This is a follow-up to my first post on code quality with updates to various tools and some thoughts on POSIX threading.
In addition to Address Sanitizer, GCC and Clang also both support
Memory Sanitizer and Thread Sanitizer to detect uninitialized
memory accesses and threading issues. Compile with the -fsanitize=memory
or
-fsanitize=thread
options to enable them, respectively.
The Thread Sanitizer is still very much a beta tool, and while it has helped with some of my recent threaded code it needs better support for locks - see below.
The LGTM service is shutting down, with everything moving over to the CodeQL tools. Github offers built-in support for running CodeQL as an “action” (continuous integration), and if you were using LGTM they will create a pull request for any of your repositories to migrate to the new tools. Github also offers integration with Codacy but I haven’t been as happy with their tools.
Travis CI decided to go non-free, so I’ve moved all of my CI to Github Actions.
Over the last year I have needed to debug PAPPL threading issues,
mainly around its use of reader/writer locks (pthread_rwlock_t
) to control
access to the various objects. Thread Sanitizer was of limited use, so
I ended up developing simple logging macros along with an analysis program to
find deadlock, recursive locking, and locked-at-free issues. For example:
pthread_rwlock_rdlock(&obj->rwlock);
... access object values
pthread_rwlock_unlock(&obj->rwlock);
becomes:
_papplRWLockRead(obj);
... access object values
_papplRWUnlock(obj);
The macros log the thread, function, object address, and action (rdlock, wrlock,
or unlock) to stderr
and then call the corresponding pthread_rwlock_xxx
function. When the test program is done running, the parse-lock-log
program
then analyses the log and reports on any issues that are found.
I have been looking at re-implementing this as a shared library that can be loaded at runtime without re-compilation so it can be used more generally and also provide live reporting in addition to post execution analysis. I’ll post more if I make any headway on this…