- handle the case where objects are just '/'. Currently they fail in invoke
with "failed to find object ''"
nul_dbus_service_start() should likely take a data parameter to
distinguish multiple instances of the same service. Method callbacks
will need to know this information. How does this work given the
"callback, data" religion?
Maybe callbacks for methods should be hooked after the service has
been started? That way they can get data pointers that may not be
available when the service starts.
Alternatively, maybe it should be possible to check whether a service
can be started without actually starting it. Ie.,
if (!service_available())
{
App *...
start ();
}
else
{
invoke();
}
This is racy though.
Another possibility would be to say "start_service()" and if that is
successful, then add the objects afterwards. For now, this is what we
are doing.
Callbacks should religiously be on the "callback, data," form to make
them bindable in languages with closures. (Broken at the moment for
objects).
Is this actually sufficient? Destroy notifiers would also be
necessary. Maybe languages just need to suck it up and generate
trampolines. We still need to know about "data" arguments so that we
can ignore them both in the trampoline and in the bound API.
Would it be possible to also use tramplines in C? Ie., have a macro
NUL_CALLBACK (function, data)
that would return a single function to call? Event code would look like this then
nuuk_connect (button, "clicked", NUL_CALLBACK (function, button, x, y, z));
and
nuuk_connect (button, "clicked", NUL_CALLBACK (function, NUL_INT (x), NUL_INT (y)));
But there would need to be a way to specify where extra information
passed to the function would go. Maybe that would just always go first. Ie.,
nuuk_connect (button, "clicked", NUL_CALLBACK (on_clicked));
...;
void
on_clicked (void)
{
printf ("clicked\n");
}
Would require a dependency on macro varargs and on libffi closures.
libffi closures are not a problem. They are supported for all
relevant, and several irrelevant, architectures. Macro varargs - the
main issue might be msvc. Actually, it doesn't have to be a macro, it
could be a vararg function. It would require some way to specify the
end of the argument list though. Or just give the number of arguments:
NUL_CALLBACK (function, 2, NUL_INT (x), NUL_INT (y));
If the NUL_INT constructors return a pointer, it would be easy to
check at runtime that the number of args is not an insanely big number
(say > 100 or < 0).
If all callbacks are defined as trampolines, then it is possible to
have them be a struct with an optional destroy notifier embedded in
it. Ie., the NUL_CALLBACK would do something like this:
nul_callback_info_t *info = malloc (sizeof (...));
info->f = make_closure (f, ...);
info->destroy = free_callback_info
where free_callback_info would free the closure and the callback info
plus any other house keeping.
Actually, we probably don't even need libffi closures. Instead
NUL_CLOSURE() creates a nul_closure_t that contains the function, the
types and the arguments. The caller of the closure is then responsible
for supplying extra arguments and for destroying it.
There will need to be functions
nul_create_closure (function, n_args, ...);
/* These functions would take care of storing/overwriting/destroying */
nul_closure_own (closure **closure, closure *closure);
nul_closure_disown (closure **closure);
/* tie the lifetime of the closure to the object */
nul_object_own_closure (object, closure **closure);
/* Takes any additional argument. These are added to the front
* of the parameter list followed by all the other
*/
nul_closure_call (closure, ...);
Could the "own" functions take additional types and then return a
function that could be called directly?
Type systems:
- sfile array, record, union, double, names,
- invoke basic types, no names
- dbus array, record, double, names
- introspection
Would be nice to unify some of these.
For sfile, the names for records, unions and lists serve two purposes:
(1) it is the name of the corresponding variable, and (2) it is the
name of the type for the purposes of pointer types.
dbus types are used directly with invoke, but for now arrays and
records are not supported. These will need to be converted to pointers
for invoke.
Dbus:
Note that proxy_invoke() will have to contain a description of the
types, because the alternative is to believe the XML from the service,
which would mean a malicious service could make us send random stuff
from the stack.
Write tool that reads the introspection XML and generates service
code.
Can we use the same type system for dbus and sfile? The sfile type
system is basically name/type pairs, which is approximately what dbus
needs as well.
