-
Carl Worth on Cairo:
- Last night he had all his slides done in OpenOffice, whose Cairo integration isn't quite done. So he wrote a quick SVG-based presentation program for his talk, redid the slides, and got it working with the Glitz backend in the server...
- Cairo is an immediate-mode 2D drawing library with a PostScript-like drawing model, plus Porter/Duff compositing. Supports different backends (displays, printers), and different operating systems.
- 2D graphics: document-centric; lots of text, images and icons; few large geometric figures; needs to match printing.
- Target users are toolkits such as GTK+, also Mono, Mozilla, OpenOffice.org, and anything that needs SVG.
- It's a high-level library, sitting directly below applications in the platform stack. It depends on a rendering system, output device, and a font system underneath.
- Backends: OpenGL (Glitz), XRENDER extension, core X graphics, in-memory images, PostScript. Planned backends: PDF, Quartz, Windows.
- Error handling in C: C has no exceptions; checking return values is tedious, so people don't bother. As a result we get many broken programs.
- Error handling in Cairo: error status is persistent within the Cairo context; core functions all return void. You can call cairo_status() when convenient; the API "shuts down" when an error occurs. All functions are well-defined after any error.
- Most of what SVG needs is done; gradients got added recently. The PostScript backend generates a giant image for printing, for now; it's just a proof of concept. The Glitz backend is 10x-100x faster than software; uses accelerated shaders for gradients, or multi-texturing for everything else.
- Future work: PDF backend; real PostScript backend (but you have to do segmentation for sections of the image that have transparency, as PS doesn't support that); API freeze for a release; optimization (but need applications first).
-
Owen Taylor on what toolkits expect from the graphics sytem:
- Trends in hardware: large screens, moderate resolutions, lots of memory, lots of bandwidth.
- Trends in user interfaces: moving away from strictly overlapping windows — popups/drawers, alpha transparency. Prioritization of information. Computers acting on behalf of the user.
- User experience in 5 years: tight integration of multimedia; animations where appropriate; richly detailed desktop.
- User experience now: no excuse for visible redraws, inconsistent response times. Cut-and-paste and drag-and-drop not Just Working. Jim Gettys mentions bad interaction between the Xfree86 implementation and the Linux scheduler — the kernel thinks that the X server is not an interactive process. Iconsistent response times suck — a popup menu sometimes appears in 1/10 of a second, then in 1 second, then in 0.5 seconds...
- Example of GTK+'s anti-expose mechanism to avoid processing expose events when we could know beforehand that they would occur: this avoids double redraws.
- Having the GDK layer between applications and, say, Xft, was a success. Xft1 was broken, then Xft2 was good, but applications required no changes.
- Successful X extensions do something that apps can't do themselves (e.g. XRENDER and MIT-SHM are successful, versus XPRINT and XIMAGE, which nobody uses because they suck). Successful extensions don't force particular applications on apps; they provide obvious benefits to applications. They don't have to be particularly pretty (e.g. hairiness in XKEYBOARD), as toolkits can wrap them prettily.
- Performance. Most apps do very few things: draw solid rectangles, draw images, draw text. We need benchmarks that accurately simulate app operation — no one draws a million arcs. We need point-to-point benchmarks: rather than asking, "how many rectangles can we draw in a second?", ask "how much time does it take from pressing a button to bringing up a menu on the screen?". Profiling is a black art; we need a cookbook of things people can do to get accurate timings. SOMEONE PLEASE OPTIMIZE XRENDER! That's what makes text slow.
- Toolkits shield applications from the imperfections and perfections of the windowing system.
- Do we need an infrastructure to reduce latency? Should toolkits move towards retained-mode graphics? What about SVG? What extra things do we need from the window system (output-only windows, input redirection to get rid of grabs, window coverage events)?
