The Plot isoline `.value` checks asserted at WithinAbs(value, 1e-10), which
is platform-fragile: libm transcendentals (exp/log/pow) differ in the last
few bits across glibc, Apple libm, and UCRT, and that propagates through
the EOS solvers into ~1e-9 to 1e-10 relative differences in the computed
range bounds. Reference data regenerated on macOS in #2749 therefore broke
the Linux CI.
Switch to `WithinAbs(v, abs) || WithinRel(v, 1e-8)`: keeps the absolute
tolerance as a near-zero safety net and adds a 1e-8 relative tolerance
that comfortably covers the observed cross-platform jitter (worst case
~5e-9 rel for p_isolines[4]). Updated the generator script and re-ran it
to regenerate src/CoolPropPlot.cpp.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* build: replace git submodules with CPM.cmake
All 11 submodules (Eigen, fmtlib, msgpack-c, rapidjson, IF97,
REFPROP-headers, multicomplex, Catch2, pybind11, ExcelAddinInstaller,
FindMathematica) are now fetched by CPM.cmake at configure time.
Set CPM_SOURCE_CACHE (e.g. ~/.cache/CPM) to share the download cache
across git worktrees and build directories — no more per-worktree
`git submodule update --init --recursive`.
Vendored deps that have no upstream release cycle (miniz, nlohmann-json,
incbin) remain in externals/ as before.
Source-level changes: angle-bracket includes for rapidjson, IF97, and
REFPROP-headers now that their directories are on the include path via
CPM-provided source dirs rather than relative paths from the repo root.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(python): wire up CPM.cmake in Python wrapper, fix include paths
The Python wrapper CMakeLists.txt was still referencing submodule paths
(externals/Eigen, externals/fmtlib, externals/msgpack-c) removed by the
CPM migration. Include CPM.cmake + dependencies.cmake from the root and
use the ${Pkg_SOURCE_DIR} variables instead. Also adds missing
rapidjson, IF97, and REFPROP_headers include dirs required by CoolProp
sources.
CI workflows drop the now-meaningless `submodules: recursive` checkout
option and add a CPM source-cache step to avoid re-downloading on each
macOS/Windows runner.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* docs: remove git submodule references after CPM.cmake migration
Dependencies are now managed by CPM.cmake (fetched automatically at
CMake configure time), so git submodules no longer exist.
- Drop `--recursive` from all `git clone` commands in Web docs and
wrapper READMEs (26 .rst/.md files)
- Remove `submodules: recursive` from all CI workflow checkout steps
(13 workflow files)
- Remove `git submodule foreach/update` calls from release.bsh and
delete the now-dead pybind11 security-workaround lines
- Drop `--recursive` from build_swigged_matlab.sh and gitMirror.bsh
- Update CONTRIBUTING.md: drop "and its submodules" phrasing
Changelog entries referencing old submodule PRs are left intact as
historical records. The --recursive in buildbot.rst is for the
Dockerfiles repo (unrelated) and is also left unchanged.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(ci): remove empty 'with:' blocks after submodule removal, apply clang-format
After removing 'submodules: recursive' from checkout steps, some workflow
files were left with dangling empty 'with:' blocks that GitHub Actions
rejects as workflow file errors. Remove the empty 'with:' in 9 workflows.
Also apply clang-format to fix spacing in REFPROPMixtureBackend.cpp,
HumidAirProp.cpp, and Helmholtz.cpp.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* Use ninja for the examples
Should help a little bit with doc building
* FIx some templates
* Fix template deduction error in count_x_for_y_many functions
The count_x_for_y_many and count_x_for_y_manyC functions were using
a single template parameter for both input (double) and output (size_t)
arrays, causing template deduction failures in Cython bindings.
Changed to use separate template parameters YContainer and CountContainer
to properly support different types for input y values and output counts.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* Fix types in PXD header too
* Docs should get ninja too
* Changes from clang-format
* Fix git diff argument order in clang-format script
The script was comparing PR_BRANCH to TARGET_BRANCH, which shows changes from the PR branch to the target (what's in target that's NOT in PR). For PR validation, we need the opposite: changes from target to PR (what's in PR that's NOT in target).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* Fix Cython template deduction errors in solve_for_x_manyC and count_x_for_y_manyC
Add explicit template parameters [double, size_t] to both function calls
to resolve template type deduction errors when compiling with Cython.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* Fix wheel building by removing return from void Cython template calls
The functions solve_for_x_manyC and count_x_for_y_manyC return void in C++.
