When converting molecules with wavy bonds (Bond::STEREOANY on double
bonds) through InChI and back, the stereo information was silently
dropped. This affected any workflow using InChI roundtrips for
canonicalization (e.g. with -SUU flag).
Two bugs in External/INCHI-API/inchi.cpp:
Reverse path (InchiToMol): The stereo0D processing loop skipped
INCHI_PARITY_UNDEFINED entries before they could reach the double bond
handler. The handler already had an else clause that correctly sets
Bond::STEREOANY, but it was never reached. Fix: only skip
INCHI_PARITY_NONE at the top level, and add a guard in the Tetrahedral
case to prevent UNDEFINED/UNKNOWN from incorrectly setting chirality.
Forward path (MolToInchi): STEREOANY double bonds were only handled by
collapsing the coordinates — InChI then produced no stereo annotation
under -SUU. Fix: also emit a stereo0D entry with INCHI_PARITY_UNKNOWN
parity so InChI's -SUU output correctly carries the "stereo unknown"
designation. StereoAtoms may be cleared for STEREOANY, so we locate
the two outer neighbors by iterating bonds.
New test testStereoAnyRoundtrip in External/INCHI-API/test.cpp covers
9 representative cases (Schiff base, oxime, cinnamic acid, chalcone,
crotonaldehyde, tamoxifen-like, retinal-like, plus two molecules with
a chiral center adjacent to the wavy bond).
Counts in rdkit/Chem/UnitTestInchi.py shift by 1 (689 same, 492
reasonable) because the new STEREOANY emission produces a more
accurate roundtrip for one entry in the test inventory.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
MolToInchi has called MolOps::Kekulize(*m, false) for years, but PR #9125 changed the default traversal to canonical=true. That pulls rankFragmentAtoms() and the canonicalization path into the InChI conversion even though the tested InChI outputs stay the same.
Validation:
- rdkit.Chem.UnitTestInchi passed before and after this change on upstream/master (18 tests, OK in both runs).
- No InChI output drift was observed between stock and patched builds on Regress/Data/mols.1000.sdf.gz, rdkit/Chem/test_data/pubchem-hard-set.sdf.gz, or the atom-order regression molecules added in Code/GraphMol/catch_graphmol.cpp.
Performance:
- Release_2026_03_1 Python MolToInchi on Regress/Data/mols.1000.sdf.gz improved from 0.40712s to 0.38871s median (-4.52%).
- Release_2026_03_1 rdinchi MolToInchi on the same dataset improved from 0.39755s to 0.37814s median (-4.88%).
- Release_2026_03_1 standalone C++ MolToInchi on /tmp/mols.1000.sdf improved from 7.66775s to 7.03474s wall time (-8.26%), from 20.57B to 19.04B cycles (-7.46%), and from 121.78M to 114.05M cache misses (-6.35%).
* External/INCHI-API: drop unnecessary mutex
A mutex was introduced in 6cfd34 from 2012 around all InChI library
calls, and InChI v1.05 release notes from 2017 announce that they fixed
race conditions to support multithreading.
We find that the InChI library mutex is no longer necessary with v1.05
and this patch removes it, which enables rdkit users to concurrently
call InChI-related methods.
This patch also updates the InChI multithreading tests to cover
`MolBlockToInchi`, so we have test coverage of concurrently making all
InChI calls.
I ran `testInchi` under asan 200x and all runs passed cleanly. I was
unable to configure the build toolchain with msan or tsan, so it is
unclear to me if those sanitizers would flag any issues.
* External/INCHI-API: add second multithreading test
This test covers case where immediate concurrent usage of the InChI
library leads to memory corruption.
* remove ASAN workaround (not needed for InChI 1.06)
clang-format run
update inchi version to 1.06 in docs
Co-authored-by: Jin Pan <jinxp1@gmail.com>
* - fix non-threaded *nix builds that currently fail because
boost flyweight introduces a dependency on pthreads
- make sure that mutexes and futures are only used when
RDK_BUILD_THREADSAFE_SSS is ON
- fix SubstructMatch failing test when RDK_BUILD_THREADSAFE_SSS is OFF
due to misplaced #ifdef's
- rename RDK_TEST_MULTITHREADED to RDK_THREADSAFE_SSS in inchi.cpp
(which is not a test)
* - the limitexternal Linux build is now single-threaded so we make
sure single-threaded builds do not break in the future
(suggestion from Greg)
* reverted unnecessary change to Code/GraphMol/FileParsers/testMultithreadedMolSupplier.cpp
Co-authored-by: Paolo Tosco <paolo.tosco@novartis.com>
* cleanup a bunch of g++ warnings
* make it work with clang
* remove some additional warnings based on CI builds
* fix that version number
* stop being verbose when building
* remove include from headers
* update implementation files
* completely remove BOOST_FOREACH (#7)
* convert those changes to use auto
* get rid of all usage of BOOST_FOREACH
Co-authored-by: Greg Landrum <greg.landrum@gmail.com>
* Fixes#3365
* update expected inchi results
Note that this actually increases the number of failures with one of the tests.
That's because I believe the expected InChIs to be wrong and these new results to be correct.
* run clang-tidy with modernize-use-default-member-init
* results from modernize-use-emplace
* one uniform initialization per line
otherwise SWIG is unhappy
Co-authored-by: Brian Kelley <fustigator@gmail.com>
* run clang-tidy with readability-braces-around-statements
clang-format the results
clean up all the parts that clang-tidy-8 broke
* fix problem on windows
* boost::thread mostly gone... still need to get rid of once
everything compiles
* replace boost::call_once
* remove link-time dependency on boost::thread
* first pass at using async
* switch to using async everywhere
* clean up the header a bit
clang-format
* clang-format
* add test for #1286
* backup
* fixes the specific failure, but needs more looking
* add an additional test
* update the inchi code to explicitly include queue
* fix a typo identified during the review
* add cis and trans to bond stereo
* compiles, does not work
* tests all pass
* Whitespace cleanup to recent changes.
* C++ test case for Bond::setStereo using Bond::STEREOCIS and Bond::STEREOTRANS
* Adding a PRECONDITION to Bond::setStereo to make sure the stereo atoms
are already specified if CIS or TRANS is being specified.
E/Z is technically defined by the topology of the molecule so the
stereo atoms are redundant (easier to understand and use!), but
ultimately redundant with the graph. However, CIS and TRANS is _only_
defined in this usage as the orientation of the atoms in the
getStereoAtoms vector.
* Exposing Bond::setStereo to Python and adding test cases to make sure
it can be used to set CIS/TRANS stereochemistry.
* verify substructure matching works
* Adding Bond::setStereoAtoms to C++ Bond class.
This allows setting the atoms to be considered for CIS or TRANS
directly without a much more costly determination of ranking that E/Z
requires.
* Wrap Bond::SetStereoAtoms into python with a new type of test case.
* docs
* Remove a bunch of warnings
The biggest change here is to treat the boost includes as system libraries.
This causes g++ to ignore warnings from boost.
This change still needs to be tested with clang and on Windows
* suppress a bunch of warnings on clang