-
Owen just gave me a good idea. You should be able to drag anything from an application into the Nautilus trash to delete it. In particular, you should be able to drag bookmarks from your file chooser into the desktop-wide trash can to remove them — you have to hit the "Remove" button right now. To make the trash can a generic trigger for deletions, we just have to make it take in a data type of application/x-delete-me with GDK_ACTION_MOVE; Nautilus will do nothing but ask the drag source to delete its data. Your drag source will get a GtkWidget::drag_data_delete signal, and you can do whatever is needed to delete your object there.
-
Got my hotel room sorted out; it turns out that Novell/Mexico had not actually paid for my room, but just made the reservation. I had a good sleep last night. I had dinner with JRB, Zana, Blizzard, and a bunch of people. Then everyone went to Flat Top's, a pool bar, where Mark proved to be quite the shark, even if he denies it.
Blizzard scores:
-
Stuart Anderson on standardization:
- X.org used to have a standardization process. Plans to move to Free Standards Group.
- Need a desktop standards group. [What about freedesktop.org?]
- Free software evolves so quickly that slow standards processes become obsolete and people stop caring about them.
- Initial versions of Xinerama got shipped incorrectly (static rather than shared libraries), people didn't push fixes upstream, things required incompatible changes later. Would a formal process have avoided that?
-
Jon Smirl on Longhorn and Mac OS X:
- 3D hardware is insanely fast. We should use it even though we use 2D most of the time.
- Web pages are now front-ends for many applications. The problem is how to present rich user interfaces in a web browser (e.g. web forms suck). Longhorn is a solution to that problem.
- Linux doesn't have a model for data binding — what translates web forms to actual database records? We do it server-side by hand; with XAML, you can do it with an XSLT transform.
- Microsoft: Direct 3D, everything accelerated. Apple: custom drivers, Quartz draws to RAM, things get composited in hardware; 2D is not accelerated, OpenGL is. Linux: we have to use OpenGL; everything accelerated.
- [Lots of MS strategy and such not really relevant to X.]
- Make OpenGL the system-level graphics API. Apps talk to Cairo (2D) or OpenGL (3D) directly. Stop development on 2D drivers; make XAA a transition to OpenGL. Remote Glitz would use OpenGL wire protocol.
Keithp walks in:
- We have the X server talking to the hardware, DirectFB talking to the hardware, and the VGA console talking to the hardware — that sucks. We need unified access, probably DRI? Talk about it at nospam-ksummit-2004-discuss@thunk.org (mailing list archive).
- We need a System Access Key (Ctrl-Alt-Delete in Windows), e.g. a non-spoofable key to get a real login prompt. Getting the key is trivial, but how to interact with the user?
- Need an API to switch video modes from applications, as well as output ports (VGA, S-Video, etc.).
-
Hideya Kawahara on Looking Glass:
- We've had 20 years with traditional 2D desktops. We have fast 3D hardware. How do we make use of it?
- When a window loses focus, it becomes partially translucent. That way it is easy to see your whole stack of windows.
- Each window has an icon with a small version of the window's contents. When the contents of a window change, so does the icon.
- You can flip a window around to write down notes. Is this useful?
- Bringing up a transient window makes the parent flip around, and the transient appears on its "back", which is now on the front. Supposedly this lets users better associate the transient with its parent.
- If you drag a window to the edge of the screen, it tilts sideways so that you can have more space near the center of the screen.
- You can view all open windows from the side, as viewing a stack books from the side of their spines — this is isomorphic to Exposé.
- The desktop background rotates in fractions of 360 degrees when you change virtual desktops. Supposedly this gives user a better spatial orientation of their current desktop — "my browser was on the side of the mountains; my word processor on the side of the lake".
- The CD selector shows the CD covers in a ring of discs; you select among them and the currently-selected disc comes to the front. What happens when you have 100 discs?
- Looking Glass is a modified X server, plus a Java implementation of the 3D desktop stuff. As this is a highly experimental area, Java's productivity is important.
Deron Johnson comes in to talk about implementation:
- There is ongoing work to free the source for Looking Glass — management has to figure out the licensing.
- Plan to add support for OpenGL capture, so OpenGL applications can run inside Looking Glass.
