Doing the suggested change for name-prefix breaks the compile, so we
can't do that one yet.

So we can use this from tests directly without dealing with linker things
for tests and what not.

There is a bunch I'd like to do to make this better now, but we have to
get some basics in place and round off some corners which never really
worked with GITypelib access.

My goal here is to make this a bit easier to use as a basic scripting
layer for Builder to use in it's shortcut bindings.

Without this, you can't call a function called "require" on a GObject as
the require keyword would have higher priority. To use the reserved keyword
you can now instead use the following, similar to C♯.

obj.@require(args)

This will give us a bit more flexibility going forward so that we can have
different syntax between the two from a scriptability standpoint.

Just makes it easier to debug later on.

Mostly used for debugging so that you can see if/why you should have missed
something in that list.

We don't care about the whole L"" stuff in the literal as that should only
be dealt with when scanning so the parser can be naive.

This was lost when we renamed the old STMT_LIST to ARGS.

These are not handled automatically for us, and most TmplExpr node creation
take `const char *` strings and therefore are not freeing these for us.

An oversight I wish we could fix for parsing performance, but what can you
do at this point.

We always had this in the syntax, but we never actually finished the
implementation of defining functions so that the user can call them.

I'd still like to change the syntax somewhat, but this at least gets
things working so we can define those functions and use them from the
test script.

This fixes an incorrect unref we couldn't hit before because we didn't
have user functions working. But in particular USER_FN_CALL does not use
the SIMPLE expr node as it has it's own.

The parameter is transfer full, so we don't need to ref the array.

We don't really need to unset the value since it's a primitive type, but
best to just do the right thing anyway.

This wasn't supposed to be changed previously.

So that we can use this in more ways without duplication.

This is mostly useful for us for tests so that we can fill in the script
files with things we expect to be without having to check return values
from the C sources directly.

And test it in test1.script

This was supposed to be on the parameter, which is already there.

And add cos/sin/tan builtins

Builtins can only take a single parameter currently, so pow() cannot
currently be hoisted asa builtin. Test something dumb as a fallback.

Like python you can do # some comment

This can be used to end things on a single line where you might not
otherwise be able to.

This can be helpful in some upcoming places.

These are just handy in case people are blindingly writing as they read
from other bits of GLib stuff.

These should be reversed to preserve the expected order.

It turned out to be too much effort to treat it as EOL.

We might want to drop names from func as well but we'll see.

This is just handy to write to stderr intead of stdout.

This way we can reuse it for anonymous functions.

This just makes calling code a bit cleaner and avoids copying a GValue
to get the boxed pointer.

Just to make quick testing as we fix things easier.

This allows creating a new function which does not have a name even though
we can't call them yet.

This is just general progress towards making an anonymous function that
has been assigned to a symbol in scope work.

Not quite there yet, but closer.

We can just use def/end for cases other than that.

This should use item.@version instead now to get the proper attribute.

 * Use -Wl,-z,relro and -Wl,-z,now when possible in release builds
 * Use -Wl,-Bsymbolic and -fno-plt when possible
 * Disable cast checks and assertions in release builds

This ensures we've built the enums headers before it can begin compiling
to avoid the race. Also we can avoid the weird include_directories()
issue from tests/meson.build.

We don't need to allocate these, so long as reference counting is correct
they never get freed and we can just re-use the same instance as a constant
sort of cheap hack.

We will probably run into additional things to fix with this, particularly
around comparisons/etc but this is a start.

That way we can check pointer equality (although we need additional
checks for object/variant/etc comparisons).

This allows you to do handy things like check types with inheritance such
as the following:

 typeof(widget).is_a(typeof(Gtk.Native))

This is a little bit of a cleanup around the build. Makes the build a
little bit more modern.

Its nice and all to force extraneous (), but its also really annoying.

We need to fix this still.

Clean up Meson a bit

See merge request GNOME/template-glib!6

For various sizes/floating point/double.

Fixes #3

meson: export package in gir

See merge request GNOME/template-glib!9

Error out build early when is a subproject without -Dpackage_subdir

See merge request GNOME/template-glib!8

As mentioned in !8

This fixes things to work where we don't have non-standard sscanf features
such as on macOS.

Obviously you don't want to evaluate the whole tree.