TODO:
    test colors ${\red x}$
    test axodraw
    test does LatexTemplate get added with plain text pagetype? (no)
    "thinness" of new images is due to use of alpha channel  (e.g. alpha is 16%
    and color is 16% results in a pixel that is 2.56%)
        Derive "white" from image
        Can dvipng handle alpha for me?
    Alpha: fix normalization of color, leaves two angles.  Easy to derive right
    color for a given alpha (fix norm of color at max).

Paper sizes:
    1. run gs in bbox mode to get sizes
    2. Define custom paper sizes (create a .ps file for gs to include)
        a. /usr/share/gs-esp/7.07/lib/gs_statd.ps for examples
        b. use /margin and similar operators to get rid of margin
        c. set imagingbbox to be same as paper size
            ImagableArea?  -- in ppd files
    3. gs -I switch to specify include file with my papersizes

    bboxes don't change with paper size (but need letter or a4 to get
    bboxes)
    See: http://raph.levien.com/Devices.htm
        for an example of how to set HW resolution using /.HWMargins and
        /PageSize in stcolor.ps
    See also /usr/share/gs-esp/7.07/lib/align.ps
    gs -gXxY forces png's to be XxY in size (doesn't work)
    Add to latex preamble: (moves text to UL)
        \oddsidemargin -86pt
        \headheight 0pt
        \topmargin -96pt
    dvipng is the answer.
        does not handle axodraw, though claims to handle ps specials (?)
        can use in --follow mode before latex finishes.
        use png16 instead
        identify equations and non-equations.  Use gs on non-equations.

latex pagetype:
    /usr/share/tex4ht/xhmlatex <file>
    with a config file:
        default font cmss-10 should be 'serif' or 'cmr10'
        encoding should be utf-8

    latexml:

dvips resolutions:
#        err, stdout, stderr = runCommand('dvips -D %f -R -mode pcscreen -o %s %s'%(res, psfileName, dvifileName))
        #err, stdout, stderr = runCommand('dvips -o %s %s'%(psfileName, dvifileName))
        # Select metafont mode
        # -mode pcscreen            (118 dpi)
        # -mode pcprevw             (118 dpi)
        # -mode pixpt               (one pixel per point) (72.27 dpi)
        # -mode mactrue             (72 dpi)
        # -mode ncd                 (95 dpi)
        # -mode macmag              (86 dpi)
        # -mode ibmvga              (110 dpi)
        # -mode declarge            (100 dpi)
        # -mode decsmall            (82 dpi)
        # -O -1in,-1in              (move offset to 0,0)

TODO:
  Resolution differences on navi/moya may be due to different paper sizes in
  dvips.  Also try dvips -D and specify resolution (affects font selection)
  rather than paper size.
  Also add -R (secure mode -- see man page)
  Chop off more of the dot horizontally?  1.5 is 1px too much at 16px charheight

    use cmr10 font.
    with centerline = dotendy (and pixel fudge), and res_fudge=1,
        14 perfect
        16 perfect
        18 1px high
        20 perfect

