v5.24.0 new features + optimizations¶

Changed¶

Now parproc streams/iterates only Result objects. Using the results stream for passing other information led to complicated code.
All update notifications in crross-process update of barz bars go through a packetz queue. Now PacketzQueue is JSON-serializable for that purpose. The actual payload of a queue passed across tasks or processes is the path to the file-based queue.
Synchronization through a queue is done by a barz.BarBroker which is opt-in for users of the library. BarBroker uses a daemon thread to read the queue and call update_row() on the Multi.
Multi.height starts at -1 as the magic number to preserve the line position before drawing of the first bar.
Keeping the files for queues happeens on a per-queue basis with Packetz(keep=True). The files are stored uner ./packetz/ by default. The format of the filenames is aaaa-HH-mmmm.pkz.jsonl, where aaaa is four letters representing the Unix timestamp day in base len(strings.ascii_loweercase), HH is the UTC hour, and mmmm is the fraction of the hour in tenth mof millisecond. The format of the files is JSONL, with one line of flattened JSON per packet. Keeping all the files for queues may be activated by setting the PACKETZ_KEEP environment variable to non-falsy.
Optimized PacketzQueue.receive() to read by line and retry if a line of JSONL was incomplete (no newline at the end).

A round of removing memory and transformation bottlenecks has OGoPEGo generating the fastest parsers among the TatSu family of PEG parser generators. When the CLI combines the parser speed with Go’s implementation of concurrency the result feels instantaneous.