NetSurf / FrameBuffer #2

今日もframe buffer portをモニョった．

FTC_CMapCache_Lookup()でsegvるのは，error handlingがタコだったのとフォントへのパスがhardcodedで間違ってたのが原因．で，まずはSDL+freetype on Xで動くようになった．


さすがに黄色過ぎるので，internal pixel-colour変換を直すと，テキストの色は正しくなった．



endian dependな部分は泥臭くもあり，面白くもある．変な風にinternal pixel-colour変換がぶっ壊れると色がぶっ飛ぶ． :DDD

 02/28 01:24:33 hiyuh
    any idea? http://dev.gentoo.gr.jp/~hiyuh/misc/BE2.patch
    http://dev.gentoo.gr.jp/~hiyuh/misc/nsfb-BE2.png
 02/28 01:25:15 hiyuh
    it looks like only image ablend is broken, I guess.
 02/28 01:25:54 Kinnison
    looks like that
 02/28 01:26:13 kyllikki
    blue and green traded in your pixel to colour
 02/28 01:32:15 hiyuh
    http://dev.gentoo.gr.jp/~hiyuh/misc/nsfb-BE2-web-colors.png
 02/28 01:32:35 hiyuh
    text color is correct?
 02/28 01:33:03 kyllikki
    yes
 02/28 01:33:29 kyllikki
    the alpha blending is broken because the values its reading in to
    mix are wrong
 02/28 01:34:23 tlsa
    opaque images are wrong too
 02/28 01:34:33 hiyuh
    wtf :)
 02/28 01:34:57 tlsa
    well the screenshot is opaque, so is the logo image
 02/28 01:35:20 kyllikki
    I have no idea then
 02/28 01:35:38 kyllikki
    maybe the image formatters are getting it wrong on big endian systems?
 02/28 01:35:58 tlsa
    they'd be bust when he runs nsgtk
 02/28 01:38:40 hiyuh
    nsgtk works correctly on BE.
 02/28 01:40:11 tlsa
    there's something more wrong than just channel swapping
 02/28 01:40:36 tlsa
    cos the logo in the download box and the logo in the mast head should
    be the same colour
 02/28 01:40:46 tlsa
    but they are completely different
 02/28 01:45:01 tlsa
    actually, that's cos the background is different and the images are
    incorrectly getting alpha values other than opaque
 02/28 01:47:36 tlsa
    does the gtk build swap bitmap data round on BE?
 02/28 01:49:11 tlsa
    cos it looks perfect except for bitmaps
 02/28 01:54:21 hiyuh
    IIRC, nsgtk's image problems on BE was only in libnsgif code base
    (maybe it was in netsurf/image/gifread.c).
 02/28 01:56:37 tlsa
    well there's something certainly wrong with images
 02/28 01:56:50 tlsa
    and nothing wrong with text colours
 02/28 01:57:20 hiyuh
    true

と言う訳で，もう少しゴニョる必要があるのだったのだった． :)

NetSurf / FrameBuffer

#netsurfがframbuffer portで盛り上がっていたので，モニョる． :DDD

 02/26 21:01:19 hiyuh
    what does NETSURF_FB_FRONTEND=able mean?
 02/26 21:01:50 hiyuh
    runtime linux/sdl detection?
 02/26 21:02:29 tlsa
    to pick whether to build nsfb for the linux, able or other framebuffer
 02/26 21:02:54 kyllikki
    the list is now able, sdl, linux and vnc
 02/26 21:03:16 kyllikki
    sdl, linux and vnc are mostly what you will be interested in

 02/26 21:24:47 hiyuh
    nsfb-{sdl,linux} doesn't work here. :(
 02/26 21:26:09 rjek
    You won't get much help without details :)
 02/26 21:26:54 hiyuh
    http://dev.gentoo.gr.jp/~hiyuh/misc/nsfb-linux-r6625-fsckedup.jpg