    x: 28px high
    cdot: 8px high
    centerline is 2px from bottom of cdot  (1/4 factor then?)
        12
        14 perfect
        16 
        18 1px high (x-height 10.1666666)
        20
    1/8 factor:
        12
        14
        16 
        18 perfect (x-height 8.6666666)
        20
    3/8 factor:
        12 perfect (x-height 7.25)
        14 perfect (x-height 6.75)
        16 1px low (x-height 7.50)
        18 perfect (x-height 9.1666666)
        20 perfect (x-height 9.3666666)
    5/16 factor:
        12  perfect (x-height 7.375)
        14  perfect (x-height 6.625)
        15  perfect (x-height 7.383333) (lower case smaller than cmr10)
        16  perfect (x-height 7.25)
        17  perfect (x-height 9.016666)
        18  perfect (x-height 9.416666)
        19  perfect (x-height 9.250000)
        20  perfect (x-height 9.116666) (cmr10 good)
        21  perfect (x-height 9.15)
        22  perfect (x-height 11.41666)
        23  1px low (x-height 11.78333)
        24  1px high(x-height 12.79166)
        25  1px low (x-height 12.14166)
        26  perfect (x-height 12.125)
        27  perfect (x-height 12.391667) (cmr10 tits)
        28  1px high(x-height 12.658333)
    --- new        centerline, dotcenter, dotpixmiss, newheight, res
        12 generates blank image for $xj$
        13 1px low   2.687500,  0.312500,   0.000000, 6.625000,   78
      * 14 perfect   2.687500,  0.312500,   0.000000, 6.625000,   84
        15 perfect   3.308333,  0.625000,  -0.066667, 7.383333,   90
            cmr tits, lower case letters different size
        16 perfect   3.375000,  0.625000,   0.000000, 7.250000,   96
        17 perfect   4.508333,  0.625000,   0.133333, 9.016667,   102
            cmr lower case top too wide
      * 18 perfect   4.708333,  0.625000,   0.333333, 9.416667,   108
            cmr upper case top too wide
        19 perfect   4.375000,  0.625000,   0.000000, 9.250000,   114
      * 20 perfect   4.441667,  0.625000,   0.066667, 9.116667,   120
        21 perfect   4.575000,  0.625000,   0.200000, 9.150000,   126
            cmr upper case top too wide, png upper case has tits on top
        22 perfect   5.708333,  0.625000,   0.333333, 11.416667,  132
            cmr lower case top too wide, png lower case has tits
        23 1px low   4.108333,  0.625000,  -0.266667, 11.783333,  138
        24 1px high  6.395833,  0.937500,   0.333333, 12.791667,  144
            cmr top too wide, png roman has tits
      * 25 1px low   4.929167,  0.937500,  -0.133333, 12.141667,  150
        26 perfect   6.062500,  0.937500,   0.000000, 12.125000,  156
            cmr lower case top too wide, lc roman has tits
        27 perfect   6.195833,  0.937500,   0.133333, 12.391667,  162
      * 28 1px high  6.329167,  0.937500,   0.266667, 12.658333,  168
        29 1px low   5.995833,  0.937500,  -0.066667, 14.008333,  174
            cmr upper case has tits, png lc has tits
        30 1px low   6.129167,  0.937500,   0.066667, 13.741667,  180
            cmr upper case top too wide, 
            0.97    png uc has tits
        31 perfect   7.195833,  0.937500,   0.133333, 14.391667,  186
            cmr top too wide, png lc has tits
      * 32 perfect   7.395833,  0.937500,   0.333333, 14.791667,  192

        15 perfect   3.308333,  0.625000,  -0.066667, 7.383333,   111
            cmr tops of letters fuzzy
        15 perfect   2.308333,  0.625000,  -0.066667, 7.383333,   108
            cmr tits, lower case letters different size
        15 1px low   1.687500,  0.312500,   0.000000, 8.625000,   104
            cmr tits, lower case letters different size
        15 1px low   2.108333,  0.625000,  -0.266667, 7.783333,   102



Startup time is 
    0.123u 0.016s 0:00.18 72.2%     0+0k 0+0io 0pf+0w

Time to get bbox for sample pdf generated by:
    http://mcelrath.org/Notes/Littlest%20Higgs%20Phenomenology
    time sudo -u zope gs -dSAFER -dDOINTERPOLATE -dTextAlphaBits=4 -dGraphicsAlphaBits=4 -dNOPAUSE -dBATCH -r130 -sDEVICE=bbox -sOutputFile=bob-%03d.png *.pdf
    0.427u 0.031s 0:00.52 86.5%     0+0k 0+0io 0pf+0w
    0.427u 0.023s 0:00.50 88.0%     0+0k 0+0io 0pf+0w
    0.423u 0.029s 0:00.49 89.7%     0+0k 0+0io 0pf+0w
    0.415u 0.036s 0:00.46 95.6%     0+0k 0+0io 0pf+0w
    0.422u 0.027s 0:00.49 89.7%     0+0k 0+0io 0pf+0w
Time to ghostscript the same thing:
    time sudo -u zope gs -dSAFER -dDOINTERPOLATE -dTextAlphaBits=4 -dGraphicsAlphaBits=4 -dNOPAUSE -dBATCH -r130 -sDEVICE=pnggray -sOutputFile=bob-%03d.png *.pdf
    2.456u 0.042s 0:02.52 98.8%     0+0k 0+0io 0pf+0w
    2.517u 0.036s 0:02.57 98.8%     0+0k 0+0io 0pf+0w
    2.478u 0.040s 0:02.55 98.4%     0+0k 0+0io 0pf+0w
    2.478u 0.039s 0:02.53 98.8%     0+0k 0+0io 0pf+0w
    2.461u 0.037s 0:02.53 98.4%     0+0k 0+0io 0pf+0w
time sudo -u zope convert -scale 1075x1520 *.pdf bobconvert-%03d.png 
    37.981u 0.439s 0:38.47 99.8%    0+0k 0+0io 0pf+0w
    eek!
time sudo -u zope mogrify -crop 204x13+204+692 bob-001.png bobnew-001.png
    0.296u 0.039s 0:00.38 84.2%     0+0k 0+0io 0pf+0w

