Fix CsvReader::current() TypeError at EOF (with regression tests)

[slashrsm]

Summary

Replaces / supersedes #3 with the same core fix plus a reproducible PHPUnit suite.

CsvReader::current() used do { … } while ($this->valid()) when column headers were set. SplFileObject::current() returns false at EOF. A do-while always runs its body once, so calling current() while the iterator is already invalid executed count(false) and threw:

TypeError: count(): Argument #1 ($value) must be of type Countable|array, false given

Why #3’s author could not reproduce

Simple “last line of a normal CSV” fixtures never call current() after the iterator is invalid. Foreach only calls current() when valid() is true, so the do-while body always started on a real line. Trailing newline / short last line / strict mode alone do not hit this path.

Real-world trigger (also noted on #3)

Port\Reader\OneToManyReader does:

$this->rightReader->next();
$rightRow = $this->rightReader->current(); // may be past last detail row
if ($this->rightReader->valid()) {
    $rightId = $this->getRowId($rightRow, …);
}

After the last matching detail row, next() leaves the detail CsvReader invalid, then current() is still invoked. That is exactly when the do-while blows up—so the last master row never completes. Matches the comment on #3 about OneToManyReader.

Direct calls after exhaustion (getRow(99), current() after iterator_to_array) hit the same bug.

Fix

Same idea as #3: only read a line when valid:

while ($this->valid()) {
    $line = $this->file->current();
    // …
}
return null;

Tests

Three new tests (all error with TypeError on unfixed master, pass with this patch):

1. testCurrentAtEndOfFileWithHeadersReturnsNull — call current() after exhausting the reader
2. testGetRowPastEndWithHeadersReturnsNull — getRow(99)
3. testOneToManyReaderConsumesLastDetailRowWithoutError — full master/detail join through the last rows

Verified locally:

• Without fix: 3 errors (count(false))
• With fix: 3 tests / assertions green
• Existing CsvReader tests still pass

Recommendation for #3

Close #3 in favor of this PR (same fix + tests + “Closes #3”). Do not merge #3 alone without tests.

Checklist

• Bug reproduced with a failing test on master
• Fix applied (while + valid check)
• Tests pass with fix
• CI green
• Review

Not merging yet — waiting on CI/review per housekeeping guidance.