Add hook to handle Content-Encoding about drakma

[hanshuebner]

Supporting other encodings would be a great addition, and additional dependencies are no big deal. As the dependency would probably require some external shared library, it would be good if it could be switched off at compile time for non-quicklisp users.

The hook would probably be installed in the wrap-stream (https://github.com/edicl/drakma/blob/master/request.lisp#L581), but I'm unsure whether a hook can be general enough, given that there may be chunking and encryption as well (i.e. does there need to be any flexibility with respect to where in the stack the hook gets to place its wrapper?).

Is a generic, public hook really useful, i.e. are there uses for it other than implementing decompression?

Would it be useful to be able to compress the request body as well? Is that even supported as per the HTTP standard?

[tmccombs]

The Content-Encoding and Transfer-Encoding headers only applies to responses.

Also ideally, it should be able to handle both the Content-Encoding and Transfer-Encoding headers.

As far as I can tell, decoding should always be done after encryption and chunking.

The only use case I can think of for having a general, public hook is that it allows users to add support for additional encoding types that are either non-standard or not implemented in drakma. That means that drakma could have support for gzip and deflate, but a user could add additional support for the less common pac200-gzip, or the un-official bzip2 if needed.

Another possible implementation would be to use a generic function that dispatches on the encoding type, then have methods for identity, gizp, defalte, etc. I think I prefer this method.

[hanshuebner]

A generic function that dispatches on the encoding type sounds compelling.
Can you come up with a patch for that?

[tmccombs]

Yes, I've started working on it.

[avodonosov]

It is an unfortunate fix, because it breaks backward compatibility.

I've just spent 3 hours debugging my application, to locate this problem.

I am stroing data on Amazon S3 and gzip it to save space. I retrieve my data with zs3:get-vector and ungzip it. Suddenly, ungzip started to signal an error. Turns out, with new drakma zs3:get-vector returns not the data I posted earlier, but ungzipped version of it.

I suggest to rollback this fix, until it's refactored in backward compatible fashion.

[hanshuebner]

Thayne, would you be able to take care of this?

2015-03-13 12:08 GMT+01:00 Anton Vodonosov [email protected]:

It is an unfortunate fix, because it breaks backward compatibility.

I've just spent 3 hours debugging my application, to locate this problem.

I am stroing data on Amazon S3 and gzip it to save space. I retrieve my
data with zs3:get-vector and ungzip it. Suddenly, ungzip started to
signal an error. Turns out, with new drakma zs3:get-vector returns not
the data I posted earlier, but ungzipped version of it.

I suggest to rollback this fix, until it's refactored in backward
compatible fashion.

Reply to this email directly or view it on GitHub
#45 (comment).

[tmccombs]

I can look at it tonight

[hanshuebner]

Thanks!

2015-03-13 15:59 GMT+01:00 Thayne McCombs [email protected]:

I can look at it tonight

Reply to this email directly or view it on GitHub
#45 (comment).

[stassats]

I don't see an issue here, if S3 sets the Content-Encoding to deflate, then it's fair game. An option to disable this could be added, but no backwards compatibility is necessary. Incidentally, there should be an option for easily specifying Accept-Encoding, probably defaulting to gzip,deflate

[avodonosov]

Fuzzy concept of "fair game" can't justify breaking someone's existing code.

Support of gzip and deflate is convenient, but doing so in backward incompatible way ads to degradation of CL ecosystem, which has lot of bugs and instabilities as it is. I regularly monitor QL libraries and every month something breaks (https://common-lisp.net/project/cl-test-grid/ql/).

Who is affected - any code already dealing with gzip or deflate encoding. It's not a theoretical speculation, my case is a concrete example of this.

[stassats]

Then "tough luck" is a better justification. Sometimes things have to be broken. It's hard to get API right the first time and foresee any future needs. Not breaking anything will result in a shit API full of crutches.

[avodonosov]

Generally speaking, if you need to improve an API, you introduce new API, instead of breaking the existing one.

If I wanted to improve drakma API, I would introduce new method drakma:http-request2 with better default behaviour and simpler parameters. Or, if the changes are very significant, I would created new library, drakma2. So the old tested code remains working using drakma:http-request, and the new code uses new function / library.

That way API evolves without and destructive network effect or efforts / time expenses. Almost always maintaining backward compatibility is easy and cheap (both for library author and the users).

About this particular case. Anyway, we need a parameter to disable / enable automatic gzip handling - I may want to download the gzipped version in order to back up it, or move to another location; in this case I don't want redundant ungzip / gzip processing.

And in my opinion the default should be backwards compatible. In the open source CL we don't have many resources, it takes years to fix even simple bugs, so deliberately breaking things is unacceptable.

Incidentally, there should be an option for easily specifying Accept-Encoding

It's easy to specify using :additional-headers

[hanshuebner]

Backwards compatibility has always only been a secondary goal with Edi's libraries. The reason for that is that maintaining backwards compatibility quickly becomes a burden for the library developer. The suggestion to add a new API for existing functionality with a different name is something that will not happen in Drakma. If you need an old version of a library, you need to get that old version.

[avodonosov]

OK.

For the record, new API for new functionality, instead of breaking the old API.

[hanshuebner]

If an existing API needs to be changed to support some new feature and it can't easily make it into a new API, then backwards compatibility will be broken. This will happen more often in non-modular, monolithic APIs like `DRAKMA:HTTP-REQUEST`, which might be a motivator for an API refactoring of DRAKMAs API to make it easier to change in incompatible ways. Decoding content automatically is the right thing to do for most users. Users who need undecoded content will need to change their code to adapt. A general utility library should do the right thing for most people by default. rather than cater for the special needs of a few, or for backwards compatibility. This is for the record, too.