time sudo -u zope mogrify -crop 0x0+2+2 bob-002.png
    0.731u 0.045s 0:00.82 93.9%     0+0k 0+0io 0pf+0w
    0.731u 0.040s 0:00.81 95.0%     0+0k 0+0io 0pf+0w
    0.756u 0.040s 0:00.84 94.0%     0+0k 0+0io 0pf+0w
    0.746u 0.039s 0:00.81 95.0%     0+0k 0+0io 0pf+0w
    0.734u 0.041s 0:00.81 95.0%     0+0k 0+0io 0pf+0w

time gs -dSAFER -dDOINTERPOLATE -dTextAlphaBits=4 -dGraphicsAlphaBits=4 -dNOPAUSE -r125 -sDEVICE=pnggray -sOutputFile=test-%03d.png test.pdf < gsscript
    5.665u 0.237s 0:06.60 89.2%  0+0k 0+0io 0pf+0w
    5.698u 0.225s 0:06.76 87.4%  0+0k 0+0io 0pf+0w
    5.696u 0.273s 0:06.92 86.1%  0+0k 0+0io 0pf+0w
time source gsscript2
    6.012u 0.317s 0:08.04 78.6%     0+0k 0+0io 0pf+0w
    5.961u 0.312s 0:06.99 89.6%     0+0k 0+0io 0pf+0w
    5.955u 0.311s 0:06.96 89.9%     0+0k 0+0io 0pf+0w



Latex runs in (almost) constant time no matter how many pages are in the file.
latex with existing LatexTemplate takes 0.2s.  
    Removing all included packages takes 0.06-0.07s.  
dvips -E takes 0.04-0.05s (indep of latex template).
gs -sDEVICE=bbox takes 0.145s.
eqn2img gives this command to ghostscript:
      fprintf(gs, "%i neg %i neg translate\n", xoffset, yoffset);
    can I get it to print the bbox without restarting?
    Bounding Box printed in postscript by default is not tight.
latex + dvips -E + gs bbox + gs + custom code = 0.25s + custom / eqn.
pdflatex + gs
(bbox) selectdevice
(test.pdf) run
1 showpage      (runs a particular page)


Alignment:
    on hawk: 
        lowercase letters are exactly 8 pixels high (a,e, not y) 
            Helvetica (Adobe) 14pt
        Top of upper case letters to bottom of 'y' is 16px.
    on draal:
        lowercase letters are 9px, upper case plus 'y' is 15px.
        (recopy TODO* from hawk)
    Perhaps make all things less than 16px to 16px.  Can measure latex height
    with a little latex:
        \cdot Wj
    Make this 16px scalable (match browser with latex)
    anything else gets expanded by 2.

    || or \left|\right| is getting second bar (first?) chopped sometimes.

    1px too much is removed in 'e'.

Documentation:
    Fix README: -imagemagick +pil
    Note about standard_header CSS template.
    Change resolutions?

1/22/2004
    Rendering comments:
        Stx is inserting <p>...</p> around every line.  When the whole page is
        rendered it does not handle the page and comments separately.  Comment
        headings should not be treated as Stx, and message bodies should only
        be treated as Stx if the PageType class calls for it.

1/31/2004
    More timing tests.  I've switched to pslatex so I can use axodraw (requires
    postscript specials so doesn't work with pdflatex)  dvips can spit out an
    eps with a tight bbox for each page, which could speed things up significantly.
    Timing tests on draal (600MHz alpha 21164), 512MB RAM)
        pslatex/dvipdf: (looks crappy)
            6.255u 0.469s 0:08.25 81.3%     0+0k 0+0io 0pf+0w
        pslatex/dvips:  (looks best)
            5.503u 0.385s 0:07.06 83.2%     0+0k 0+0io 0pf+0w
        pdflatex:       (looks better)
            5.696u 0.359s 0:07.12 84.8%     0+0k 0+0io 0pf+0w
    appears to be only 10pt and 12pt available in latex
    -dAlignToPixels=1
        affects neither ps or pdf mode.

