[Numpy-discussion] 1.5 release schedule proposal
Tue Jul 27 13:34:22 CDT 2010
Pauli Virtanen wrote:
>> These are a couple relevant tickets;
> The relevant thing for 1.5.x would probably be to check that fromfile/
> fromstring handle malformed data in a well-specified way, and being
> compatible with previous behavior.
> I don't think we want to start changing the behavior (error handling etc)
The error handling I refer to is internal -- what happens when the
routine that a specific dataype uses to try to parse the file fails --
that information is currently not passed up the chain -- so fixing that
is not a change to the API, unless you consider unspecified behavior in
the face of malformed data part of the API.
> or feature set of fromfile at this point. It's a low-level function, and
> meant to be one.
yes, it sure is -- I'm still convinced that adding newline handling (and
maybe comments) will massively improve its usefulness -- on the one
hand, it's low level, on the other hand, its only point is performance;
you can use any number of python routines to do the same job otherwise.
I think with a few changes, it could be a high performance routine that
could be used by higher level routines (such as loadtxt) to do more
> The correct solution is probably to improve the speed (and possibly
> extend the feature set) of `loadtxt`, especially as the previous thread
> started from the observation that although `loadtxt` did what was
> desired, it was too slow for big files.
That would be fine, too. Actually, as I dug into the fromfile code, I
almost decided that it was just too complicated, and it would be better
to simply write something else from scratch, maybe in Cython -- I'm
still on the fence about that. If I did write something from scratch it
would probably look something like loadtxt().
There was a thread on the Cython list recently about this.
Christopher Barker, Ph.D.
Emergency Response Division
NOAA/NOS/OR&R (206) 526-6959 voice
7600 Sand Point Way NE (206) 526-6329 fax
Seattle, WA 98115 (206) 526-6317 main reception
More information about the NumPy-Discussion