Skip to content
GitLab
  1. 05 Aug, 2022 2 commits
  3. 24 Jul, 2022 3 commits
  5. 23 Jul, 2022 1 commit
  7. 03 Jul, 2022 1 commit
  9. 17 Jun, 2022 4 commits
  11. 16 Jun, 2022 2 commits
  13. 27 May, 2022 1 commit
  15. 03 May, 2022 1 commit
  17. 21 Dec, 2021 1 commit
  19. 17 Nov, 2021 1 commit
  21. 04 Oct, 2021 1 commit
  23. 30 Sep, 2021 2 commits
    • Uli Schlachter's avatar
      Use xml.etree.cElementTree where appropriate · 65169c1a
      Uli Schlachter authored 
      

This makes the code added in the previous commit consistent with the
rest of the code base. From Björn's suggestion:

  This should import xml.etree.{,c}ElementTree conditionally on the
  version of the Python interpreter used to import this module.

  xml.etree.ElementTree is preferred for Python >= 3.3, as it uses the
  fastest possible implementation automatically. Earlier versions of
  Python need xml.etree.cElementTree as they may not have an
  implementation that can be used in a platform generic way.

Suggested-by: Björn Esser's avatarBjörn Esser <besser82@fedoraproject.org>
      65169c1a
    • Uli Schlachter's avatar
      Add missing fields to errors · 6d72110e
      Uli Schlachter authored 
      All X11 errors have the same fields. There are no differences.

In a perfect world, the XML could thus just say "define an error" and
xcbgen would do all the rest. However, the world is imperfect and we
already have a mixture of fields defined in the XML. Some of the XML
even defines trailing padding, while most does not.

This commit makes xcbgen add all fields to X11 errors, but those that
are already defined in the XML are skipped and left as-is. Due to the
structure of the code, this requires pretending that a different XML was
read, i.e. the code now modifies the in-memory structure of ElementTree
to add the missing fields to the in-memory representation of the XML.
This is the simplest way that I found to append elements. The existing
mechanisms can only prepend fields.

The approach taken by this commit was suggested by Peter Harris. Thanks
a lot for this idea, it's a lot simpler than my previous approach.

Fixes: xorg/proto/xcbproto#16


Signed-off-by: Uli Schlachter's avatarUli Schlachter <psychon@znc.in>
      6d72110e
  25. 22 Sep, 2021 1 commit
    • Povilas Kanapickas's avatar
      xinput: Add XI v2.4 · 151ee698
      Povilas Kanapickas authored 
      XInput 2.4 introduces touchpad gesture support which includes 6 new
events: Gesture{Pinch,Swipe}{Begin,Update,End}.
      151ee698
  27. 02 Sep, 2021 2 commits
    • Povilas Kanapickas's avatar
      xinput: Add length specification for DeviceClass struct · f0db8b7d
      Povilas Kanapickas authored
      f0db8b7d
    • Povilas Kanapickas's avatar
      Add element to specify expression that defines length of a struct · c36dde3f
      Povilas Kanapickas authored 
      Currently the layout of a struct is used to compute its size. This works
fine in case of structs of fixed size. However this introduces
forwards-compatibility problems in cases when the struct has multiple
variants and the exact variant is specified by the value of some field
(e.g. in the case of <switch> elements). Future revisions of protocols
may introduce new layout variants, in which case the old code does not
know the size of the struct variant and can't parse the incoming byte
stream.

Instead of relying on knowledge about the layout of data structures we
should instead use the length field for length information. This way
when old client libxcb communicates with newer server it can at least
ignore unknown struct variants.
      c36dde3f
  29. 22 Jul, 2021 1 commit
    • Olivier Fourdan's avatar
      xfixes: Add ClientDisconnectMode · 78d0652a
      Olivier Fourdan authored 
      

The Xserver itself is capable of terminating itself once all X11 clients
are gone, yet in a typical full session, there are a number of X11
clients running continuously (e.g. the Xsettings daemon, IBus, etc.).

Those always-running clients will prevent the Xserver from terminating,
because the actual number of X11 clients will never drop to 0.

To solve this issue directly at the Xserver level, this add new entries
to the XFixes extension to let the X11 clients themselves specify the
disconnect mode they expect.

Typically, those X11 daemon clients would specify the disconnect mode
XFixesClientDisconnectFlagTerminate to let the Xserver know that they
should not be accounted for when checking the remaining clients prior
to terminate.