 02/26 21:27:11 hiyuh
    color is fscked up completely. :)
 02/26 21:27:22 rjek
    Oh, an endianess problem.
 02/26 21:27:30 rjek
    TTOTD: Don't use big-endian machines.
 02/26 21:27:33 rjek
    :)
 02/26 21:28:09 hiyuh
    keyboard/mouse doesn't work too, so I did force powerdown this
    powerbook. :DDD
 02/26 21:29:22 hiyuh
    tested only x86 boxes?
 02/26 21:29:27 rjek
    And ARM.
 02/26 21:29:33 rjek
    Extensively on ARM, in fact.
 02/26 21:29:39 hiyuh
    ah, k
 02/26 21:31:10 jmb
    awesome
 02/26 21:31:17 jmb
    kyllikki: that rocks
 02/26 21:41:02 jmb
    hiyuh: I assume keyboard/mouse work correctly under sdl?
 02/26 21:43:45 jmb
    as for endianness, it's likely in the plotters
 02/26 21:45:18 hiyuh
    jmb: sadly, nope. sdl's one said "Unable to init SDL: Unable to open
    mouse" then segv immediately (was tested on X/fbconsole).
 02/26 21:45:29 jmb
    hiyuh: oddness
 02/26 21:46:59 jmb
    hiyuh: I assume other SDL apps work?
 02/26 21:47:16 hiyuh
    jmb: let me check
 02/26 21:47:51 jmb
    hiyuh: because the SDL_Init() call we have is entirely standard :)
 02/26 21:51:09 hiyuh
    jmb: it works on X, but it doesn't work on fb. maybe, causes of
    "sdl's ones didn't work on X/fbconsole" are different.
 02/26 21:52:13 jmb
    hiyuh: right
 02/26 21:57:48 hiyuh
    jmb: I guess X blocks mouse device access from SDL when I was on
    fbconsole, I'll check w/o running X (no VT switch).
 02/26 21:57:50 jmb
    there. updated the application
 02/26 21:58:02 jmb
    hiyuh: wouldn't surprise me
 02/26 21:58:09 hiyuh
    yup, bbl

 02/27 00:17:11 hiyuh
    jmb: cause of SDL_init error comes from my udev rules were
    fscked up, sorry. even though, sdl's one still rocks. :)
    http://dev.gentoo.gr.jp/~hiyuh/misc/nsfb-sdl-r6625-fsckedup-270.jpg

 02/27 00:18:34 Kinnison
    hiyuh: are colours in web pages wrong too?
 02/27 00:18:40 Kinnison
    Or just the furniture?
 02/27 00:21:05 hiyuh
    nope, it can not show any page ATM.
 02/27 00:21:51 Kinnison
    Odd
 02/27 00:21:56 Kinnison
    why not?
 02/27 00:23:24 tlsa
    Kinnison: web page colours were wrong in is photo:
    http://dev.gentoo.gr.jp/~hiyuh/misc/nsfb-linux-r6625-fsckedup.jpg
 02/27 00:27:06 Kinnison
    hmm
 02/27 00:27:39 Kinnison
    presumably this is kyllikki needing to support endianness stuff in
    his plotters?
 02/27 00:28:54 hiyuh
    fb_plotters_ablend()?
 02/27 00:30:37 Kinnison
    that's the alphablend isn't it?
 02/27 00:31:16 hiyuh
    I guess. it calculates r/g/b.
 02/27 00:50:09 kyllikki
    hi
 02/27 00:50:14 tlsa
    re k
 02/27 00:50:59 kyllikki
    right for big endien the least invasive way is to have different
    plotter c files built
 02/27 00:51:44 kyllikki
    e.g. fb_32bpp_plotters_be.c
 02/27 00:52:14 kyllikki
    because although the machines words are BE it appears the framebuffer
    display is LE
 02/27 00:52:39 kyllikki
    I suppose I could make it generic but...
 02/27 00:52:44 rjek
    How much of the code is actually different?
 02/27 00:52:47 rjek
    Might #ifdefing be better?
 02/27 00:53:08 kyllikki
    rjek: I really didnt want a load more #ifdeffery
 02/27 00:53:20 rjek
    Well, if large bits of the function will be different, I agree.
 02/27 00:53:30 kyllikki
    I am thinking of ways to remove the existing ifdefery
 02/27 00:53:52 kyllikki
    rjek: its every time an output colour is computed
 02/27 00:54:04 kyllikki
    so its not everso often
 02/27 00:57:04 kyllikki
    no, i see it
 02/27 00:57:08 kyllikki
    I can generalise it
 02/27 00:57:24 kyllikki
    in fact i ought to
 02/27 01:02:11 tlsa
    is it easy to add the rest of those pointers?
 02/27 01:06:42 kyllikki
    what #define can i use to determine endiness at compile time?
 02/27 01:09:55 hiyuh
    ? http://dev.gentoo.gr.jp/~hiyuh/misc/netsurf-gifread-try2fixonBE.diff
 02/27 01:11:02 kyllikki
    cool
 02/27 01:14:53 kyllikki
    try that?
 02/27 01:15:04 kyllikki
    I *think* my shifts are right
 02/27 01:15:50 kyllikki
    you get to test if #if __BYTE_ORDER == __BIG_ENDIAN is right
 02/27 01:16:45 hiyuh
    ACK
 02/27 01:19:38 *
    kyllikki waits paitently

