User:MitchellNCharity/Old: Difference between revisions

From OLPC
Jump to navigation Jump to search
(-bq. bq merged the paragraphs.)
(→‎Handwriting recognition links: moved to their own page)
Line 265: Line 265:
**assemble test list
**assemble test list



== Handwriting recognition links ==
*http://xournal.sourceforge.net/
*http://www.adebenham.com/gournal/
*http://www.dklevine.com/general/software/tc1000/jarnal.htm


== Why aren't more people developing for OLPC? ==
== Why aren't more people developing for OLPC? ==

Revision as of 21:40, 4 August 2007

Hi. This is my slush pile. Currently ordered most recent at bottom.

I'm an ex-MIT Laboratory for Computer Science "systems" person, who dabbles in science education reform, dynamic language implementation, and assorted other things.

I'm most interested in using the xo to help kids develop a feel for the physical world, by integrating "Powers of Ten" perspective, pervasive order of magnitude estimation and approximate quantitative reasoning, fermi questions, accessible foundational physics, storytelling and tangible representations.

But the xo has to exist for any of it to happen, so that's my main objective. "Widen bottlenecks as you hit them" has me mostly: editing the wiki; wrestling with disk images and otherwise trying to make it easier for developers to get started (and other people to get a taste); and trying to figure out how folks can best help the project 'make the base system solid and compelling, and win some orders'.

Google search: wiki.laptop.org and laptop.org (which in includes the mailing lists).

Graph paper background

Graph background 01.png

File:Graph background 01.ps

Converging lines to eyeball angles

A rough draft experimental tool to measure angles by holding up the laptop screen to your eye and looking along sight lines. Objective: measure the angle between things in the environment. The idea: hold up the laptop, and put your eye to the edge, so the lines all appear to converge at you.

Show-stopper(?): kentquirk reports a 20% viewing angle limit on B2 in both modes (with color artifacts at 45%). 20% seems perhaps borderline. But experiments on mockups suggest 20% could still leave you with ~1/3 degree accuracy. My current plan is to try it on a laptop.

I briefly tried it on an XO. Worked just fine. Seemed better than 1/4 deg. :) MitchellNCharity 15:14, 18 May 2007 (EDT)

The camera is an another way to measure environmental angles.

Here is a quick draft. I'll wait for the reality check before fleshing it out. Lines are at 1 degree intervals. The eye convergence is currently set at 100 mm from the screen edge. Converging lines 01.png

SVG might be a better tech. Or even python.

Random brainstorming

In User_talk:MitchellNCharity. And Talk:Educational_content_ideas#Laptop as flexible physical object (ruler, etc).

Hmm. Email went by on devel which suggested that using xo's for photometry was an unanticipated idea, possibly best noticed sooner rather than later due to hardware/microcode impact. Perhaps scavenge the brainstorm cruft for any other similar things which might have fallen through the cracks, and write up a post? (Doable) MitchellNCharity 21:05, 3 June 2007 (EDT)

How Big Are Things? at 800dpi might show up life-size on the laptop.

No, because the browser is scaling, so it's not 200dpi. Umm, why did I do 800 dpi? You can use the calibration page to get arbitrary corrections. (Doable: determine browser scaling and fix.) MitchellNCharity 21:05, 3 June 2007 (EDT)

Invitation by physical proximity. Creating a set of laptops using physical proximity. Say kids A,B, and C are physically adjacent, and wish to collaborate. Or you want laptop X to be, say, a remote control for laptop Y. Currently, you have to mentally map the already physically localized laptops into user space, and then gather the users by hand. It would much nicer if folks could do this automagically. Eg, a group points laptops at each other, hit a key, and bang, everyone's invited. The low resolution audio-delay determination of distance should be perfectly adequate for this. MitchellNCharity 13:23, 14 June 2007 (EDT)

Using one xo as a remote keyboard/mouse for another. Eg, when the other is in ebook mode, on the ground, with a microscope, sample, and slides perched on it. One side is an activity RemoteControl, the other an unmodified activity plus... what? MitchellNCharity 08:33, 18 June 2007 (EDT)

