PEP 305 - CSV File API

Dave Cole djc at object-craft.com.au
Sun Feb 2 13:46:26 CET 2003


> Then MyRow has to understand that particular file format.  You could
> just as well do [MyRow(row) for row in csv.reader(file("some.csv"))] (or
> some iterable equivalent).
> 
> More interesting would be something like:
> 
> reader = csv.columnreader(file("some.csv"))
> 
> with columnreader being like:
> 
> class columnreader(object):
> 
>     def __init__(self, *args, **kw):
>         self.reader = reader(*args, **kw)
>         self.iter = iter(self.reader)
>         class Row:
> 
>             def __init__(self, row):
>                 self.row = row
> 
>             def __getitem__(self, item):
>                 try:
>                     # If they are trying integer access:
>                     return self.row[item]
>                 except TypeError:
>                     # else name-based access:
>                     return self.row[self.headers[item]]
> 
>         headers = self.iter.next()
>         self.headers = {}
>         for i, header in enumerate(headers):
>             self.headers[header] = i
>             # I'm not sure if this will work:
>             setattr(Row, header, property(eval('lambda self: self.row[%i]' % i)))
>         self.rowClass = Row
> 
>     def __iter__(self):
>         return self
> 
>     def next(self):
>         return self.rowClass(self.iter.next())
> 
> This isn't necessarily hugely fast.  For more ideas on that db_row I
> believe might be of interest, as it does this for SQL results, in a
> manner that supposedly is good both for speed and memory
> (http://opensource.theopalgroup.com/).  It's a similar problem,
> where you get back lists (if you want to be fast) where each item
> has the identical attributes, but those attributes aren't known
> until you start getting the objects, and you'd probably want to
> trade flexibility for speed anyway since the objects are generally
> intermediates.
> 
> Anyway, they do it with a C extension I believe.

If you were going to do it and you needed it to go fast, a C extension
type a bit like the PyStructSequence would be the way to go.

- Dave

-- 
http://www.object-craft.com.au






More information about the Python-list mailing list