Version 0.5.0

Version 0.5.0#

Version 0.5.0 ships the new ccp application, a Django web app that replaces the Streamlit app and runs as a desktop window or in the browser. It also adds automatic digitizing of performance-curve PDFs, an agent API so AI agents can work with the case open in the app, a faster thermodynamic core and a more robust point solver.

Highlights#

  • The new ccp application (ccp_web) is included in the package. Install it (python >= 3.10) with pip install "ccp-performance[desktop]" (or [web] for the browser only) and start it with ccp-app.

  • ccp-app now opens the new application. The Streamlit app is still available with ccp-app --streamlit.

  • New ccp.digitize reads vendor performance-curve PDFs, with a curves digitizer page in the app.

  • AI agents can read and change the case open in the app, run its calculation and post notes, through ccp-web agent or an MCP server.

  • Point creation converges for every polytropic method and fluid of the Evans and Huble case set, and the calculation pipelines are faster.

  • head_isentropic / eff_isentropic return the exact isentropic head defined by PTC 10 and ISO 5389 instead of a closed-form approximation.

Running the application#

pip install "ccp-performance[desktop]"   # native window (pywebview)
pip install "ccp-performance[web]"       # browser only
ccp-app                                  # desktop window, or the browser if it cannot open
ccp-web serve --port 8000                # always the browser, on a chosen port
python -m ccp_web                        # same as ccp-app, where console scripts are not on the PATH

The web, desktop and hosted extras need python >= 3.10; on python 3.9 they install nothing, and ccp-web says so. ccp-app, ccp-web and python -m ccp_web open the desktop window when pywebview, a GUI toolkit for it and a display are available (on Linux the desktop extra installs no toolkit: install pywebview[qt], or pywebview[gtk] with the WebKit2GTK system libraries) and fall back to the browser otherwise; ccp-web desktop reports the missing piece instead. The app applies its migrations at start and keeps cases, files and results in the user data directory (CCP_DATA_DIR overrides it). REFPROP is optional: without RPPREFIX ccp uses CoolProp. The curves digitizer page also needs the digitize extra and the Tesseract OCR binary.

New features#

  • ccp web application (#170) — the ccp_web Django app ports the four Streamlit pages: straight-through and back-to-back performance tests (ASME PTC 10), curves conversion, and performance evaluation of plant data with online monitoring and the HTML report. Calculations run as background jobs with progress and cancel. .ccp files import and export losslessly in both apps. The app runs as a single-user desktop application (SQLite in the user data directory, secrets in the OS keyring) or, with CCP_PROFILE=hosted and the hosted extra, as a multi-user service (deploy/hosted/). New console script ccp-web (desktop, serve, worker, manage, agent) and new extras web, desktop and hosted.

  • Curves digitizing (#171) — ccp.digitize (digitize_pdf) reads a vendor performance-curve document and returns every operating case with its speed lines, legend speeds and suction conditions, from the PDF vector content or from the rendered image for scans. The app gets a curves digitizer page with a plot editor to review the traced points. Needs the new digitize extra and Tesseract OCR.

  • Agent API (#174) — AI agents can see what the user has open in the app, read inputs and results, change the open case, run its calculation and post notes the user sees. Every agent write is undoable from the app. The desktop launcher issues a fresh token at each start (--no-agent turns it off); the ccp-web agent CLI and the MCP server (ccp-web agent mcp) need only the base package. The agent skill gains a live_app recipe.

  • Faster thermodynamic core (#168) — states built without a phase get it imposed after the first flash (ccp.config.DEFAULT_PHASE), which removes the phase-stability analysis from the point solver flashes, and Impeller interpolates and shares points instead of deep-copying them.

Bug fixes#

  • Exact isentropic head (#172) — ccp.point.head_isentropic returned Schultz’s closed form \(\frac{n_s}{n_s-1}(p_2 v_{2s} - p_1 v_1)\) instead of the isentropic enthalpy rise \(h(p_2, s_1) - h_1\) that ASME PTC 10-1997 (para. 2.6.2) and ISO 5389 define. The closed form assumes a constant isentropic exponent along the path, which fails for dense or near-critical gases: for CO2 at 80 → 250 bar eff_isentropic gave 0.800 against an exact 0.779. The difference is negligible for near-ideal gases (0.07 % for N2 at low pressure). Point, f_schultz and every head_pol_* method are not affected.

  • Point solver robustness (#169) — every discharge closure is bracketed by the suction and the isentropic discharge and solved with ccp.roots.solve_monotone, so Point creation converges for every polytropic method and fluid of the Evans and Huble 2017 case set.

  • sandberg_colby_multistep points (#164) — Point with polytropic_method="sandberg_colby_multistep" no longer raises TypeError.

  • Point solver errors (#166) — a point that cannot be calculated raises ValueError naming the solver error instead of an AttributeError later on.

  • State.T_critical units (#167) — the units argument is honored; docs now state that the default polytropic method is sandberg_colby.

  • Lossless compressor test points (#170) — StraightThrough and BackToBack save points defined by a measured discharge with disch_p and disch_T, so a reloaded compressor converts its points exactly; older files still load.

API changes#

  • ccp-app opens the new application instead of the Streamlit app; the Streamlit app is started with ccp-app --streamlit (app extra) (#170).

  • Without the web extra (or on python 3.9), ccp-web prints how to install it instead of failing with ModuleNotFoundError; ccp-app runs the Streamlit app when only the app extra is installed and otherwise prints the same hint.

  • States built without a phase get ccp.config.DEFAULT_PHASE = "gas" imposed when the density check confirms it; set ccp.config.DEFAULT_PHASE = None to restore the previous behaviour, or pass phase=False to a single state (#168).

  • Impeller shares the points it is given instead of deep-copying them (#168).

  • Point raises a ValueError for a suction with vapour quality between 0 and 1 before solving; Evaluation instead emits a PhaseWarning for a wet cluster-centre suction and converts with the unconstrained state (#169, #170).

  • head_pol_sandberg_colby_multistep and eff_pol_sandberg_colby_multistep take disch_s as their third argument; nstep moved to the fourth position (#164).

  • check_similarity reports the PTC 10-2022 Reynolds envelope shared with Point.reynolds_limits, and the unused phi3 was removed (#169).

  • head_isentropic and eff_isentropic return the exact values; the previous closed form is available as head_isentropic_pvn / eff_isentropic_pvn. head_pol / eff_pol (the \(pv^n\) closed form without correction) are renamed head_pol_pvn / eff_pol_pvn; the old names are deprecated aliases that emit a DeprecationWarning (#172).

Infrastructure and packaging#

  • The wheel ships ccp_web (#175) — the published wheel and sdist contain the ccp_web package with its templates, static files and migrations. The release workflow builds the wheel from the sdist and checks both before upload: every file git tracks under ccp/ and ccp_web/ must be in them, and no compiled bytecode may be. A CI job installs the wheel with the web extra in a clean environment (python 3.10 and 3.13), applies the migrations and serves a page.

  • packaging is now a declared dependency (it was only installed through plotly), and the web extra requires django-allauth>=65.5, the first version that honours the hosted profile’s e-mail login settings. The license is declared as an SPDX expression (setuptools >= 77 to build), and the Python classifiers list 3.9 to 3.13.

  • CI runs the web app tests in both the desktop and the hosted profile (#170).

  • A PyInstaller spec for the desktop app (build/ccp_web.spec) and a Docker setup for hosted deployments (deploy/hosted/) were added to the repository (#170).