For dbus we will need structs of basic types, arrays of basic types,
and probably arrays of structs of basic types. Arrays will be
represented in C as "int n; const type *array". I suppose it would be
possible to generate the fully nested types, but it's pretty crackful
that that's even supported in dbus.
For the introspection types, we don't need the names, so those types
will probably have to be specified in some other ways. One question is
if it is only the specification or also the actual types that are
different.
Or possibly make sfile use anonymous types.
Why do we not need names for the introspection types? The parameter
names would be quite useful actually.
Things from glib:
- Consistent API between things like GCheckSum and GBase64.
GCheckSum looks better to me offhand. Note, these should also be
able to take nul_byte_queue_t's (or nul_strings())
nul_checksum_t
nul_base64_t
nul_iconv_t (iconv wrapper)
Consider making them objects inheriting some sort of stream
interface. This could also be done separately, and in a way that
would allow errors to propagate. Ie., build a stream out of
smaller streams.
nul_compound_stream (
nul_convert_stream ("LATIN-1", "UTF-8"),
nul_checksum_new (G_CHECK_SUM_SHA1),
NULL);
nul_object_connect (stream, "event", on_event, NULL);
nul_stream_feed (stream, "asdfasdf");
nul_stream_end (stream);
The event system here could really benefit from extensible unions
so that you can have inherited streams with more complex events.
- Note GData is used in GObject, and by GScanner, which is also used
for various things. GHook is also used in GObject (signals), plus,
weirdly, an enum from ghook.h is used in gmain.c. Also, gpattern
is used in gtestutils.c
- These things should be improved:
- ghash.h open addressing, no primes
- gmain.h simpler API, better threads
Better child watch
- gmarkup.h cut and paste expat
- grand.h Delete the context version
- gsequence.h Should gain aggregators
- gstring.h Should probably be replaced with byte array
(which in turn must get equivalent API to
gstring)
Is GStringChunk useful? probably not
- gthreadpool.h maybe (replace with executor?)
- gasyncqueue.h probably
- G*Array Direct pointers, byte queue
- New things:
- sparse table maybe
- byte queue
- sfile
- block sequence maybe, or could become default for gsequence.
- FreeList maybe
- watch API
- GObject should go away, replaced by new, simpler object system
(See below)
- Note: GIO depends on GObject, so the right order is:
- Add new object system
- Port GIO to it
- delete GObject
- GIO in separate library?
- GThread integrated into main library, and simplified
- Careful API review of what remains
- These things are crack:
- gbacktrace.h though G_BREAKPOINT is useful
- gdataset.h
- ghook.h
- giochannel.h
- gnode.h
- gpattern.h
- gprimes.h
- gquark.h possibly
- gslice.h
- gscanner.h
- galloca.h
- gmem.h the malloc wrappers are useful, chunks most go
- gcache.h
- gcompletion.h
- gtree.h
- grel.h
-=-=- Notes on main loop
New, even simpler design
Basically, having one main thread is good enough. JobQueues/MainContexts
can be implemented on top of that. Executor could be called ThreadPool.
JobQueues should have a reference to an executor.
job_queue_set_read (job_queue_t *job, int fd, read_func, data);
job_queue_set_write (job_queue_t *job, ...);
job_queue_add_idle/timeout().
Whenever something is queued up for a job queue, the jobqueue's
run-next time is computed, and if it needs to run at some point, it is
added to the executor along with the timeout. The executor will
maintain a sorted list of things that need to run (sorted by when they
need to run). When there is nothing to do it waits for that timeout to
expire (or for something to happen), otherwise it schedules things in
order.
When something is added that should run "as soon as possible", it is
added with the current time. This ensures that timeouts etc. will
eventually run (because their deadline will eventually be in the
past).
So when an fd fires for a job queue, this will happen:
1. callback is added to the job queue's queue. (Insert sorted with
current time as deadline).
2. Current run-time is computed, and the job queue is added to the
executor.
The job queue queue probably needs its own mutex, separate from the
job queue's.
New, simpler design:
There is an 'Engine' which is both Executor and Mainloop. It uses an
Epoll object to handle the polling. This epoll object does not do any
locking by itself.
