Message9832

Author jeff.allen
Recipients jeff.allen, zyasoft
Date 2015-04-14.07:00:41
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <1428994842.53.0.318581921341.issue2320@psf.upfronthosting.co.za>
In-reply-to
Content
I'm -1 on the whole integer file descriptors thing anyway. I see it present in CPython, but it's all a bit too FORTRAN for me.

We tried doing without them, and I think that nearly worked, but not quite. There was discussion of this on python-dev and BDFL support for this divergence, but it seems to have aged off everywhere. The essence was that integer FDs are a low-level intrusion from the platform and some kind of file-like object should always be preferred. Butthey keep popping up.

The idea that a file-like objects is convertible to an int is a step forward, but I don't see how we can complete the idea unless the reverse is also possible: each method that takes an fd is able to convert it to a file-like object through global mapping/function.

No need for a lecture here on the drawbacks of global shared data.

Short-term a skip if nt is sensible. You'll find other skips/ifs in the tests around use of fileno(), conditioned on is_jython that could now be conditional on os._name. But if inte fds are desirable for compatibility with CPython, then they're desirable cross-platform. At first sight, the Windows file handle is a substitute ... but I'm a tad uncomfortable: Java makes this stuff private for a reason.
History
Date User Action Args
2015-04-14 07:00:42jeff.allensetmessageid: <1428994842.53.0.318581921341.issue2320@psf.upfronthosting.co.za>
2015-04-14 07:00:42jeff.allensetrecipients: + jeff.allen, zyasoft
2015-04-14 07:00:42jeff.allenlinkissue2320 messages
2015-04-14 07:00:41jeff.allencreate