 02/27 01:29:08 hiyuh
    lol
 02/27 01:29:09 hiyuh
    http://dev.gentoo.gr.jp/~hiyuh/misc/nsfb-sdl-r6635-onX-fsckedup.jpg

 02/27 01:29:46 hiyuh
    mouse/keyboard works. :)
 02/27 01:29:49 kyllikki
    hmm thats 32bpp?
 02/27 01:30:12 hiyuh
    dunno, just run nsfb on X.
 02/27 01:30:23 kyllikki
    well 32bpp is the default
 02/27 01:32:49 hiyuh
    let me try -linux.

 02/27 03:39:52 hiyuh
    on 32bpp nvidiafb, -linux shows red screen, keyboard/mouse doesn't
    work and cpu usage goes 100%  then I did force poweroff.
 02/27 03:40:01 hiyuh
    FYI, generic ppc machine uses offb which is 8bpp by default.
 02/27 03:40:05 hiyuh
    kyllikki: ^^
 02/27 03:40:39 kyllikki
    right, the 8bpp plotters may indeed be funted
 02/27 03:41:07 kyllikki
    hiyuh: I will take a look, maybe later maybe morrow
 02/27 03:41:15 kyllikki
    tired atm
 02/27 03:41:20 hiyuh
    k, thx
 02/27 03:41:24 kyllikki
    np

つーわけで，次回に持ち越し．
今のところ，一番まともなのは-sdlを32bpp nvidiafbで動かす場合かな．
nouveauの中の人は{nvidia,riva}fbはstateがゴチャゴチャになるからofonlyにしろって言うし，offbは32bppにならねーし，どーすっか．


で，NETSURF_FB_FONTLIB=freetypeにしたら，FTC_CMapCache_Lookup()でSEGVるのを発見．
gcc-4.3.3のbugかもしれんので，熟成の方向で． :P

tanslate(mercurial_ref, "ja_JP.UTF-8");

例のAでLな"ぷろじぇくつ"の残務も一段落したし，matsuu先生がブツの日本語訳を御所望なのでモニョる．
PowerBookにmedia-gfx/inkscapeをemergeするのはgnome depでコンパイルがダルいので，会社のパソコンにWindows版のinkscapeをぶち込んで仕事をしているフリをする． :DDD

で，出来たので晒す．


もっと大きいのが欲しいひとはこことここにあるので，御自由にどうぞ．
SVGはここに突っ込んだ．訳がヘチョいのとフォントがMS-Gothicになってるのが御愛嬌と言うことで． :P

emerge nouveau

AでLな"ぷろじぇくつ"に20kLOC overのVHDLを書きまくって，なんとか終わったのでnouveauで遊ぶ．

一応，upstream(?)のもあるにはあるが頑なにGallium3D対応にしようとしないので，今までmedia-libs/mesaだけ野良buildしていたがebuildにしてみた．proj/x11.gitとかからlive ebuildをパクってきて，モニョる． :DDD

まだTTMがアレなので，アレだのう． :S

if (!open("/dev/code", O_RDONLY)) { ...; }

NWCCの中の人とのメールの一部．

俺の好きなマイナーなブツに限らず，dev codeを拝めないことは良くある．メジャー/マイナーに関わらず，dev codeは爆弾な訳で，だからこそ見る側からすれば面白かったりするのだが，出す方は迷いもある．作り手である分，被害が具体的に予想出来てしまうのが原因，多分．

DSCMで気軽にブツを晒せる昨今になっても，changesetを綺麗にしてから出直して来いとか言ってrejectするメジャードコロでもあるまいし，自家製爆弾くらいを公開したって別に良くね？一つや二つ程度の地雷を踏んだ程度でupstreamの追っかけを止めるくらいなら，最初っから追っかけなんかやってんな，と．LICENSEは読んだか？ちゃんと，「爆発しても知らねーからな！！1」って書いてあるよ，だいたい．爆弾だったとき文句を言われのが嫌？別に爆弾じゃなくても文句なんていつも言うだろ，実際． :DDD