Signed-off-by: Olivier Fourdan's avatarOlivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Simon Ser's avatarSimon Ser <contact@emersion.fr>
      78d0652a
  31. 02 Jun, 2021 1 commit
  33. 08 Oct, 2020 1 commit
  35. 03 Oct, 2020 1 commit
  37. 04 Jun, 2020 1 commit
  39. 01 Jun, 2020 2 commits
  41. 21 Mar, 2020 1 commit
    • Uli Schlachter's avatar
      Parse a field's "enum=" correctly · 2b3559c1
      Uli Schlachter authored 
      

In xv.xml, there is something like this:

  <struct name="ImageFormatInfo">
    [...]
    <field type="CARD8" name="byte_order" enum="ImageOrder" />
    <pad bytes="2" />
    <list type="CARD8" name="guid">
      <value>16</value>
    </list>
    [...]
  </struct>

When parsing this, the Field instance for "guid" ended up with .enum ==
"ImageOrder". This is because the loop that parses complex type did not
unset a variable across iterations, meaning that the last "enum"
property "stuck" and was also used for all following fields.

Fix this by simply moving the initialisation of the "enum" variable
inside of the loop.

Signed-off-by: Uli Schlachter's avatarUli Schlachter <psychon@znc.in>
      2b3559c1
  43. 22 Feb, 2020 2 commits
  45. 28 Dec, 2019 2 commits
    • Uli Schlachter's avatar
      Allow access to the original type in the XML · e79f6b01
      Uli Schlachter authored 
      

xcbgen 'helpfully' transforms things to C types already so that libxcb
does not have to do so. Thus, even though the XML says that a field has
type CARD8, xcbgen will claim uint8_t. This might be a bit weird, but is
so far totally fine.

However, the type mapping that xcbgen uses is not injective. All of
CARD8, BYTE and BOOL get turned into uint8_t and the original type is
lost.

This is totally fine for libxcb, but programming languages other than C
do have built in boolean types. My personal problem is with Rust, where
providing a boolean for an integer argument causes a compiler error.
This results in (relatively) ugly "0 / 1" instead of "false / true".

This commit adds a new xml_type member to SimpleType. This type contains
the original string that appeared in the XML file.

Since libxcb creates instances of SimpleType itself and to avoid
breaking the API, the new argument to SimpleType.__init__ defaults to
None.

Signed-off-by: Uli Schlachter's avatarUli Schlachter <psychon@znc.in>
      e79f6b01
    • Uli Schlachter's avatar
      Fix size computation of imported lists · 3cc42f6d
      Uli Schlachter authored 
      

XFixes contains a CreateRegion request:

   <request name="CreateRegion" opcode="5">
     <field type="REGION"    name="region" />
     <list  type="RECTANGLE" name="rectangles" />
   </request>

This request contains a list of type RECTANGLE. This struct comes from
xproto and is thus not contained in xfixes itself.

Normal "Struct"s have their resolve() method called early, because they
appear in the module itself. However, in the CreateRegion case, this
struct is imported and thus does not get resolved. Instead, ListType's
resolve() method calls self.member.resolve(module). Thus, only at this
point is the struct resolved.

Why is this important? Struct.resolve() is the same as
ComplexType.resolve() and this function does self.calc_size() at the
end. Thus, only after the struct was resolved is its size known. Before
that, the size is just set to 0 (this happens in ComplexType.__init__).

However, ListType.__init__ already computes its size member based on its
member. At this point, this is still 0 so the list ends up believing its
size to be zero.

Fix this by recomputing self.size in ListType.resolve().

Signed-off-by: Uli Schlachter's avatarUli Schlachter <psychon@znc.in>
      3cc42f6d
  47. 02 Nov, 2019 1 commit
  49. 27 Apr, 2019 1 commit
    • Daniel Stone's avatar
      pkg-config: Add sysroot prefix · bd7a481a
      Daniel Stone authored and Jon Turney's avatar Jon Turney committed 
      

xcb-proto's pkg-config file exports static file paths for where to find
its Python files and the XML definitions.

When used to discover cflags, library paths, etc, pkg-config will
prepend ${PKG_CONFIG_SYSROOT_DIR}, if any, to those paths. This makes
the use of a sysroot prefix transparent to users. However, since it
doesn't know about paths in custom variables, it cannot automatically
prefix this.

Adding ${pc_sysrootdir} to these absolute paths makes the behaviour
match, and XCB builds work out of the box in a sysroot.

Signed-off-by: Daniel Stone's avatarDaniel Stone <daniels@collabora.com>
Reviewed-by: Emma Anholt's avatarEric Anholt <eric@anholt.net>
      bd7a481a
  51. 17 Feb, 2019 1 commit
  53. 14 Mar, 2018 1 commit
  55. 28 Feb, 2018 1 commit
    • Daniel Stone's avatar
      Release xcb-proto 1.13 · 94228cde
      Daniel Stone authored 
      

This release adds support for variable-sized lists of FDs (e.g. sized
by a field in the request/event), as well as several other cleanups and
fixes in the core code.

DRI3 has been bumped to v1.2, using the new support to support
multi-planar buffers and DRM modifiers.

Present has also been bumped to v1.2, adding a new token for the server
to indicate to the client that it should reallocate its buffers.

RandR has been updated to 1.6, adding support for DRM leases to allow
clients to directly control outputs.

Signed-off-by: Daniel Stone's avatarDaniel Stone <daniels@collabora.com>
      94228cde