Addressing child/parent disassembly of xo's -- A "visible xo" document might reduce the appeal of xo disassembly? (If it didn't encourage it instead;). Taking apart and reassembling the xo seems a reoccurring user activity. As does the xo not quite coming back together (missing screws, disconnected wires). And I assume some static fried chips, though I've not seen that reported. Nicely connectionist, but maybe we want to make this less common? Maybe not. If so, perhaps each xo could have a "visible xo" document in the library, showing the inside of the xo so thoroughly, in such detail, that looking inside had less appeal? "Been there, seen that, why bother?" Maybe it gets done once, the pictures are seen to show it all, and other folks don't bother. Maybe. MitchellNCharity 00:49, 20 June 2007 (EDT)

xometry (noun) - measuring with an xo. :) MitchellNCharity 11:29, 20 June 2007 (EDT)

No. "xometry" is the easuring of xo's. A much less interesting concept. :) MitchellNCharity 11:15, 28 June 2007 (EDT)

Help wanted. wiki page?. mailing list?

  • "Nerd with hardware wanted". Examples: "I don't have hardware. Could you download this draft activity and stopwatch how long it takes to do mumble?". "I have a B2. Could someone with a B3 run a profile of this code, so I can better prioritize my performance optimizations?".
  • "User(s) with hardware wanted". Examples: "Could you download this 'put a ruler on the screen' draft activity, try going around measuring things, and then provide user feedback? Do you find you are dinging up your screen?".
  • "User testing wanted". Examples: "I have a new sugar game. Player testing and feedback wanted".
  • "Quick experiment wanted". Examples: "What color is the top left pixel?"; "At how shallow an angle can you read the xo screen?"; "If you browse this webpage image of converging lines, can you successfully use it to eyeball angles? How precisely?".

I'm not sure the community is large enough to sustain this yet, as turnaround time is usually important. And .xo (un)install isn't easy yet. MitchellNCharity

Printable xo mockups: cheap and easy creation of xo mockups for developers without hardware to do crude user testing. Printable to-scale images of various xo profiles. Tool/recipe for sugar screen dump -> printed screen mock. Is there actually any demand for this, or is it just specific to my "measure the world with your xo" interests? MitchellNCharity 11:21, 28 June 2007 (EDT)

Created Life-size laptop pictures.

puzzle construction. cut up images/drawings, play and share. MitchellNCharity 09:59, 8 July 2007 (EDT)

NASA's Visible Earth images for OLPC countries - Possible demo content?

NASA's Visible Earth has images sorted by country. Perhaps they might make good demo fodder. Some are whole country, others just a feature. Some are topical (some fire, or erosion, etc). Most have descriptive text (in english). Some stereo pairs.

TCC Tiny C Compiler - Include C toolchain on OLPC laptop?

My impression is a gcc toolchain is not planned for the laptop, due to space limitations (a fuzzy recollection - I now can't find it mentioned). So I note TCC, a Tiny C Compiler for x86. GCC compatible, small and fast. Same author as QEMU. So a C toolchain would fit on the laptop.

I also note its libtcc, which permits disk-less run-time C compilation and linking. Just in case anyone encounters a need for it.


Wow, I've never seen that!

From Educational content ideas.

If you have never seen snow, a short video clip of snow falling in the woods could have a high "wow" factor. The kind of thing kids watch again and again and again. Similarly for ocean, mountain, desert, farm, rainforest, savanna, metropolis, cow, whale, raptor, etc, etc. TV and movies take some of the edge off, and you can surf... but consider an example. Say I want a "cow" moment. Just as if I walked up to one. I want a cow face in front of mine, with a ear twitch, blink, and swipe of a big wet nose with a long flexible tongue. What I don't want is fifteen minutes of wading through the internet, google, youtube, etc, trying (unsuccessfully) to find something similar. I want someone to have already groveled around and collected wonderful breathtaking clips. Volunteers? MitchellNCharity 18:44, 1 May 2007 (EDT) (further thoughts on my user page)

  • And perhaps flesh out each clip with pictures, which can be more diverse, and higher resolution, without blowing space budgets.
  • Exposure to the unfamiliar world.
  • Office, lab.
  • Might creating these be a task for kids using the olpc camera?
  • Also "task videos" - riding a train, a plane; filling out paperwork; ... And familiar tasks in unfamiliar contexts, eg, women surgeon washing hands, US "all employees must wash hands" bathroom sign, a paw being washed, chimp washing hands, mumble.

