Is the behavior of multi-field convert-with-row contractual?

[chrullrich]

petl.transform.conversions.convert() supports converting multiple fields in one call, and supports passing the whole row to the conversion function. It (currently) performs the conversions such that I can (in the simplest case) use a single convert() call to exchange two values because the second conversion function sees the original state of the row, not the result of the first conversion:

>>> import petl
>>> t = petl.io.base.fromcolumns([[0], [1]], header=['a','b'])
>>> t
+---+---+
| a | b |
+===+===+
| 0 | 1 |
+---+---+

>>> mapp = {"a": lambda a, row: row.b, "b": lambda b, row: row.a}
>>> t.convert(mapp, pass_row=True)
+---+---+
| a | b |
+===+===+
| 1 | 0 |
+---+---+

If the second conversion saw the result of the first one, both fields would be equal.

This is not spelled out in the documentation. Can I rely on this behavior, at least to the point that it would be mentioned in release notes if it ever changed?

[juarezr]

It would be interesting to clarify the desired behavior and add a test case for assuring that future changes don't break it.