- X server gets modified to render to a buffer. Display server loads the buffer as a texture map, and puts it on a 3D surface. Similarly, input events are redirected from X window coordinates into surface coordinates; events are snooped with the Xevie extension, which originally was designed for accessibility purposes.
- Keith talks about the XTest and Composite extensions, and how you have to translate event coordiantes each time a compositor introduces a break in the window stacking order — what to do to avoid round-trips when each part needs to talk to the other to do coordinate conversion?
- Owen says, what happens with drag-and-drop? GTK+ must implement its own version of DIX's XYToWindow() for DnD; if a compositor changes the stacking order, this is no longer going to work.
- Room for improvement: render directly to a texture rather than a temporary pixmap; support 3D audio (Java supports that) so that applications with audio, which may not be in the center of the screen, sound in the right position in your ears.
-
Dave Reed and Dave Smith on Croquet:
- Croquet just blew my socks off. Although the immersive 3D world is sort of research-y and unusual, it is the malleability of the environment that is interesting. In principle the world is a huge canvas with sub-viewports that can expose other regions of the canvas within themselves. Then, we may be able to take ideas from that and reuse them within Mono/GNOME. I'm sure other people would get equally excited by the collaboration infrastructure in Croquet.
- This is one of those things where one can't describe it; you just have to see it.
-
I'm at the XDevConf in Cambridge. First good thing of the day: bagels and coffee available for attendees. I didn't have time to get breakfast, so this is great. Plus, the hotel is practically next door from Quantum Books, which is so unbearably tempting...
Owen, Havoc, and Trow are here. It's great to see them again.
Russ Nelson was summarizing the talks on IRC, and I took some of his notes to paste them here.
-
Jim Gettys's talk:
- The X11 drawing model was intended to be replaced since the beginning, but nobody had the expertise.
- Rob Pike and the Plan 9 guys showed that Porter/Duff compositing can be done fast enough, even in software, for the basic imaging model.
- 2D is fundamentally different from 3D.
- Glitz shows that Cairo can run quickly.
- New features don't get used unless they work on every X server. Dual code paths suck, as they never get tested.
- X is but one component of a desktop environment. Good citizenship may involve fixing other components.
- Application startup is painful. Latency kills you, not lack of bandwidth. Round-trips to the X server depend on latency. For GTK+, we have about 130 round-trips at application startup; we may be able to eliminate around 90% of them (we used to have about 130 round-trips; this got improved by batching/caching calls to XInternAtom()).
- X through SSH is replacing the X security extension. Fat clients are really expensive;
- X request stream is dominated by image data (gradient buttons, glyphs, etc.). This compresses extremely well, at a factor of around 300.
- Still, we often keep duplicated image data in the client and the server. How to avoid that without going down the dreaded path of XIE? And how do we avoid re-compressing a stream that is already compressed (e.g. don't compress the part of the protocol stream that is already transferring a JPEG).
- Toolkits should do a lot of the work (changes in screen size, for example).
- Xlib needs to handle loss of connection properly; think of wireless networks. We should be able to preserve and migrate sessions, but we should then handle authentication issues. Think of GNU Screen but for X servers.
- XInput needs support for hot-plugged devices, DBUS/HAL integration. We need support for networked input devices; think of a conference room with input devices for each attendee's table.
- We need network audio. Current audio servers all suck. Need low latency for VoIP; need synchronization facilities for audio/video.
-
Keith Packard's talk:
- Eye candy: translucent windows, magnifiers, window animation, flash-free window manipulation.
- Right now we have visible window management effects — resize a window, the WM resizes first, then the application's content. The plan is to double-buffer the whole desktop, and have cooperative buffer updates. The perceived effect is that everything is cleaner/faster — in Mac OS X everything is smooth, so it appears to be faster, while it is measurably slower.
- The plan is to have windows have their own OpenGL buffers for their content. Have the video card composite everything. Applications would draw everything off-screen. Have a Compositing Manager take care of compositing everything to the screen.
- But then, input is currently untransformed — if you had a compositing manager that changed window positions/sizes (a la Exposé), you would have to transform coordinates in input events.
