Fix out-of-bounds std::vector access causing R session abort

[ssppkenny]

Summary

Fixes a std::vector out-of-bounds access in src/word2vec/lib/word2vec.cpp:72-74 that causes an R session crash when training word2vec models on Linux distributions with _GLIBCXX_ASSERTIONS enabled (Arch Linux, Fedora, etc.).

Root Cause

std::copy(&_trainMatrix[wordIndex * m_vectorSize],
          &_trainMatrix[(wordIndex + 1) * m_vectorSize],
          &v[0]);

When wordIndex is the last word in the vocabulary, (wordIndex + 1) * m_vectorSize == _trainMatrix.size(). The expression &_trainMatrix[size()] calls std::vector::operator[] with an out-of-bounds index, which triggers the assertion:

/usr/include/c++/16/bits/stl_vector.h:1253: Assertion __n < this->size() failed.

Using &_trainMatrix[index] to get a past-the-end pointer is undefined behavior in C++.

Fix

Replaced &_trainMatrix[index] with _trainMatrix.data() + index, which is standard-conforming — data() + size() yields a valid past-the-end pointer, while operator[](size()) does not.

Also replaced &v[0] with v.data() for consistency.

Reproduction

library(word2vec)
word2vec(x = c("i saw the queen yesterday"), type = "cbow", dim = 15, iter = 20)

On Arch Linux / Fedora, this crashes the R session immediately with a std::vector assertion failure.

Testing

Verified the fix by rebuilding the package from source and running word2vec training and prediction — no crash, correct embeddings returned.

Related

See #(issue-number) for the original bug report.