How Big? - a possible activity

One thing I'm working on is an exploratory technology prototype for a "How Big?" activity. It's still vanilla pygtk rather than sugar. Currently, cairo image scaling speed is the biggest challenge, and appearance on an actual laptop display the biggest unknown. MitchellNCharity 22:13, 12 May 2007 (EDT)

Screenshots looked great on a B2 XO! :) Leaving speed and space (10 MB or so:/) as current challenges. And perhaps doing the sugar/hippo ui. MitchellNCharity 15:22, 18 May 2007 (EDT)

Activity howbig 01.png Activity howbig 02.jpg Activity howbig 04.jpg Activity howbig 03.jpg

Doables:

  • include a "How Big?" screenshot of south-america.


Towards a better calculator - augmented and bantering

Calculators on computers are vastly more boring than they could be.

Examples:

  • Enriched keycaps. The key has a "3" on it. When pressed, or randomly, that "3" could become "lll" (3 hashes). Or the local word "three". Or accompanied by the word "three". Or 3 dots in a line, or equilateral triangle, or right triangle. Stationary or spinning. Or a triangle with lines rather than dots. With lines emphasized one at a time. Switching from dots to lines one at a time. Accompanied with counting. Etc, etc.
  • Slide the decimal point (either grabbing it, or with buttons), zero padding and going exponential as needed.
  • Say the number. The "kid randomly hits keys, generates content, and wants adult/computer to say it" game.
  • Sidebar banter, see below.

Sidebar banter

A calculator doesn't need much screen realestate even in full color. So we use it! Everything described for keycaps can be show on the side, for the whole number. And it can be manipulatable. Add another dot, the number goes up. Grab and move the pie wedged edge, and the fraction changes. Much of what is normally taught about numbers in isolation can be "fun stuff happening on the side". Counting; counting with hash marks; counting on fingers; how to draw a number; place-value system; geometric characteristics of numbers (ways to arrange dots; shapes); decimal point; number line; exponential notation; fractions; pie charts; graphs. And a whole lot more. It's not even hard. Just a framework, and then one hippo widget at a time as folks get around to it.

Other examples of "so much more one could do":
  • Notable numbers - calc says "1111", banter that town population is "1300". Population (class, school, town, etc), age (students, teachers, parents), time (minutes, hours, etc), durations (of school day, holidays, etc), distance (in km, travel time, etc), length (of... everything), mass, prices, food consumption, energy/sunlight, characteristics of materials, etc, etc. Which of course sets them up to be able to do fermi questions. Also mathematical properties (prime, factors, etc).

Notes:

  • Banter should be replicable. For doing repetition. And so having found a neat one, you can show it to someone. Ideally even on the others' machine. How? A banter history with sharable entries?


Vision screening

Sj expressed some interest in vision testing, so...

Lighting

Illumination levels have to be within certain ranges. Order 1k lux. I'm not sure yet how critical it is, or whether we can compensate. And I don't have a feel for how large the range is subjectively, and thus how amenable to "give the human instructions".

Can use a laptop in photometer mode to judge the illumination on another.

What is the backlit illuminance? #olpc silent. I don't know if the backlight might be useful or not.

Screening plans

Visual acuity

There is various better-that-Snellen literature. Some won't apply, as we can show letters one at a time and choose arbitrary sizes. Some will... but I've no medical closed-literature access anymore. :(

Internationalization

Symbol charts tend to be culture specific. Letter charts script specific. For our target age group, letter charts are recommended. Near-term, we can just use latin. One can get the letter sized right in any script, but judging the letter's difficulty requires experimentation to create a baseline. Not a big deal, order 100 people should do it, but one more thing to worry about.

Misc

Game Jam feedback

Random ideas...