- The compositing manager is actually another X client.
- Synchronized updates: pend updates until client drawing done. Have a synchronization facility between window manager and clients.
- Mouse input: X server figures out which window corresponds to the global mouse coordinates; it also defines the stacking order. This needs to be moved outside of the server.
-
Three new extensions:
- Xfixes: little things, Region objects.
- Damage: monitor rendering to drawables, send events indicating damaged windows.
- Composite: off-screen window contents, split window hierarchy.
- Arbitrary visuals: ARGB visual for transparency; implemented with the Composite extension.
- Keith started an Ultra Sexy(tm) demo; run a regular server and move windows around; things flicker as usual. Then, start the client for the compositing manager; everything is smooth now and it does appear faster.
- Another cool demo; use a client to set the opacity of a window arbitrarily.
-
Jim Gettys again:
- The Microsoft model of everything-in-client-PCs drives cost of ownership way up.
- If we bother to make X good for remote operation again, and with security, interesting things can happen.
- We should not require sysadmins at home!
- We need fine-grained access control. You go over to a free computer and want to display your stuff there, but it should not interfere with what is running in it already.
-
Eamon Walsh on fine-grained access control for X:
- Need to bring the X server into the Secure Linux project; need control on the level of individual protocol requests.
- For the kernel, SELinux provides hooks for security implementations, but there's nothing special for the network. There are some user-space changes as well.
- In the SELinux research agenda, there is local GUI security — the kernel and low-level user-space enhancements are practically done. Also pending are labeled networking and a network-wide policy.
- X clients can exchange information in several ways, including window properties, so that has to be secured.
- Just like the kernel, the X server could have a cache of security policy decisions.
- Goals for security framework: based on existing work; easily extensible; non-intrusive (callbacks rather than local code); works at dispatch layer to avoid performance issues; provides framework for arbitrary access control extensions.
- With the XC-Security extension, clients go to the server via a decision engine in the middle (Xext/security.c), that just has trusted/untrusted attributes. The idea is to replace that middle-man with a mechanism for callbacks (Xext/xace.c); those would make the security decisions.
- You can install new hooks, and the security people can actually implement the callbacks.
- KeithP suggests that instead of a flat list of hooks, it should be hierarchical for if a new security model wants to do more specific things.
- Tytso: what about return codes? It's different to know whether an atom exists than whether its value is actually accessible. Reply: the middle-man can modify the return values however it wants.
- Non-intrusiveness: that's why the core code is separate from the actual security implementation — middle-man versus the callbacks. The whole framework is a compile-time option.
- Sample hooks: CORE_DISPATCH, EXT_DISPATCH (replace XC-Security shadow dispatcher); RESOURCE_ACCESS, DEVICE_ACCESS, PROPERTY_ACCESS (replace SecurityCheck*Access() functions); MAP_ACCESS, BACKGRND_ACCESS (replace untrusted child & background "None" checks — windows with a background of None can be used maliciously to get the content of underlying windows).
- Can get client credits from several sources (basic PID, UID, GID, or a security context).
- For now, there is an AUTH_AVAIL security hook for remote clients.
- Future work: trusted window labeling (e.g. have your window manager decorate secure windows differently from normal windows — how do you pass this information? via a secure window property, for example).
-
James McQuillan on the Linux Terminal Server Project:
- Diskless think client; boot over network. Xserver -query to log in via XDMCP. Can run apps locally. Low hardware requirements.
- Widely used for schools (k12ltsp), kiosks, internet cafes.
- Current security is X through SSH. In the future, Kerberos, NFSv4, IPSec, or other options.
-
Ted T'so follows up on security:
- Kerberos is nice from ~200 users onwards. SSH is nice for up to about 100 people, but it gets messy when you need to remove authorized keys from many machines (e.g. someone from your organization leaves), or what if a key gets compromised. However, SSH is dead easy to configure. IPSec is nice, but managing certificates is cumbersome (and potentially buggy, as BSD learned).