Contexts can be created; they are like jobqueues, except they also
allow you to add fd's and idles and timeouts. They can do that since
they have access to the engine. These are completely unlocked; if you
want to access a context from more than one thread, you must lock them
yourself. It is guaranteed that only one callback will be active for a
context at the same time. From within that callback it is then safe to
remove other callbacks etc. In general such a context can act pretty
much like one thread.
A web server will consist of a listening context, which will create
other contexts as necessary and add the fd's to them with the
appropriate callback. Or maybe just call back with the new fd. I guess
there should be a generic listener object that will create its own
context and call back. Then an http module can just create a listener
and do whatever it wants in response.
One possible problem. The thread that creates a context and adds the
fd not necessarily the one that will execute the first callback. Can
these interfere? Maybe the solution is just that the creating thread
should not do anything with the context after it adds the fd. A
better idea may be to just have a context_dispatch() function that you
must call after creating the context. Until this is called, nothing
the context is not active.
Beginning of this new design is implemented and can be built with
build2.sh.
Old design (referenced above)
Highlevel design for a web server:
- Executor
takes callbacks and executes them, possibly in parallell
or out of order. The callbacks are likely to be called in
a different thread than the one that queued them.
Should support both push_back() and push_front(). Push_front()
is needed to do parallel requests with good latency. If
a request can be parallellized, the parallel tasks should
be put at the front of the queue.
- EPoll
Simple class that wraps epoll()/poll(). Note: must support
oneshot polling and rearming.
- JobQueue
Maintain a queue of jobs, uses executor to execute them
in order. Two jobs in the same JobQueue will never run
at the same time.
- Mainloop
Uses executor, epoll.
- Takes filedescriptors and callbacks. calls
back when descriptor is readable/writable/etc.
- The callbacks are put on a job queue, which is passed in
- Each callback is a oneshot - ie., after calling, the
filedescriptor is not polled again until it is rearmed.
The mainloop will have a method to rearm filedescriptors.
This ensures that the mainloop can start a new poll()
whenever it wants without waiting for all the callbacks
to finish.
- Also takes timeouts that can be canceled. Timeouts also
need to be put on a queue.
Sketch of implementation:
polljob()
{
timeout = compute min_timeout ();
if (timeout > 0)
poll_armed_describtors(timeout);
call all timeouts (ie., put them on queues);
call all callbacks (ie., put them on queues);
schedule (polljob);
}
- MainContext
- The thing clients will deal with
- Uses the main loop
- Filedescriptors
maincontext is responsible for rearming the
descriptor after the client callback has been
called.
- Timeouts
- Idle handlers
- Everything associated with a main context happens
serialized - ie., as if only one thread executed it.
- This means a client structure doesn't need to be locked.
- Has a get_executor() method so that clients can parallelize
if they want to.
- Current thinking is that if we have main contexts, who
really needs a main loop? The only thing you could do
with it is to pass it to main contexts.
OTOH that's true of several of the objects here.
see notes at top of maincontext.c
- Worth noting that stuff that has to be passed in to
create a client object must be available to the
listener callback. But see http_server.c for an example
- Err, the epoll in itself is not enough for maincontext,
since it needs to be shared between maincontexts. Ie.,
who would call epoll_wait(). We do really need a main
loop that will call back.
- Listener
Listens on a port. Calls back with a file descriptor when
someone connects.
- Connection
initialized with a MainContext and a filedescriptor.
Creates events when something happens
- Http:
Has a Connection
parses http, emits events such as
"get hostname pagename query"
"post etc etc etc"
- ContentProvider
-=-=- Notes on object/component system:
/*
signals (no class handlers, no return values).
