How to use precomputed metric/kernels?

What is the correct way of using precomputed metric (or kernel) with StackingCVClassifier and StackingCVRegressor functions? Below is a minimal modified version of your first StackingCVClassifier example code using precomputed metric with KNeighborsClassifier (three flavors). It won't work and gives an error:
"ValueError: Precomputed metric requires shape (n_queries, n_indexed). Got (51, 150) for 48 indexed."

from sklearn import datasets
from sklearn import model_selection
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from mlxtend.classifier import StackingCVClassifier
from sklearn.model_selection import GridSearchCV
import numpy as np

iris = datasets.load_iris()
X, y = iris.data[:, 1:3], iris.target

RANDOM_SEED = 42

clf1 = KNeighborsClassifier(n_neighbors=1,metric='precomputed')
clf2 = KNeighborsClassifier(n_neighbors=3,metric='precomputed')
clf3 = KNeighborsClassifier(n_neighbors=5,metric='precomputed')
lr = LogisticRegression()

np.random.seed(RANDOM_SEED)
sclf = StackingCVClassifier(classifiers=[clf1, clf2, clf3],
                            meta_classifier=lr)

print('3-fold cross validation:\n')

# create pre-computed distance matrix XX
n=X.shape[0]
XX=np.zeros((n,n))
for i in range(n):
    for j in range(n):
        XX[i,j] = np.mean(np.square((X[i,:]-X[j,:])))

# verify that GridSearchCV works
GS_clf = GridSearchCV(clf1, {'n_neighbors':[1,3,5,7,9,11]},cv=10)
GS_clf.fit(XX,y)
print(GS_clf.best_estimator_)

for clf, label in zip([clf1, clf2, clf3, sclf],
                      ['KNN1',
                       'KNN2',
                       'KNN3',
                       'StackingClassifier']):

    scores = model_selection.cross_val_score(clf, XX, y,cv=3, scoring='accuracy')
    print("Accuracy: %0.2f (+/- %0.2f) [%s]" % (scores.mean(), scores.std(), label))

Start by reproducing the failure at the StackingCVClassifier entry point with the provided Iris example and precomputed distance matrix, then compare it with StackingCVRegressor. Done should be established by determining whether precomputed metrics can be supported without the (51, 150) versus 48 indexed shape error, and by covering the behavior with tests.