Intake procedure:

  • have a punch-list; tinderbox&milestones?
  • wiki page; who am
  • skills inventory
  • photo
  • svn repo up
  • xo or emulation up; hello world activity running
  • on irc
  • automate more

Support:

  • "best way to learn topic x" pointers (with better docs it won't matter, but for now...)
  • mit zephyr-like announcement channel, ie, popups. less invasive than group shout, more visible than irc chatter.
  • irc in/out flag - eg key developer sleeping -> "he's there but not responding"
  • "events" to give time texture. brownies&lemonade; cookies&milk; accompany with demo.
  • code review; experts on call; "lab TA's" (mit sense)
  • TDMultiplexing experts - need to self monitor, or timeout, or have badguy, or... better: expert moves to questioner, so easier to leave, other questioners can come and draw them away. virtual "raised hands". request assistance in domain x queue? skill level escalation mechanism.
  • pairing with offsite mentors (individuals or teams)
  • active facilitation role needed/useful - emergent social organization generally suboptimal. someone whose job is it to know peoples interests, status, and plans; make sure useful/interesting connections happen; smooth the path for folks getting stuck; watch for large scale issues/problems.
  • visible team status; maybe support folks help maintain; entire jam current state at a glance. attempting the benefit of everyone walking around asking "what's up?", without the cost.
  • faq/bulletin board - eg, "both sets of game keys unfortunately currently have the same codes, and they are foo".
  • might use intro tutorials to bootstrap "folks gather at lounge tables for mini-seminars" behavior.
  • generalize/explore/expand - task-oriented efficiency, "on a roll", principle programmer model, vs "rtfm/c and good luck".

Misc:

  • really really need good mono-color explanation
  • questions re degree of intended olpc-gnash integration
  • dedicated xo's for each kid attending (even unanticipated kids w/o xo's in room full of xo's just seems wrong)
  • have a water cooler
  • extra batteries. this jam, most everyone was stationary and could be plugged in. i suspect this will change. "coding better than charging". maybe.
  • quiet room automated enforcement/encouragement? ie, sound meter.

Excellent job by the Olin folks, Sj and Mchua.

Determining the bounding box of a cairo surface

I've no idea where in the wiki this should go, so...

There is no easy way to determine the bounding box of a cairo surface. Two approaches come to mind:

1. During drawing, you can use stroke_extent() on individual strokes (and fill_extents() for fills). So create an "extent logging" wrapper proxy for surface classes, which records extents as the strokes and fills go by. This might be a nice solution, but I haven't tried it.

2. Kludge. Render twice, first to a large postscript surface, and then run

gs -sDEVICE=bbox -dNOPAUSE -dBATCH file.ps

Given an output of

%%HiResBoundingBox: 0.000000 359.981989 245.002985 410.021987

the bounding box is 245 by 50.

Wiki change proposals

Some change proposals get quickly adopted. Some not. Here I track a few of the latter.

  • wiki sidebar, section "search", needs a google link. Some searches are just intractable with the build in engine. via google or via google. I often walk the path of 'try search box, fail, then homepage->here->google'. For instance, you can't search for "art". But you _can_ on Wikipedia. So unless they are using a different search backend, perhaps this is simply a misconfiguration.
  • wiki sidebar, link "Learning parables" should be removed. (discussed elsewhere)
  • wiki sidebar, a link For educators should be added. (discussed elsewhere)
  • wiki sidebar, organization/grouping needs to be looked at with a fresh eye. I've grown used to it now, but early on found it annoyingly counter-intuitivly grouped.

Some not proposed, because others of a similar flavor went nowhere:

  • wiki sidebar, Mailing_lists should link to a real page, so we have a place to put google links for searching the lists.

Display simulation for screenshots

simulate_display.py takes a screenshot, a lighting ratio, and creates a simulation of what an xo screen would look like. For those of us without hardware to check for color artifacts.

Here is a photograph of the real screen EToys - new display.jpg.