nul_object_emit (object, "birnan", x, y, z);
nul_object_block (object, "birnan");
nul_object_unblock (object, "birnan");
/ nul_object_connect (object, "birnan", on_birnan, DATA);
*/
const nul_type_t *
nul_toggle_button_get_type(void)
{
return nul_define_class (
"nul_toggle_button_t",
nul_extends (nul_fish_get_type()),
nul_implements (nul_scrollable_get_type()),
nul_doc (
nul_doc_since (2, 14, 0),
nul_doc_class_blurb (
"A type of button that retains its state"),
nul_doc_class_long (" \
Yeah, it's going to be quite annoying to type this stuff asd \
But then, it's hopefully not that much documentation that must \
be written this way. \
\
A better scheme may be to have ToggleButton.txt that are run \
through a preprocesser which generates a header file which \
defines a macro TOGGLE_BUTTON_LONG_DOC_STRING, which we then \
just use here. \
\
Or maybe writing documentation as C strings is not \
that bad -- the main issue is likely going to be \
changing existing documentation and manually putting in \
newlines. Maybe we don't actually need newlines. We can just \
require <p> tags. \
\
Yeah, not that bad, actually. \
\
\
Creating two <structname>GtkToggleButton</structname> \
widgets. \
\
void make_toggles (void) { \
GtkWidget *dialog, *toggle1, *toggle2; \
dialog = gtk_dialog_new (); \
toggle1 = gtk_toggle_button_new_with_label (\"Hi, i'm a toggle button.\"); \
\
/* Makes this toggle button invisible */ \
gtk_toggle_button_set_mode (GTK_TOGGLE_BUTTON (toggle1), TRUE); \
\
g_signal_connect (toggle1, \"toggled\", \
G_CALLBACK (output_state), NULL); \
gtk_box_pack_start (GTK_BOX (GTK_DIALOG (dialog)->action_area), \
toggle1, FALSE, FALSE, 2); \
\
toggle2 = gtk_toggle_button_new_with_label (\"Hi, i'm another toggle button.\"); \
gtk_toggle_button_set_mode (GTK_TOGGLE_BUTTON (toggle2), FALSE); \
g_signal_connect (toggle2, \"toggled\", \
G_CALLBACK (output_state), NULL); \
gtk_box_pack_start (GTK_BOX (GTK_DIALOG (dialog)->action_area), \
toggle2, FALSE, FALSE, 2); \
\
gtk_widget_show_all (dialog); \
} \
\
\
\
\
Ok, that's more painful, especially since you don't get any \
syntax highlighting. It also distracts from the rest of the class \
definition. \
\
Also, MSVC does not support longer literals than 2048; an ANSI C \
is only required to support 509 characters. \
\
So a preprocessor that generates a macro that expands to a comma \
separate list of literals. \
"),
NULL),
NULL);
nul_define_constructor (
...);
nul_define_property_rw ("fish", nul_type_int32_limits (0, 0, 200)),
nul_define_property_r ("moose", nul_type_int32()),
nul_define_property_cr ("fwr", nul_type_string()),
nul_define_method (
"set_label",
nul_return_type (NUL_TYPE_VOID),
nul_arg_in ("label", NUL_TYPE_LABEL),
nul_arg_out ("size", NUL_TYPE_INT),
nul_arg_in ("fisH", NUL_TYPE_FISH),
NULL);
nul_define_method (
"birnan",
NUL_TYPE_INT,
nul_type_closure (
NUL_TYPE_VOID,
NUL_TYPE_INT,
NUL_TYPE_DOUBLE,
NULL)),
nul_define_virtual_method (
"..."),
nul_define_signal (
"toggled",
NULL);
NULL);
}
nul_toggle_button_t *
nul_toggle_button_new_with_label (const char *label)
{
return nul_object_new (nul_toggle_button_get_type(), NULL);
}
/* ... */
const nul_introspect_info_t *
nul_introspect (void)
{
/* This should be updated everytime
* new types are added
*/
return nul_introspection_info (
nul_introspect_types (
nul_object_get_type,
nul_toggle_button_get_type,
nul_int_get_type,
nul_rectangle_get_type,
NULL),
nul_introspect_functions (
...));
}
types:
objects
structs (represented as pointers)
unions? (tagged unions, that can extend other unions,
useful for extensible events, represented as
pointers, but contents is a full type. Ie.,
a union of structs will be a pointer to
a (tag + struct)).
If we add them, make sure they can represent a
dbus variant.
basic types
arrays (represented in C as an integer and pointer to
contents; or as a pointer to nul_queue? with
similar semantics as boxed types)
You only get arrays of basic types probably.