When trying to generate isomeric or canonical smiles from SDF with large numbers of atoms the error:

An InChI could not be generated and used to canonise SMILES: null
Could not generate InChI Numbers: Too many atoms [did you forget 'LargeMolecules' switch?]

is thrown. See Mailing List

Attaching detailed explanation from Andrew Dalke of the issue below:

CDK uses InChI to generate absolute SMILES. Here's a comment from the code:

 * Create a absolute SMILES generator. Unique SMILES uses the InChI to
 * canonise SMILES and encodes isotope or stereo-chemistry. The InChI
 * module is not a dependency of the SMILES module but should be present
 * on the classpath when generation absolute SMILES.

If you remove either the SmiFlavor.Canonical or the SmiFlavor.Isomeric bit flag from your output flavor then you'll get a SMILES, though it won't be an absolute SMILES.

More specifically, CDK uses InChI to generate the atom labels used during canonical SMILES generation, in cdk/smiles/SmilesGenerator.java there's a code path which looks like:

        // apply the canonical labelling
        if (SmiFlavor.isSet(flavour, SmiFlavor.Canonical)) {

            // determine the output order
            int[] labels = labels(flavour, molecule);

where the labels() is:

private static int[] labels(int flavour, final IAtomContainer molecule) throws CDKException {
    // FIXME: use SmiOpt.InChiLabelling
    long[] labels = SmiFlavor.isSet(flavour, SmiFlavor.Isomeric) ? inchiNumbers(molecule)
            : Canon.label(molecule,
                          GraphUtil.toAdjList(molecule),
                          createComparator(molecule, flavour));

Thus, if SmiFlavor.Canonical and SmiFlavor.Isomeric are set, it ends up using code in cdk/graph/invariant/InChINumbersTools.java which configures InChI to do the atom order assignments, via the 'auxiliary information':

public static long[] getNumbers(IAtomContainer atomContainer) throws CDKException {
    String aux = auxInfo(atomContainer, new InchiFlag[0]);
  ...

static String auxInfo(IAtomContainer container, InchiFlag... flags) throws CDKException {
    InChIGeneratorFactory factory = InChIGeneratorFactory.getInstance();
    boolean org = factory.getIgnoreAromaticBonds();
    factory.setIgnoreAromaticBonds(true);
    InChIGenerator gen = factory.getInChIGenerator(container, flags);
    factory.setIgnoreAromaticBonds(org); // an option on the singleton so we should reset for others
    if (gen.getStatus() == InchiStatus.ERROR)
        throw new CDKException("Could not generate InChI Numbers: " + gen.getMessage());
    return gen.getAuxInfo();

That calls into the InChI, which has the check (actually, it's in a few places, all with the same idea):

max_num_at = ip->bLargeMolecules ? MAX_ATOMS : NORMALLY_ALLOWED_INP_MAX_ATOMS;
if (nNumAtoms >= max_num_at)
{
    TREAT_ERR( *err, 0, "Too many atoms [did you forget 'LargeMolecules' switch?]" );
    *err = 70;
    orig_inp_data->num_inp_atoms = -1;
    goto err_exit;
}

where

#define MAX_ATOMS 32766
#define NORMALLY_ALLOWED_INP_MAX_ATOMS 1024

The InChI flag is enabled with the flag 'LargeMolecules', https://github.com/dan2097/jna-inchi/blob/master/jna-inchi-api/src/main/java/io/github/dan2097/jnainchi/InchiFlag.java#L47

/** Allows input of molecules up to 32767 atoms [Produces 'InChI=1B' indicating beta status of resulting identifiers]*/

so it appears that changing cdk/graph/invariant/InChINumbersTools.java line 49 from:

    String aux = auxInfo(atomContainer, new InchiFlag[0]);

to have LargeMolecules in that 'new InchiFlag' would make this work.

However, I'm not a Java developer and don't know how to make this change nor test it. I can say it does not seem to be user-configurable.

I am a Python developer, and I can reproduce the error using my 'chemfp translate' tool, which uses a Java/Python bridge to work with the CDK. The following uses RDKit to translate a FASTA sequence to an SDF with 1079 atoms:

% python -c 'print(">megatryp\n" + "W"*77 + "\n")' | chemfp translate -T rdkit --in fasta --out sdf | head -6
megatryp
RDKit

0 0 0 0 0 0 0 0 0 0999 V3000
M V30 BEGIN CTAB
M V30 COUNTS 1079 1232 0 0 0

I can have it go from FASTA to SDF using RDKit then have CDK read the SDF to produce the SMILES generation failure:

% python -c 'print(">megatryp\n" + "W"*77 + "\n")' | chemfp translate -T rdkit --in fasta --via sdf -U cdk --out smi
Error: CDK cannot create the SMILES string (input title='megatryp'): An InChI could not be generated and used to canonise SMILES: null, file '', line 1, record #1: first line is '>megatryp'. Skipping.

(the --via defaults to 'sdf' so I'll omit that in the rest).

I can configure CDK SMILES writer to use the Default flavor, but without the 'Canonical' option, to show that work-around gives a (non-canonical) SMILES:

% python -c 'print(">megatryp\n" + "W"*77 + "\n")' | chemfp translate -T rdkit --in fasta -U cdk --out smi -W flavor=Default,-Canonical | fold | head -2
NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)
NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)NC(C(=O)

Here I'll disable Isomeric instead, so it should be canonical but not isomeric, which might be okay for you:

% python -c 'print(">megatryp\n" + "W"*77 + "\n")' | chemfp translate -T rdkit --in fasta -U cdk --out smi -W flavor=Default,-Isomeric | fold | head -2
O=C(O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(
NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(NC(=O)C(

That's the flavor you pass into SmilesGenerator().