Original ticket http://projects.scipy.org/numpy/ticket/2184 on 2012-07-11 by trac user khaeru, assigned to unknown.
The documentation for genfromtxt() reads:
When the variables are named (either by a flexible dtype or with names, there must not be any header in the file (else a ValueError exception is raised).
If names is True, the field names are read from the first valid line after the first skip_header lines.
The cause of this seems to be in [https://github.com/numpy/numpy/blob/master/numpy/lib/npyio.py#L1347 numpy/lib/npyio.py at lines 1347-9]:
if names is True:
if comments in first_line:
first_line = asbytes('').join(first_line.split(comments)[1:])
The last line should read first_line = first_line.split(comments).
first_line = first_line.split(comments)
With the current code, the input line:
# Example comment line
will be transformed to:
Example comment line
resulting in columns named 'Example', 'comment' and 'line' (this is what the warning in the documentation is about).
But also the input line:
ColumnA ColumnB ColumnC # the column names precede this comment
the column names precede this comment
resulting in columns named 'the', 'column', 'names' …etc. In this instance actual column names present in the file are inappropriately discarded.
By taking the  portion of the split instead of [1:]:
while not first_values
trac user khaeru wrote on 2012-07-11
Sorry, bad title. Also, what's the difference between the Trac issues list and https://github.com/numpy/numpy/issues ?
Title changed from Remove to npyio.py: genfromtxt() handles comments incorrectly with names=True by trac user khaeru on 2012-07-11
npyio.py: genfromtxt() handles comments incorrectly with names=True
atmention:rgommers wrote on 2012-07-12
We opened Github issues only a few weeks ago, we're in the process of transitioning all Trac tickets to it. When that's done we'll close Trac, or make it read-only. For now you can use either one.
Suggested fix looks correct.
trac user khaeru wrote on 2012-07-12
Oh, I see — well, I also posted a branch with this fix and a pull request: numpy/numpy#351