Documentation says about g_vfunc_get_offset():
"Obtain the offset of the function pointer in the class struct.
The value 0xFFFF indicates that the struct offset is unknown."
But g-ir-compiler did set the value to 0 when the offset is unknown.
This patch fixes it.
https://bugzilla.gnome.org/show_bug.cgi?id=628270
One of the first big changes in this rewrite is changing the Type
object to have separate target_fundamental and target_giname properties,
rather than just being strings. Previously in the scanner, it was
awful because we used heuristics around strings.
The ast.py is refactored so that not everything is a Node - that
was a rather useless abstraction. Now, only things which can have
a GIName are Node. E.g. Type and Field are no longer Node.
More things were merged from glibast.py into ast.py, since it isn't
a very useful split.
transformer.py gains more intelligence and will e.g. turn GLib.List
into a List() object earlier. The namespace processing is a lot
cleaner now; since we parse the included .girs, we know the C
prefix for each namespace, and have functions to parse both
C type names (GtkFooBar) and symbols gtk_foo_bar into their
symbols cleanly. Type resolution is much, much saner because
we know Type(target_giname=Gtk.Foo) maps to the namespace Gtk.
glibtransformer.py now just handles the XML processing from the dump,
and a few miscellaneous things.
The major heavy lifting now lives in primarytransformer.py, which
is a combination of most of annotationparser.py and half of
glibtransformer.py.
annotationparser.py now literally just parses annotations; it's
no longer in the business of e.g. guessing transfer too.
finaltransformer.py is a new file which does post-analysis for
"introspectability" mainly.
girparser.c is fixed for some introspectable=0 processing.
Rather than have the scanner/parser handle both e.g. "glong" and
"long", simply use the GLib types everywhere.
This commit adds TYPE_LONG_LONG and TYPE_LONG_DOUBLE to the
scanner types; however, rather than add them to the typelib,
they're just marked as not-introspectable.
The reason for the warning was that g_irepository_get_version() expects
the typelib to be already loaded, but enumerate_version() can be called
on typelibs that are not.
Logically speaking, the already loaded version of a namespace is part of
the currently available versions, and can be forgotten if we only
consider the versions available in GI_TYPELIB_PATH, as it could have
been loaded using g_irepository_require_private().
As a side effect, it meant that bindings relying on enumerate_version()
(like pygobject) were not able to require private versions through their
classical requirement scheme.
This patch fixes it by adding the loaded version to the unsorted list of
available versions returned by g_irepository_enumerate_versions()
This patch also uses g_list_prepend() instead of g_list_append() in that
function.
https://bugzilla.gnome.org/show_bug.cgi?id=625983
We never actually include multiple modules in the compiler,
so just nuke that. Also rather than passing around GIrModule
consistently pass around a GIrTypelibBuild structure which
has various things.
This lets us maintain a stack there which we can walk for
better error messages.
Also, fix up the node lookup in giroffsets.c; previously
it didn't really handle includes correctly. We really need to
switch to always using Foo.Bar (i.e. GIName) names internally...
We didn't show the right error message if we failed to find
the symbol; fix this by removing error printing from the
middle of the dumper, and add it correctly to the toplevel
dump entry point.
Entries in the GI_TYPELIB_PATH environment variable are added to the
global search path in reverse order - instead, add entries in the
same order in which they are specified.
Take a GError * for typelib loading code, validate the header. This
fixes bizarre errors from gjs where g_irepository_require would happily
load old typelibs.
Previously we had both e.g. GI_TYPE_TAG_LONG and GI_TYPE_TAG_INT64,
but in fact the typelib is already machine-specific, so it makes sense
to just encode this as a fixed type. The .gir remains abstract.
We also remove size_t from the typelib; one would never want to treat
it differently than an integer.
time_t is removed as well; while bindings like gjs had special handling
to turn it into e.g. a JS Date object, I don't think we should encourage
people to use these POSIX types in their API. Use GTimeVal or the like
instead.
Because the typelib is now really machine-specific, we need to remove
the -expected.tgirs from git. (We could potentially add a check
which wasn't just a literal diff later)
https://bugzilla.gnome.org/show_bug.cgi?id=623774
This patch adds support for instantiable fundamental object types,
which are not GObject based. This is mostly interesting for being
able to support GstMiniObject's which are extensivly used in GStreamer.
Includes a big test case to the Everything module (inspired by
GstMiniObject) which should be used by language bindings who wishes to
test this functionallity.
This patch increases the size of the typelib and breaks compatibility
with older typelibs.
https://bugzilla.gnome.org/show_bug.cgi?id=568913
Moving to <doc> allows us to better preserve whitespace. XML has no
facility for whitespace-preserving attributes.
Second, for arrays and lists, both types with unknown element_type can
occur in the current scanner; it's least wrong if we write out an
<any> type.
Any annotation where the key has a dot in the name will go into the
attribute list. For example
* @arg: (foo.bar baz): some arg
the parameter @arg will get the attribute with key foo.bar and value
baz. This also works for.
* Returns: (foo.bar2 baz2): the return value
Also add tests for this new feature.
See https://bugzilla.gnome.org/show_bug.cgi?id=571548
Signed-off-by: David Zeuthen <davidz@redhat.com>
Rectify an assumption that nodes are ordered according to offset
- since this assumption was not true, attributes ended up being not
ordered either and the bsearch() when looking up attributes failed
mysteriously. Instead of making such assumptions, simply sort the
list of nodes we want to extract attributes from.
The total attribute size computation was wrong as we didn't properly
descend into subnodes. This resulted in memory access violations
when writing the typelib (because not enough data was allocated).
Instead of having a separate function for this, just include the
attribute size in the existing function.
See https://bugzilla.gnome.org/show_bug.cgi?id=571548
Signed-off-by: David Zeuthen <davidz@redhat.com>
This work allows us to move closer to replacing gtk-doc, among other
things. We add a generic attribute "introspectable", and inside the
typelib compiler if we see "introspectable=no", we don't put it in the
typelib. This replaces the hackish pre-filter for varargs with a much
more generic mechanism.
The varargs is now handled in the scanner, and we emit
introspectable=no for them.
Add generic metadata to Node with references to file/line/column,
which currently comes from symbols.
Add scanner options --warn-all and --warn-error.
https://bugzilla.gnome.org/show_bug.cgi?id=621570