Doables:

  • doc or fix pgm "the snapshot must have at least some color to work" bug
  • move this to {a section in Display, it's own page, it's own page redirecting to a section in display (so I don't have to keep updating links)}.
  • rename "simulation"->fake. Simulation implies ground truthing, or at least the expectation of successful truthing. No such exist at present.
  • the bogus border is unnecessary now - remove it.
  • cairo canvas wrapper which renders to pixmap, and simulates for some (ENV?, time cycle?) specified lighting. Can overwrite canvas class?
  • find out what color the top left pixel is.
  • maybe find the etoys page and truth?
Hi, maybe it would be usefull to tell that xwdtopnm and pnmtopng are needed for the script. In ubuntu they're in package netpbm, btw.HoboPrimate 22:45, 25 June 2007 (EDT)

Doables - week of 25 June

  • find out what's going on with the mini content release. can I get ruler/grid on it?
  • shift to a public repository. coderanger and/or dlo. ask cr if he's ok with a "common" tree, to which he/I can give others access. maybe dlo for releases, and common for dev. common is an opportunity to nudge dev towards a more social model, rather than me being just another person who surmounted the barrier to entry.
  • write yrgb->rgb mapper. in Lab.
  • generate palette using color mapper.
  • break activities into smaller chunks. use this to prototype a 'common' layout.
  • explore 'bundle tree as a single python file'? to make activity (re)assembly/renaming more fluid.
  • figure out how much of the sub-activity code is still useful.
  • vision screening. get Sj his demo..
    • contemplate scope
    • visual acuity -> standalone
    • create toy driver. review va math and optotypes.
    • create vision screening page. oy, there's a slog. (postmortem note: assume all activity projects will escalate; even though current hardware makes it a pain, do full proj docing as you go).
  • clean up protractors. explore colorization.
  • add in 'angle site lines'
  • contemplate pragmatics of string inclinometer. generate protractor for it. separate activity? embedded docs? lesson plan? use this to explore content context.
  • experiment with "numbers on the diagonal" with grids
  • tooling up: explore creating a widget, gtk timers, gtk mouse and move (eg, for click-n-drag ruler), gtk dragging, gtk/pygame integration. web embedding. is widget layering/stacking possible?
  • explore possible constraint/data propagation infrastructures. sigh.
  • stopwatch. quickie or tool driver? both? tooling path: widget. then integrate with a dot-per-second widget.
  • coin stack widget. for widget exploration, and random prettiness.
  • proto bantering infrastructure: calc widget + coin stack + dots + propagation (likely out of scope for week)
  • Camera dupe with angle grid overlay. gstreamer level or cairo? gstreamer would make it a trivial mod, ease upstream tracking. another approach would be a click-n-drag angle "ruler". The 'how far is your friend' gets positive reactions. perhaps combine with 'angle sight lines' in a laactivity.
  • do some vocabulary creation. "little-a activity" isn't a long term solution. sound out T. just how much of problem is "activity" being rebound going to be?
  • toolbar icons should have ENTITY stroke and fill_color overridden just like frame icons. at the moment (build443), it seems for compatibility, you should use toolbar's FFFFFF/none in both places, which sacrifices previewability. an itch, and opportunity for first sugar core contribution. Perhaps "Icon could have a way to set entities", "the code to do entities is in canvasicon.py right now", "we need to factor it out in a way that allows both canvasicon and icon to use it" (marcopg).
  • explore rulers/grids in other programming environments. squeak, what else?
  • howbig
    • sugarize proto. check performance, clarity.
    • project plan.
    • try 'get image from net, as url or bits, and save to journal'.
    • screen ruler. scale meta info -> image/journal (exists?) or file.
    • prototype stacks ui
  • screen faker
    • proj space exploration and plan sketch. is there really a need?
    • clean up code
    • doc pgm bug
    • can parasitize cairo canvas? then, any low-hanging performance tweaks? current local+pos input permit different pixel access mechanism? inline C? :).
  • wiki infrastruct
    • look for download/upload tools. tool driver: grab activities, add 'Ticket for component' links. Next step maybe an auto-generated 'activity page status' report page.
  • olpc hardware access
    • missed meeting. :( when's next window?
    • assemble test list


Why aren't more people developing for OLPC?

<semi-flame>
A month or few ago, someone (approximately) asked me "Why aren't more people developing for OLPC?". My answer was that, to a zeroth approximation, they can't. I believe this is still the case, so I'm revisiting the topic here.