- While large corporations may want to use something like Kerberos, individual end users may keep using SSH for convenience.
- X supports Kerberos, but a) it has likely bit-rotted, b) nobody knows how to set it up.
- Everything boils down to security requirements. What do your users need to do? What do you need to secure? If you arrive to a lab as a visitor and plug in your laptop, what should you be able to do?
-
Sigh.
-
Why I love Lisp inside Emacs, part N:
(defmacro with-system-time-locale (time-locale &rest body) `(let ((old-locale system-time-locale)) (setq system-time-locale ,time-locale) (unwind-protect (progn ,@body) (setq system-time-locale old-locale)))) (with-system-time-locale "C" (format-time-string "%A")) => "Monday" (with-system-time-locale "es_MX" (format-time-string "%A")) => "lunes"In general, the with-foo convention makes code much more readable and robust: consider Sawfish's with-server-grabbed, or any kind of with-mutex construct.
That was to fix a bug in the elisp code that generates my RSS feed; if I ran Emacs under a non-English locale, dates inside the RSS would get generated in Spanish, which is not valid.
-
The meeting between the GNOME Foundation and the Mozilla team is really good news! Tomorrow I'm flying to Boston for the XDevConf, and one of my plans was to get together with Miguel/Owen/Keith to see what we could do about a canvas-based drawing model for GTK+, based on Cairo. The Blizzard Man will also be attending, so we may as well throw in some Mozilla/XUL discussion into the mix.
-
Here are my slides from Colombia:
-
Enanos, simios, y otras alimañas: El Proyecto GNOME (HTML) — gnome.sxi (819 KB)
-
Moda 2004 para programadores de GNOME (HTML) — new-api.sxi (316 KB)
-
-
Today I gave my talk on a few APIs that people should be using in their GNOME programs: GtkFileChooser, EggRecent, gnome-vfs, and some of the new features in GTK+. It went really well; I'm glad that the slides came out so nicely.
Something happened to my laptop, though, and now the display is iffy. When I suspend it, the LCD slowly turns yellowish-green in a way that would be awesome for a screen saver, but it is disconcerting instead. And when I unsuspend it, the screen appears to be turned off, but then the green/yellow comes back. I have to really turn off the display with the CRT/LCD function key, then turn it back on so that it will display fine. I hope the stupid screen doesn't throw the towel on me, as I'll be traveling next week as well.
-
Last night I talked for a bit to Ismael Olea about things that we may need to add to gtk-doc to make reference documentation easy to translate. We have no concrete plans yet, however. Drinking in a really nice bar got in the way of the discussion.
-
I'm in Medellín, Colombia, at the Congreso Internacional de Software Libre. Oralia didn't come this time, though, and I am missing her a lot. It is the distance and delayed communications that hurt. Also, hugging together in bed at night when it gets cold.
Today I gave my talk on the history and status of GNOME, and how people can get started as contributors. After the talk I got many questions from people who are doing programming in GNOME, and the problem seems to be, mostly, that we lack enough programmer's documentation, and as a nice bonus, translations thereof. Things like Havoc's book are basically impossible to get here without outrageously expensive shipping costs, and the online version is of course obsolete, as it refers to GNOME 1.x.
-
Currently, we only have these convenience functions to set/get a GtkSelectionData:
gboolean gtk_selection_data_set_text (GtkSelectionData *selection_data, const gchar *str, gint len); guchar *gtk_selection_data_get_text (GtkSelectionData *selection_data);
These handle UTF8_STRING and the historical TEXT and COMPOUND_TEXT selection types from X.