When Cython sees a return statement with a void function call containing
template arguments with commas (e.g., [double, size_t]), it wraps the call
in the __Pyx_void_to_None macro. This macro is a simple preprocessor macro
that cannot handle the commas in template arguments, treating them as macro
argument separators instead.
The fix is to remove the return statements, making these simple void function
calls. This prevents the __Pyx_void_to_None wrapping and allows the wheel to
build successfully.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
* Expansions are fully wrapped, looking good. Next step is the set of expansions that is the 1D approximation
* Get 1D approx working via cython
* Count solutions
* SuperAncillary class is working
>1000x speedup for water
Time for C++!
* Superancillaries are working!
In C++, speedup is more than 2000x. In Python, more like 150x because of Python <-> C++ overhead
* Add pmax check for PQ superancillary calls
* Update tests
* Allow T limits to be obtained
* Implement get_fluid_parameter_double for getting superanc value
* Add tests for getting parameters from superanc
* Script for testing superancillaries for sphinx
* Microoptimizations; don't help speed
The limiting factor remains the clear function, which takes about 30 ns
* Add R125 superancillary
* Use the release from fastchebpure for the files
* Drop a .gitignore in the unzipped folder
* Update superancillary injection script
* Turn on superancillaries by default
* Missing header
* Many int conversions in superancillary
* Another int cast
* More annoying solution for boost iter max
* Fix warnings
* One more warning
* Clear up the calculation of rho
* Update docs_docker-build.yml
Use arm64 since the containers were built on mac
* Superfluous ;
* Update backend.py
* Get the critical points working for superancillaries
* Fix wrapping changes of xmin&xmax methods
* squelch warnings
* Version 0 of jupyter notebook for docs
* Try to add the notebook to the docs
* Add jupyter notebook for superancillary
* Lots of updates to superancillary notebook
* More updates to docs
* Skip pseudo-pure for superancillary docs
* Fix output of superancillary figures
* Add superancillary plots to docs for the page for each fluid
* Make a placeholder figure for fluids without superancillary
* Add superancillary plots to task list
* Bump to release fixing m-xylene
* Relax the location of the REFPROP stuff
* Change default name for R-1336mzz(E)
* No need for figures to be so large
* Don't need REFPROP setting
* Bump to fastchebpure release with methanol
* Benchmark caching options
* Benchmark more granularly
* Add the fast methods to public API for HEOS class
* Back to memset - can memset with 0 but no other value
* Fix how caching is managed in Helmholtz class
* Close to final implementation
Perhaps a tiny bit more optimization possible?
* Update function name
* Make message more accurate
* Fix init order
* Expose update_QT_pure_superanc to Python
* Fix when _reducing is set for pures
* Fix the post_update
* Indent
* Notebook
* Notebook
* Make ln(p) construction lazy
Only really matters for debug builds
* Also make reference non-const
* Inject superancillary for methanol
* Make the superancillary loading entirely lazy in debug
* Fix PH bug for Nitrogen
Closes#2470
* Force the clear to be called on SatL and SatV
To invalidate them at start
* Default is non-lazy superancillary loading
* Add CMake option to have lazy-loading superancillaries [skip ci]
Not a good idea unless doing very narrow testing
- Streamlined the build of the HTML documentation based on a private docker image.
- Invoke doxygen and Sphinx to create the final HTML pages.
- Employ a cache for the GitHub actions to store the output of the expensive tasks.
- Restructured the GitHub actions builders and introduced a naming scheme.
- Fixed a few typos in the docs.
- Automatically publish the documentation to GitHub pages.
This fixes many issues:
- Python packages can be built with XCode 10, closes#1782
- Handling of CMake options for OSX stdc++ is streamlined
- TR1 warnings from MSVC 2017 have been silenced
- Nightly builds fixed, closes#1778
- Documentation builds fixed, closes#1762
- Python source packages are updated with recent Cython, closes#1777
* autopep8 whitespace only:
--select="E101,W1,W2"
* revert single file: web2py_online.py
Should this be a .py file at all?
* revert file Tickets/1443.py because the diff looked strange:
maybe there is an encoding problem?
* Rename web2py_online.py to web2py_online.rst
