaps you'll also have found and fixed a few bugs.
You can use the environment variable MIRAI_FLAGS
to get cargo to provide command line options to MIRAI. The value is a
string which can contain any of the following flags:
--diag=default|verify|library|paranoid
: configures level of diagnostics. With default
MIRAI
will not report errors which are potential 'false positives'. With verify
it will point out
functions that may contain such errors. With library
it will require explicit preconditions.
With paranoid
it will flag any issue that may be an error.--single_func <name>
: the name of a specific function you want to analyze.--body_analysis_timeout <seconds>
: the maximum number of seconds to spend analyzing a function body.--call_graph_config <path_to_config>
: path to configuration file for call graph generator (see Call Graph Generator documentation). No call graph will be generated if this is not specified.--print_function_names
: just print the source location and fully qualified function signature of every function.--
: any arguments after this marker are passed on to rustc.You can get some insight into the inner workings of MIRAI by setting the verbosity level of log output to one of
warn
, info
, debug
, or trace
, via the environment variable MIRAI_LOG
.
See the developer guide for instructions on how to build, run and debug MIRAI.
See the CONTRIBUTING file for how to help out.
MIRAI is MIT licensed, as found in the LICENSE file.