tsize support

[allenap]

This adds tsize support. In doing so, I had to add an attribute to IReader - size - which returns the size of the file to be read in bytes.

[shylent]

To be honest, I would never make it a property. Why? Because it looks like attribute access, but it actually results in a filesystem hit (for the first time only, though, due to behaviour of t.p.filepath.FilePath.getsize). I would rather have it as a method.

Still, it is not that much of a problem, I think. More of a style thing.

[shylent]

I am ready to merge this. I had a little.. thing with size being a property of IReader. I am not sure how you are going to treat my comments on your code, but anyway this particuar issue is of no real importance. If you want size to be a property, so be it - I will merge.

[allenap]

Thanks for the review! I'll change the size property to a get_size() function, so that it's more obvious that work might be done (i.e. it can defer).

[allenap]

Mmm, actually that might be tricky when it comes to dealing with options. It might be easier to allow get_reader() and get_writer() to defer, but that might also be too lazy. What's your opinion?

[shylent]

First of all it didn't seem reasonable to me to return a Deferred from get_reader and / or get_writer, because it is not meant to do any actual work. On the other hand, now that I think of it, it actually might do some useful work (actually, it already does that - it opens the file right in the reader's __init__). So there is an excuse for returning a Deferred from get_reader / get_writer after all.

Returning a Deferred from get_size, though, is a different story, since it will require some changes in RemoteOrigin{Read,Write}Session.startProtocol (since processOptions will return a DeferredList).

It seems to me, that the first approach (making get_reader / get_writer) will be a little less messy to implement.

[allenap]

get_reader() and get_writer() can now defer. That was a little more work than I at first guessed, but I think it's still fairly clean.

[allenap]

If it helps, I've got a pre-canned diff for the last block of changes I've made.

[shylent]

Okay, I can't really work on it right now, but I will definitely review all
the relevant changes tomorrow.

Sorry for the delay.

On Wed, Jul 11, 2012 at 9:15 PM, Gavin Panella <
reply@reply.github.com

wrote:

If it helps, I've got a pre-canned diff for the last block of
changes I've made.

---

Reply to this email directly or view it on GitHub:
#5 (comment)

[allenap]

Ta! I know you're busy, so don't stress about it. It's now been packaged for Ubuntu and the package carries (or will) a patch to bring trunk forwards with my changes. When you're happy with those changes and they're landed we'll drop that patch.

[shylent]

To be honest, if I have such a situation (when a method may or may not defer), I assert that it must always defer, so that the interface is simplified. If the implementor does not need or want to return a Deferred, one can always use twisted.internet.defer.succeed. On the plus side, you automatically know that whatever is returned from the method, you can always call addCallback on it.

[allenap]

An advantage of making it optional is backward compatibility.

[shylent]

I'd rather not use inlineCallbacks in my code (IMO they complicate debugging and invite ambiguity), but that is purely a matter of taste (especially in this case).

[shylent]

I've left some comments, one of which might be of some value (or maybe not). Otherwise, looks good.

[allenap]

Thanks for all your comments; I agree with them all. I've had to work on some other, related, bits for a while, but I am going to get this done next week.

[allenap]

I've removed both inlineCallbacks on TFTP._startSession() and made get_reader() and get_writer() both always return a Deferred. I've added a helper decorator, tftp.utils.deferred, that wraps a function in a call to maybeDeferred.

[allenap]

Hi, if there's nothing else outstanding that needs doing, can you pull this now? The Ubuntu package is carrying an 800+ line diff against trunk and I'd like to get that to near zero. There are two more pull requests after this one, one medium sized, and another very small. Thanks, Gavin.

[shylent]

I've merged the pull request. However, I've replaced calls to unittest.TestCase.assertIsNone with assertTrue(foo is None), since they were failing on my python / twisted version (did it in master in my repo). Hope that is not a problem.

[allenap]

That's great, thanks for merging!