DFSG NEW Queue

DFSG, Licensing & New Packages Team

Reviews for quanty

All reviews for this package from team members (across all versions).

Reviewer Version Allocated Started Status Completed Comment
awm 0.9.0-1 1 month, 14 days ago 1 month, 14 days ago accepted 1 month, 14 days ago Hi, Thanks for fixing all the previously advised issues. This is much better :-) There are still some relatively minor points that should probably be cleaned up... debian/copyright is not fully accurate in two places (both verifiable from the tree): - Missing copyright holder for the erfcx/Faddeeva code in Core/BasicMath/BasicMath_Complex.cpp. Lines ~859–996 embed Steven G. Johnson's erfcx ("written by Steven G. Johnson, October 2012", Comment block at line ~875) and the companion FaddeevaW/FaddeevaW_Im/w_im_y100 functions (same suite, referencing "algorithm 816" / "M. Zaghloul"). This is the well-known MIT-licensed code from the stevengj/Faddeeva repository. Two problems: the file carries no Copyright (c) 2012 Massachusetts Institute of Technology, Steven G. Johnson notice or MIT text (MIT requires reproducing the notice in substantial portions), and debian/copyright's Files: * stanza attributes the whole file to the Quanty contributor list under "CC-BY-4.0 or Expat" with no mention of Johnson/MIT. Since the licence (Expat) is compatible with the catch-all, this is not a DFSG violation, but the holder attribution is wrong/incomplete. Fix: add a Files: Core/BasicMath/BasicMath_Complex.cpp stanza with Copyright: 2012 Massachusetts Institute of Technology, Steven G. Johnson / License: Expat, and ask upstream to restore the notice. - Journal PDFs and their data are mislumped into the catch-all. debian/copyright Files: * claims every file in the tree is held by "Maurits W. Haverkort and Quanty contributors" under CC-BY-4.0/Expat. That is not supportable for third-party published works shipped in the tree: - Debug examples/Tutorials/20_NiO_Crystal_Field/NiO_Experiment/ and …/40_NiO_Ligand_Field/NiO_Experiment/: Phys. Rev. B 1998 Alders.pdf, J Synchrotron Rad 2009 Verbeni.pdf, plus companion data (NIXS_dd_JSR_16_469_2009, XAS_L23_PRB_57_11623_1998, RnlNi_Atomic_Hartree_Fock). Copyright for these journal articles belongs to APS/IUCr and their authors, not to the Quanty project, and their redistribution terms are unknown to me (binary, unreadable). The catch-all stanza therefore asserts a licence (CC-BY-4.0/Expat by Quanty) that the copyright holders have not granted. - Also Debug examples/Tutorials/20_NiO_Crystal_Field/NiO_data.zip (1.5 MB binary data) and the various out.wan / CrF4_WFC / ORCA .out files are data of unverified provenance swept into the same stanza (more a hygiene than a legal point). - Fix (recommended): Files-Excluded the two NiO_Experiment/ directories (and optionally the .pdf/.zip and debug artifacts), add a Repacksuffix: (the watch already has Dversionmangle: s/\+ds\d+//), so the next repack carries +ds. At minimum, debian/copyright must stop attributing these third-party works to the Quanty contributors. Also worth noting upstream-side (not a Debian blocker): the top-level LICENSE file says "The full license texts are provided in: LICENSE-CC-BY-4.0, LICENSE-MIT", but neither file exists in the tarball — only LICENSE is present. The Debian debian/copyright documents everything itself, so the package is fine; upstream's own license-file story is broken. Thanks! View
awm 0.0~git20250119191535.560ae00-1 1 month, 23 days ago 1 month, 23 days ago rejected 1 month, 23 days ago Hi, Several issues that should be fixed here... Licensing / DFSG ================ 1. Inaccurate debian/copyright — bundled Lua is mislabeled as CC-BY-4.0. debian/copyright claims Files: * ... License: CC-BY-4.0 for the whole upstream tree, but the tree embeds a full copy of the Lua 5.2.3 interpreter (LuaInterface/lua/), whose header states Copyright (C) 1994-2013 Lua.org, PUC-Rio (LuaInterface/lua/lua.h:26) and is MIT-licensed. The copyright file never mentions Lua or its MIT license, so it violates the machine-readable format's requirement to identify all files and their licenses. This bundled Lua is also actually compiled into the binary via #include "lua/includeLua.c" (LuaInterface/main.cpp:24). 2. The effective license deviates from the CC-BY-4.0 that debian/copyright claims. LICENSE adds an obligation beyond CC-BY-4.0: "For our scientific code this means you need to site the proper papers in scientific work that is based on this code or derivatives of this code." This is an extra, field-of-use-style requirement that debian/copyright neither reproduces nor flags, and it's the part of the licensing most likely to raise DFSG concerns. (CC-BY-4.0 itself is generally treated as DFSG-free, so the copyright inaccuracy + added clause are the real problems, not the base license per se.) If you can persuade them to, I think that upstream would be better to word it as something like "... this means you should cite the proper papers ...", which would not override the CC-BY-4.0, so you should propose that to them. It also corrects a spelling mistake (site -> cite). 3. debian/copyright upstream holder is just Copyright: Quanty with no years/names. I guess if you don't know the years, you don't, but perhaps they can be discovered from the upstream repository? Is that really the copyright holder, or should it list the paper authors? Packaging Quality Observations ============================== None of these are showstoppers, but some of them definitely seem worth fixing. The vendored Lua should be explained, ideally, and definitely exposed. 4. Test loop bug (duplicated in two files). debian/rules:13 iterates for i in $$(find TestScripts -name *.Quanty) but then only echoes $$i while always running the hardcoded TestScripts/BitMaskToIndex.Quanty. The exact same bug is repeated in debian/tests/control:1. Only one script is ever exercised. 5. Redundant tests. The same test runs in override_dh_auto_test (debian/rules:12) and as an autopkgtest (debian/tests/control) — duplicate coverage. 6. Vendored Lua, no system liblua, no Build-Depends. The build (debian/rules:10) compiles only LuaInterface/main.cpp and relies on the amalgamated #include of the bundled Lua, rather than linking Debian's liblua. debian/control has no lua Build-Depends, and debian/copyright doesn't disclose the vendored copy (see #1). liblua*-dev should be a Build-Depends and the bundled copy dropped or at least documented. 7. Hardcoded -O2 in debian/rules:10 after $(CXXFLAGS) This can conflict with the optimizer level chosen by dpkg-buildflags (e.g. a -O3 from CXXFLAGS would be overridden by the trailing -O2), so reproducible/optimized builds aren't fully honored. 8. debian/watch version scheme likely mismatches the changelog. watch uses Mode: git + Matching-Pattern: HEAD + Git-Date: %Y%m%d%H%M%S, but the changelog version is 0.0~git20250119191535.560ae00-1 with a Dversionmangle: s/\+ds\d+// that doesn't align with the ~git... suffix. uscan/gbp may fail to match and update the version properly. View

Back to Dashboard