Changelog

A small follow-up to 1.4.1, mostly fixing things that 1.4.1 broke.

• fixed Percentile calculation on small samples (<5 records) returned NaN instead of the actual value.
• fixed --parallel hanging on input streams smaller than the chunk size. Affected anyone piping short outputs.
• fixed Terminal width detection on macOS Sonoma when running inside Warp.

• fixed Crash when a JSON line contained a number larger than int64. We now fall back to string representation and emit a warning to stderr.
• fixed before / after filters now accept timestamps without timezone, assuming the system zone.
• changed The summary table now right-aligns numeric columns, which is what you wanted all along.

Adds support for compiled inputs and parallel parsing. This is the largest release since 1.0.

• added New compile subcommand that converts an input stream into a packed binary format. Repeated queries against the same dataset are 8–12× faster.
• added --parallel N flag for filter and summary. Order is not preserved when used.
• added gap analysis for timestamp aggregations: shows the largest gap in the stream and where it occurred.
• changed Default percentiles are now p50/p90/p99. The old defaults (p50/p95/p99) are still available via --percentiles 50,95,99.
• changed Minimum Go version bumped to 1.22 (was 1.21).

• added matches operator for regex filtering. Patterns are compiled once per query, not per record.
• added Configuration file support (~/.config/codissy/config.toml). Aliases turn out to be the killer feature.
• added Shell completion scripts for bash, zsh, fish (codissy completion <shell>).

• fixed Memory usage when streaming very large files. Previously the deduplication buffer was unbounded.
• fixed ANSI escape codes in log content are now stripped before parsing rather than leaking into table cells.

• added Support for the Common Log Format (Apache & Nginx access logs).
• added Time-bucketed summaries: summary --bucket 1m, --bucket 1h, etc.
• added --limit N on pick for stopping after N matches without consuming the rest of the stream.
• changed --no-color now also disables hyperlinks. Most terminals ignored them but a few rendered escape codes literally.

• added pick subcommand for projecting fields without aggregation.
• added JSON output mode (auto-enabled when stdout is not a TTY).
• added Logfmt input format. Detected automatically.

First stable release. Public API is now considered frozen until 2.0.

• Filter expressions with comparison, boolean, and string operators.
• Summary statistics with default percentiles.
• JSON input format with auto-detection.
• Coloured tables in TTY mode.

Last 0.x release. Started using it in production at $work for tailing the staging environment. Mostly held together.

Initial proof of concept. Read JSON from stdin, filter on a single field. The kind of thing you write on a Saturday and then can't stop tweaking.