However, we have no convenience functions for other well-known data types that get passed primarily through drag and drop, such as text/uri-list and application/x-color. Applications who handle those data types have duplicated code, different kinds of bugs as a result, and all sorts of evilness. I would love to have something like this instead:
/* Sets "application/x-color" from the 16-bit RGB values */ gboolean gtk_selection_data_set_color (GtkSelectionData *data, const GdkColor *color); /* Parses from "application/x-color", handling correct and incorrect * formats (old versions of Qt or KDE set the selection incorrectly). * Returns an error if the selection type doesn't have the right atom * or if parsing fails. */ gboolean gtk_selection_data_get_color (GtkSelectionData *data, GdkColor *color, GError **error); /* Sets "text/uri-list" from a list of char; uses the format * defined in RFC 2483 section 5. */ gboolean gtk_selection_data_set_uri_list (GtkSelectionData *data, GList *uri_list); /* Parses "text/uri-list" according to RFC 2483 section 5. * Returns a list of char *, or an error if the selection type doesn't * have the right atom or if parsing fails. */ gboolean gtk_selection_data_get_uri_list (GtkSelectionData *data, GList **uri_list, GError **error);If you are interested in stuff like this, see the corresponding mail.
-
New section for the GtkFileChooser documentation: file names and encodings. Read this if you don't know why g_filename_to_utf8() exists.
-
The day my dad and his wife came to pick up my sister, Oralia made fabada for lunch. We put the remains in the freezer. Today, back from Mexico City, I heated the leftovers and had them for breakfast. Fabada does taste better after the day it was made; it was delicious.
-
Material for gnome-love aspiring hackers: make GtkRange disable the scrollbar buttons when the thumb is at the limits.
-
Up to now, the new features to switch keyboard layouts had never worked for me. It turns out that I had not enabled the group-switching shortcuts in the Layout Options page of the keyboard capplet. Now pressing both Shift keys together switches layouts. Also, I had an old .server file with the same IID as that for the new keyboard layout panel applet, so the panel was not picking up the information for the new applet — so I couldn't launch it at all.
I wonder why none of the group-switching shortcuts were enabled by default. It would also be nice to automatically launch the keyboard applet as soon as you add keyboard layouts in the control center.
-
When my dad and his wife went to Xalapa to pick up my sister and ourselves last Friday, they brought along with them some cold cuts from Perote, a town along the highway to Xalapa where they make fantastic delicatessen. Their lomo embuchado is incredibly good. On the way back to Mexico City we also stopped there, and Oralia and I got some more goodies for my mom.
-
Last Saturday we had coffee with Juan José Martínez, Toño Salgado, and his girlfriend Ingrid, all members of the Xalapa LUG. Juan José told me about a little record store downtown, in the second floor of the otherwise uninteresting Pasaje Enríquez. It's called Promúsica, and it is quite good. The owners actually know what they have in stock, and they do know the music and make recommendations. I got some Honegger for myself and the sountrack of The Hours, by Philip Glass, for Oralia. Why Honegger? I've been building a small CD collection of music that Alejo Carpentier mentions in his books, and so far everything has been good. He had quite a taste for contemporary composers, something I appreciate a lot.
Toño came by on Monday afternoon. He's learning to program in Python, and he needs a hacker-mentor. He's very interested in music software, so perhaps in the future he'll be able to write something useful.
-
The file chooser supports tilde expansion now, so you can type "~/foo.txt" in the Control-l dialog and it will do the right thing.
-
My sister Carla came from Mexico City yesterday, and she'll be staying here until the weekend. It's great to see her again. She's rather nervous about finishing elementary school and going into secondary school hell. I wish I had had someone to tell me what it was going to be like.
-
Chernobyl photo gallery. This is truly frightening.
-
Yesterday I added support for selecting hidden files in GtkFileChooser. You can make it show hidden files in two ways: one is by bringing up a context-sensitive menu on the file list and toggling an option; the other one is by going to the Location dialog with Control-L and typing the name of a dotfile.
In other news, the file chooser now displays a busy cursor while loading folders. It's much more comfortable to use with that in place. These features are in CVS HEAD for now, and they should appear in GTK+ 2.4.1 soon.
-
Today we had a lovely walk through the Parque Bicentenario. It had just rained and all the plants and trees looked fresh, the stone paths were shiny, and the air was cool.
-
Some people were surprised about the boob job, which is of course an April Fool's joke. It's no joke, however, that I would love to have one of those scarves.