The official story is that it is easy to start developing for sugar. You can use the Live CD, or use sugar-jhbuild, or run in emulation, or use real hardware. That this last is even on the list, gives a feel for how narrow the current audience is.

The Live CD. Last updated 3 months ago. Doesn't run 1200x900. While handed out after talks as an easy way to start developing, I have never actually seen sign of anyone developing on one of these. But, to be fair, I've seen the occasional email suggesting someone is trying, and I've never sent a survey to the list, so maybe they are out there. I would be unsurprised to find no one is actually using this.

Emulation. At least on some platforms, emulation now works. Sort of. It took something like a man month of effort to get things like network and 1200x900 working, and minimally documented. On some platforms at least. Though it hasn't been bootable this last month. With some additional workarounds, as of today, it again boots. Yay. Oh, and there's never been sound or video. Nor internationalization. The core developers don't use emulation, and there is no automated testing.

Sugar-jhbuild. "It's easy" is often said. Meaning it might work on 32-bit Fedora or Ubuntu. If you can manage to download all the sources, from servers which are not infrequently down. And then build them. There was, briefly (a couple of weeks), a working tinderbox. Lots of red. It is not clear whether this is really a viable option for people, and on what platforms. It well might be. Or not. Once you get it working, it can be kept working, as it's use by the core development team shows. But we don't even know how often an attempted download will succeed. No testing.

So maybe Live CD, emulation, and sugar-jhbuild, are all working just great. Sort of. Or maybe they all have prohibitive barriers to use. It's hard to tell without testing and otherwise gathering that information. And thus hard to address problems.

The status quo seems to be: those few who scale a substantial barrier to entry, find some stable niche, and hang out there. And that no one is tasked with road building.

Road building. Which brings us to the development model. OLPC is currently using what I think of as the historical Red Hat closed open-source development model. Closely held repos; dependency on non-public build context; no public compute infrastructure for community tool building, so a tool impoverished environment; still in "the world consists of the core developers, and those which distract them" mode, never having grown into a diverse self-sustaining community in which the core developers are embedded. It's not that community involvement was tried and failed. It seems to never have been an objective. Contrast the resources, and approach, which have gone into the OLPC wiki.

For folks who have never been part of an open development project, let me use an example. Say I, as a random non-core community member, notice the tinderbox is broken. In say the perl6/pugs project, I would already have commit access to the one repo (by virtue of merely not having avoided the collective community effort to give a commit bit to everyone who looks even slightly interested). The tinderbox code is in the repo. Shell access to the public services machine is slightly more closely held - if one of the many random community members with access isn't on irc, I would have to send an email with... my email address. In summary, I could fix the tinderbox. I, or anyone else in the community. In OLPC, I can nag one of the vastly overcommitted core developers to fix it. One of only two or three candidates. This difference in barrier is the difference between having a tinderbox, and irc doc-bots, and commit-bots, and, well, eval-bots doesn't really apply for sugar, but the difference between having a rich environment or an impoverished one.

There are lots of variously skilled python programmers and others out there. And OLPC has a potential attractiveness that literally no other OSS project comes anywhere even vaguely close to matching. But this vast resource is just not being tapped. And TRIAL-2 is less than a month away.

TRIAL-2 is less than a month away. I could send mail out to python lists saying "Have some time? Off for the summer? Come change the world! 1000 kids could be using your software within two months. With your help, maybe a million by end of year". Inspiring. Sure to draw a crowd if explained. But what would I then suggest they do? "Well, if you are on fedora or ubuntu, I can get you started with an old disk image running under qemu, no sound. Or you might be able to run sugar-jhbuild, if you've a day, though I've never managed it."

We need a single unpartitioned repository which is just about as easy to suck people into as the wiki is. We need a semi-publicly accessible infrastructure machine. We need a serious effort at development community building.

And, well, I have no idea when it has to get done by. At best, you get that kind of information by reading between the lines, and never being sure if you've read correctly. But to have any impact at all by month end (July), it would have to happen explosively.
</semi-flame>