2/2/2004
    code='$\mathrm{Y}$' topsize=7.866667, bottomsize=4.133333, centerline = 7.866667, ne
    wheight=15.733333, widentop=0.000000, widenbottom=0.000000, chopx=2.000000
    code='$\mathrm{j}$' topsize=6.866667, bottomsize=7.133333, centerline = 6.866667, ne
    wheight=14.266667, widentop=0.266667, widenbottom=0.000000, chopx=2.000000

    code='$m$' topsize=3.866667, bottomsize=4.133333, centerline = 3.866667, newheight=8
    .266667, widentop=0.266667, widenbottom=0.000000, chopx=2.000000
    code='$G$' topsize=7.866667, bottomsize=4.133333, centerline = 7.866667, newheight=1
    5.733333, widentop=0.000000, widenbottom=0.000000, chopx=2.000000


    16px    centerline + 0.0666666
    17px    centerline + 0.1333333
    18px    centerline + 0

2/3/2004
    gs does not produce pnggray files that are the correct resolution for an 8.5x11 page.
    scale factor is 1.063 (DPI larger than requested)
    Is it only imaging a sub-region of the page?
    produces the proper resolution:
        gs-gnu 7.05 daneel 
        /usr/local/bin/gs 8.11 on moya
        gs-aladdin 7.04 on moya
        gs-gnu 7.07 on moya
        gs-esp 7.07 on moya

    Size/Align/Res combinations that look good:
        18  0     0.97  Times New Roman
        17  0     1.03  
        16  0.5   0.97  

StructuredTextLatex changes:
    remove ZWikiPage.applyWikiLinkLineEscapesIn (not needed)
    remove ZWikiPage.protectLine
    remove ZWikiPage.links (not used)           (note dtml-accessable but not used)
    remove ZWikiPage.canonicalLinks (not used)  (note dtml-accessable but not used)
    wikilink() called from DTML should really call ZWikiPage.renderLink

6/4/2004:
    How to do mathml with tex4ht:
        1) include:
            \usepackage[xhtml,mathml,no-DOCTYPE]{tex4ht}
           in document preamble
        2) latex $1.tex
           tex4ht -f/$1.tex -i/usr/share/texmf/tex4ht/ht-fonts/ -cunihtf
           t4ht -f/$1.tex
        3) 
            import sys
            from xml.dom.ext.reader import Sax2
            from xml.dom.ext import Print

            reader = Sax2.Reader()
            document = reader.fromStream(sys.stdin)
            for math in document.getElementsByTagName('math'):
                #... math is the MathML, in order it appeared in the document.

        Can use one run of latex if we only look at the last N pages of the dvi
        file:
            dvips -p 228

    times on navi:
    dvips -e 0 may help with characters moving about...
    latex (w/tex4ht usepackage)                                 0.94s
           w/o tex4ht usepackage)                               0.25s
    dvicopy -page-start 228 -max-pages=1 blah.dvi blah2.dvi     0.001s
    dvips -E blah2.dvi blah2.ps                                 0.06s
    gs ... (creates a full-page png :()                         0.2s
        gs creating all images at once: (average of 229p)       0.0947s
    convert blah2.ps blah2.png                                  0.136s

    The altimg tag may be parsable by XSLT into an img tag.
    Konqueror is pickier about XML than Mozilla -- use it to check
MathML rendering errors:
    mi { margin: -0.15em; } does not get applied to mi's in super/subscript.
    spacing if <mi>sin</mi> sin is inconsistent -- i-n has more space than s-i.
6/24/2004:
    Plone interaction:
        got plone quick installer working

        Plone uses a <base href="http://localhost/Plone"> tag which gives the
        WRONG page for the one you're looking at, and makes <a name> links
        fail.

        restore ALLOWED_PAGE_TYPES_IN_PLONE?  need to remove wwml for plone...

        Plone skin w/header is CMFPlone/skins/plone_templates/header.pt