A lookup tag with no type attribute would crash the decompiler, even if
that was valid. This wasn't caught by the tests since blueprint never
generates such XML.
Also fixed a bug in the tests that caused decompiler-only tests not to
run.
Add more tags to the list of things the decompiler can handle. This
required some changes to track the containing object class in the
DecompileCtx, since some objects use the same tag names.
The improved support means we can test the decompiler on most of the
test suite. Any new test samples will by default be tested to ensure the
decompiler produces the original blueprint file.
Also, updated the decompiler to always use double quotes.
GtkBuilder XML uses enum nicknames, full names, or integer values, but
we accept GIR names, so passing those through doesn't work if the name
has an underscore (which traditionally turns into a dash in the
nickname). Avoid the problem by always writing the integer value of the
enum member.
For normal compilation, use .typelib files rather than .gir XML files.
This is much faster.
Rather than using libgirepository, which would try to actually load the
libraries, we use a custom parser.
The language server will still read XML because it needs to access
documentation, which is not in the typelib, but that's generally fine
because it's a long lived process and only has to do that once.
I want all language elements to be defined in the same folder, rather
than having the "core" language split by parsing/AST/completions and the rest
split by language element.
`blueprint-compiler port` interactively ports a project to blueprint.
It will create the subproject wrap file, add it to .gitignore, decompile
your GtkBuilder XML files, emit code to copy and paste into your
meson.build file, and update POTFILES.in.
It can't quite handle all of the features the forward compiler can, so
it will skip those files.
