Changelog

All notable changes to pyadps are documented in this file. The format is based on Keep a Changelog, and version numbers follow Semantic Versioning.

[Unreleased]

Critical: the Streamlit app would crash (server segfault, not a normal Python exception) on the first file upload for anyone installing via pip install pyadps. pyarrow was never a direct dependency — it came in transitively through streamlit’s unbounded pyarrow >= 7.0 constraint, so a fresh install resolved the newest release (25.0.0), which is incompatible with the pinned numpy~=1.26.4 inside pyarrow’s pandas-to-Arrow conversion (used internally by st.dataframe()). Fixed by declaring pyarrow = "~=17.0" explicitly. Existing dev-branch checkouts with an already-installed compatible pyarrow were never affected, which is why this didn’t surface in local testing.

New binary_reader module with xarray.Dataset output
xarray accessors for domain-specific operations (ds.fixed_leader, ds.variable_leader, …)
CF Convention compliant variable attributes
Comprehensive user documentation (Read the Docs)
Selectable export components (velocity, echo intensity, correlation, percent good) on the Write File page, replacing the earlier velocity-only/full-dataset choice
Short (u, v, w) velocity variable naming option alongside the CF-style long names
Processing pipeline flowchart on the Streamlit home page and in the docs, showing the six-step pipeline plus the decision logic within Time Diagnostics, Sensor Health, Signal Quality, and Profile Operations
“Background” section (motivation, deployment context, current coordinate-system limitation) on the README, Streamlit home page, and docs front page
“Processing Guidelines” section on the Web Application docs page, with per-page tips on what to check at each step

Replaced the ReadFile class with a function-based API (pyadps.read())
Migrated from custom data structures to xarray.Dataset throughout
Improved error handling with an ErrorCode enum
Three-beam mode in the Signal Quality QC pipeline is now scoped to the Percent Good check only; it previously also appeared on Correlation, Echo Intensity, and False Target, where it was either non-functional or not reliably grounded on ensemble-averaged data
Renamed the installed console scripts for a consistent, collision-resistant naming scheme: run-pyadps → pyadps-gui, run-auto → pyadps-auto, run-cat → pyadps-cat (new; combines multiple ADCP binary files)

Documentation examples calling ds.header.* right after a plain pyadps.read() call, which silently misreports file status since include_header defaults to False (installation.md, io/accessors.md, io/index.md)
Sample config.ini in processing/config.md used incorrect key names in the [QCTest], [ProfileTest], and [VelocityTest] sections that don’t match ProcessingConfig.from_ini() — a hand-edited file following the old example would have had those settings silently ignored
ProfileOperationRunner.regrid() documentation listed nonexistent parameters and the wrong default interpolation method
pd0_parser error code 5 documented under the wrong name (WRONG_RDIFILE_TYPE instead of WRONG_ADCPFILE_TYPE)
Write File page documentation describing the pre-redesign Export Data tab
pyadps-auto console script pointed at a module that no longer existed and crashed immediately on use; it now has a working CLI
Time Diagnostics’ “Fill Time Gaps” (forward-fill) could silently leave gaps as NaN instead of filling them when the optional bottleneck package wasn’t installed; it’s now a declared dependency

See the binary_reader documentation for migration instructions from v0.3.3.