Skip to content

tych0/xcffib

Repository files navigation

xcffib Build Status

xcffib is intended to be a (mostly) drop-in replacement for xpyb, the original python binding for xcb.

Installation

For most end users of software that depends on xcffib or developers writing code against xcffib, you can use the version of xcffib on pypi. To install it, you'll need libxcb's headers and libxcb-render's headers (these are available via sudo apt-get install libxcb-render0-dev on Ubuntu). Once you have the C headers installed, you can just pip install xcffib.

If you're interested in doing development, read on...

Development dependencies

You should be able to install all the language deps from hackage or pip. .github/workflows/ci.yaml has an example of how to install the dependencies on Ubuntu flavors.

Hacking

See the Makefile for examples on how to run the tests. Your contribution should at pass make check before it can be merged. The newtests make target can be used to regenerate expected haskell test data if the tests are failing because you made a change to the generated python code.

Hacking on new xcbproto versions

Sometimes (more often recently), xcbproto makes some updates that we need to do some work for. These often require some updates to xcb-types as well. First, hack your changes into xcb-types and cabal install them, then git clone the version of xcbproto you want to somewhere, e.g. ~/packages:

~/packages $ git clone https://gitlab.freedesktop.org/xorg/proto/xcbproto.git

Finally, you can build/test xcffib against this custom version of xcb-{proto|types} with:

make XCBDIR=~/packages/xcbproto/src check

Hacking on new xcb-types versions

To go along with new xcbproto elements, sometimes you need to hack on newer versions of xcb-types. Newer cabals require you to do something like:

echo packages: ../xcb-types/xcb-types.cabal ./xcffib.cabal > cabal.project

In order to find locally modified versions of xcb-types.

Differences

In general, you should s/xcb/xcffib/g. Explicit differences are listed below, however I don't think these will prevent any porting, because these were either not public APIs, or not actually generated (in the case of the exceptions) by xpyb. I think most porting should Just Work via the regex above.

  • xcb.Exception is spelled xcffib.XcffibException and is also a parent of all exceptions generated by xcffib.
  • xcb.ConnectException is gone, it was unused
  • xcffib.ConnectionException is raised on connection errors
  • xcb.Iterator is gone; similar functionality is implemented by xcffib.pack_list.
  • xcb.Request is gone. It was an entirely internal and unnecessary interface.
  • xcffib.Connection.send_request takes slightly different (but more sensible) arguments.
  • Everywhere xcbproto says char, xcffib uses a char. That means on input for a <list type="char"/>, you can use a python string literal. xcffib also gives you a string of length 1 out for each element in such a list, instead of an int. Finally, there is a helper method called to_string on xcffib.List, to convert these string-like things into native strings. This means that for things like xproto.STR, you can just do the_str.name.to_string() instead of ''.join(map(chr, the_str.name)).
  • As above, void is also packed/unpacked as chars, since the convention is to use it as string data, e.g. in xproto.ChangeProperty.
  • The submodule xcb is gone. The top module re-exported all these constants anyway, so they live there now. i.e. xcb.xcb.CurrentTime is now just xcffib.CurrentTime.

Enhancements

  • When sending requests with nested structs you no longer have to pack the contents yourself. For example, when calling xproto.FillPoly, you used to have to convert the POINTs you were passing in to some sort of buffer which had them struct.pack'd. Now, you can just pass an iterable (or xcffib.List) of POINTs and it will be automatically packed for you.

  • Most of the lower level XCB connection primitives that were previously not exposed are now available via xcffib.{ffi,C}, assuming you want to go out of band of the binding.

  • Checked vs. Unchecked requests are still supported (via Checked and Unchecked function calls). However, there is also an additional optional parameter is_checked to each request function, to allow you to set the checked status that way. Additionally, requests that are (un)checked by default, e.g. QueryTree (CreateWindow), have a QueryTreeChecked (CreateWindowUnchecked) version which just has the same default behavior.

  • The FooError BadFoo duality is gone; it was difficult to understand what to actually catch if you wanted to handle an error. Instead, FooError and BadFoo are aliases, and both implement the X error object description and python Exception (via inheriting from XcffibException).

  • You can now create synthetic events. This makes it much easier to work with ClientMessageEvents. For example:

    e = xcffib.xproto.ClientMessageEvent.synthetic(format=..., window=..., ...)
    conn.core.SendEvent(..., e.pack())

Why haskell?

Why is the binding generator written in haskell? Because haskell is awesome.

TODO

  • xprint and xkb support. These will require some non-trivial work in xcb-types, since it won't parse them correctly.