Problems embedding rxvt-unicode

[Tuukka Kataja]

>> At least on my system (Ubuntu 10.04), xterm has a switch "-into" that
>> works like "-embed" on rxvt-unicode.
>
> I am intrigued, but I just looked at xterm sources, and see no traces of
> xembed support. At all. As such, the -into option has nothing whatsoever
> to do with -embed. (Which is consistent with the documentation of both
> -into and -embed).
>
> Since you claim otherwise, how do you come to the conclusion that it works
> like -embed?
>

I came to the conclusion, because "tabbed" acted visually and functionally
as though xterm had support for XEmbed. I can start multiple xterms with
"-into" and they appear to "tabbed" exactly as should and behave normally.

I didn't have the xterm sources at hand, so I wasn't able to check the
implementation. From the man page I assumed (wrongly apparently), that
"This is used to embed xterm within other applications" ment using XEmbed.
Sorry.

>> I'm under the impression that scrotwm (being a window manager) doesn't
>> have to support XEmbed.
>
> Indeed, but bear with me - we are trying to disentangle your exact setup
> form your mails, and you send a lot of conflicting statements.
>

Please, see the end of the message for an attempt to clarify.

>> XEmbed is a protocol between only the embedding application
>> (suckless.org's tabbed) and the embedded application (rxvt-unicode), is
>> it not?
>
> Right!
>
>> Because of the XEmbed specification, which implies this in a couple of
>> places:
>
> I can't find it anywhere in the xembed specificaiton, and none of the
> parts
> you quoted it mention it either.
>
>> This to me says: The embedder takes care of mapping/unmapping embedded
>> windows and the embedded clients can ask for this to happen through
>> XEMBED_MAPPED.
>
> and rxvt-unicode fully handles XEMBED_MAPPED (which is not MapNotify).
>
>> in the embedding application. (Map requests are also redirected, but
>> XEmbed actually handles map requests separately... see the description
>> of
>> the XEMBED_MAPPED flag.)"
>
> And rxvt-unicode doesn't rely on map requests, but instead uses
> XEMBED_MAPPED.
>
>> Note the sentence in parenthesis. All of this indicates in my mind that
>> expecting MapNotify events while being embedded isn't a good idea. After
>> all, the XEmbed spec instructs the embedder to select the
>
> Your mind works very strangely, as MapNotify isn't mentioned at
> all. Moreso, map notifications in general are not handled by xembed at
> all, and not mentioned at all. If you disagree, where in the xembed spec
> does it handle map notifications?
>
>> SubstructureRedirectMask. Therefore (if I understand X correctly)
>> MapNotify events won't come to the embedded window unless the embedder
>> specifically sends it them.
>
> You don't understand X correctly - MapNotify is an event sent when the
> window is mapped (and the window selects for it). It doesn't matter *how*
> the window is being mapped.
>
>> And the XEmbed spec doesn't require this, but
>
> The Xembed spec doesn't require support for keyboard events either. The
> xembed spec is about embedding, not about basic X functions - all the X
> functionality is still there EXCEPT when it disagrees with the xembed
> specification.
>
> Or to put it otherwise: where in the xembed spec does it mention
> XCreateWindow? If it doesn't mention it, how does one create windows?
>
> Obviously, xembed does not invalidate the remaining parts of X.
>
>> gives it's own method. Similar things are said in the spec for
>> XEMBED_WINDOW_ACTIVATE and XEMBED_EMBEDDED_NOTIFY.
>
> Note the absence of XEMBED_WINDOW_MAPPED or something like it.
>

I stand corrected.

>> Also, the fact that rxvt-unicode is getting VisibilityNotify events
>> (with
>> VisibilityUnobscured or VisibilityPartiallyObscured) should tell
>> rxvt-unicode that it's windows are in fact mapped and visible, and
>> therefore screen updates are required.
>
> I think it is much preferable to get embedders (or WMs) right instead of
> implementing hack workarounds for each and every application out there.
>

I agree. Using VisibilityNotify events was based on my misunderstandings
of X.

>> All that said, I still can't figure out why the MapNotify events come
>> through in GNOME but not in scrotwm. I checked, and scrotwm isn't seeing
>> _any_ events from the embedded rxvt-unicode window, as it shouldn't
>
> It would be possible that "tabbed" does something different when it is
> running under scrotwm, than when it is running under gnome.
>

Tabbed is a very simplistic piece software, and doesn't care about the WM.
This I have confirmed by tracking the execution of tabbed. Also, the same
things happen using the example embedders.

>> But I still think the absence of MapNotify events shouldn't matter to
>> rxvt-unicode when it's embedded.
>
> Replace MapNotify with KeyEvent - is it still true for you? Why should
> applications cope with partial breakage? Wouldn't it be better to do
> it the right way, so not every embedding app needs to deal with broken
> embedders?
>

Again, I stand corrected..

>> > If rxvt-unicode calls mapwindow, and the wm makes it visible, then
>> > rxvt-unicode is entitled to know about that.
>>
>> When rxvt-unicode is being embedded, the WM isn't making it visible,
>> because non-toplevel windows aren't the WM's problem.
>
> It wasn't clear to me how your setup works (it still isn't, but it seems
> clearer to me). Indeed, the wm isn't responsible here.
>
> And the more I think about it, I think that your whole description of whta
> happens is likely wrong - I would really expect the mapnotify to be sent.
>

Let's see if I can clear things up a bit. I'm running Ubuntu 10.04. I'm
able to log in either starting GNOME (Metacity I gather as the WM) or
without GNOME using scrotwm as the WM. I'm trying to use "tabbed" to have
a tabbed interface to rxvt-unicode. Yes, I know about the perl-extension,
but this started as a test to see how to embed rxvt-unicode (planning to
do it in a project of mine).

Now, when I am logged in to a GNOME session everything works fine.

When I log to a scrotwm session, rxvt-unicode works fine when not
embedded. But when I attempt to embed rxvt-unicode with "-embed" no text
is output, but the terminal works (which I can see from the scrollbar and
blindly entered commands). This happens regardless of whether I'm
embedding into "tabbed" or to the examples in rxvt-unicode's
doc/-directory.

I started investigating what was happening, and determined (using a
primitive technique of fprintf's to stderr) that a MapNotify event is
never being received by rxvt-unicode and therefore the screen isn't being
updated. In the gnome session, the MapNotify _is_ received. If there is
something that I'm missing which makes this conclusions wrong, do tell me.
I just want to get the embedding to work, regardless of what piece of
software is acting up.

 - Tuukka