Codemirror supports bidi well enough, but all lines are left-aligned.
Want gedit-like automatic paragraph direction. (for starters, per-line direction would be good.)

Every keypress in first line creates new entry in browser history, because the URL fragment changes.

Should somehow delay/debounce it. Note that simple constraining it to cursor out of first line won't work in co-author's browsers which similarly track the first line but have cursor elsewhere.

Mar 5, 2014
\newcommand{\Z}{\mathbb{Z}} | mathdown
mathdown.net

Mar 4, 2014
\newcommand{\Z}{\mathbb{Z}} | mathdown
mathdown.net

Mar 4, 2014
\newcommand{\Z}{\mathbb{Z} | mathdown
mathdown.net

Mar 4, 2014
\newcommand{\Z}{\mathbb | mathdown
mathdown.net

Mar 4, 2014
\newcommand{\Z}{\mathb | mathdown
mathdown.net

Mar 4, 2014
\newcommand{\Z}{\math | mathdown
mathdown.net

Mar 4, 2014
\newcommand{\Z}{\mat | mathdown
mathdown.net

Mar 4, 2014
\newcommand{\Z}{\ma | mathdown
mathdown.net

Mar 4, 2014
\newcommand{\Z | mathdown
mathdown.net

Mar 4, 2014
\newcommand{ | mathdown
mathdown.net

Mar 4, 2014
\newcomband | mathdown
mathdown.net

Mar 4, 2014
\newcomand | mathdown
mathdown.net

Mar 4, 2014
\newcoman | mathdown
mathdown.net

Mar 4, 2014
\newcoman | mathdown
mathdown.net

Mar 4, 2014
\newcom | mathdown
mathdown.net

Mar 4, 2014
\newco | mathdown
mathdown.net

Mar 4, 2014
\newc | mathdown
mathdown.net

Mar 4, 2014
\new | mathdown
mathdown.net

Mar 4, 2014
\ne | mathdown
mathdown.net

Mar 4, 2014
\ned | mathdown
mathdown.net

Mar 4, 2014
\n | mathdown
mathdown.net

Mar 4, 2014
| mathdown
mathdown.net

Decode #foo in URL and scroll to # Foo heading inside document.
(Doable with pure id="foo" on line DOM? Unlikely if CodeMirror handles scrolling.)

Make # in headings clickable and result in such permalink.

Provide quick summary of supported syntax.
For now would suffice to link to http://cben.github.io/mathdown/?doc=help

• Either use Docverter or host pandoc myself on heroku.
• One-click button to then import into writelatex and/or sharelatex.

Far away:

• Markdown lint that warns about suspicious markdown (e.g. list not preceded by newline).

• A way to even open cached mathdown when you're offline.
  Look into app manifest / chrome extension / firefox OS app?
• Cache last seen content of every document in localStorage.
  Bonus if having any mathdown open does background updating on all known docs.
• Be able to edit offline and sync on reconnect.
• index.html?doc=... prevents caching of index.html for new documents.
  history.pushState() might help, assuming you have one tab open.

When it happens, usually affects all formulas.
Not permanent, reload of same document usually fixes it.

Error message is something about 'VARIANT'.
(Perhaps mathjax/MathJax#363 ? But that was fixed in v2.2.)

TODO: try mathjax/unpacked/ to debug.

• show wrapped lines with same indent as current line.
  See http://codemirror.net/demo/indentwrap.html and https://groups.google.com/d/msg/codemirror/oag8nhmJX20/u1QluJgtjkwJ
• use mode's smart .indent()
  • markdown.js doesn't implement any smart indent.

This might not be feasible until I make indents monospace (#8) because the known approach tweaks line CSS from within renderLine event, where we can't measure pixels, only count chars.

Since a document may be edited by other users (central example: about, help!), formulas should not be allowed to create javascript: links.

http://docs.mathjax.org/en/latest/safe-mode.html

In the future (#9), makes sense to similarly block [foo](javascript:evil) markdown links.

Currently mathdown doesn't work offline.
It seems that firepad would initialize and allow pasting text if only firepad.js was available. But it's proprietatry so not sure I can (or want) to put firepad.js in my git. And I'm not keen on starting to use npm/bower and losing the conveniece of Just a Bunch of Files...

Many things (including math) perform better if CM does scrolling and doesn't have to create DOM for invisible parts of doc. Blockers:

• #1 Printing
• Use CM search UI because browser search will be incomplete (also helps with cben/CodeMirror-MathJax#11).

Almost not an issue now but blocker for #9:
Since the URL is secret, directly linking to external sites would expose the URL in Referer: header.

Should probably use an extra redirect. Is there a way to keep the link structure search-engine friendly? Links on public pages should count as links. OTOH, if they're publicly editable they probably shouldn't to deter spam?

Alternitive: I'm again tempted to keep secret portion in #fragment, which should not (though sometimes did) leak via Referer.

Yesterday I pushed ad146e1 which updated CodeMirror but not Firepad, resulting in a combination where firepad doesn't sync any changes ("pos exceeds the bounds of the AnnotationList").
I knew all about it and even prepared a correct update on a new-deps branch but forgot about it, and only tested the CM commit in my dirty checkout which also had new firepad... FAIL! I should fire myself.

But my main point here is that my laughable smoke test didn't catch that I completely broke mathdown (pages did load but edits weren't saved! That's even worse that being visibly down!)

I really need a functional test opening same page from 2 instances and seeing edits propogate.
Should check if I can reuse existing firepad tests...

Also, JS console errors should be blockers. See https://code.google.com/p/selenium/issues/detail?id=148

Search engines, or Javascript-less frontends (links, curl) etc. should be getting a at least a plain-text version of the document content (in a text area). Bonus points for:

• rendering links as links.
• full server-side CodeMirror styling.

Either way needs node.js on the server.

E.g.

$$\pi$$

works but

$$
\pi
$$

does not.

For any longer formula, line breaking comes in handy.

This probably means:

• a way for user to supplies a bibtex file (global file per user? research Zotero / Mendeley / citeulike etc integration...)
• autocomplete of citation keys
• tooltip showing full citation, behave like a clickable link (#9)

http://johnmacfarlane.net/pandoc/README.html#footnotes

• Highlight footnotes [^1] as link.
• Don't hightlight First word in [^1]: First rest... differently.
• Don't highlight continuation lines as code.
• Otherwise feature parity with links #9

• mode to show one section at a time
  • readonly vs edit modes?
• beamer export
• http://slid.es/ or similar export? Better yet, edit markdown here, use reveal.js or similar for beautiful view?

This is about editing math-heavy text in fully-initialized document; see #19 for init performance.

• cache typeset DOM, reuse if text & token class didn't change
• if changed, use jax methods like .Text(newmath)
  http://docs.mathjax.org/en/latest/typeset.html#manipulating-individual-math-elements
  • use .Rerender() if only token class changed.

Could Hub.Update() do the jib instead a custom caching scheme?!
http://docs.mathjax.org/en/latest/api/hub.html#Update

• pick "supported" table format(s)
• use monospaced font in tables
• align help, a-la Org mode

http://mangomarkdown.com/ does this. I'm not quite convinced it's a good idea — it becomes too close to a WYSIWYG editor (with an obscure way to format things) to my taste.

But here is a very simple idea for implementation that might just work: enable styleActiveLine option, use markdown mode's option to style formatting chars, hide them with CSS except for current line!

The initial render on document load is horribly unacceptably slow (saw 1 minute on a math-heavy page).
The unrendered text does show very fast (#15) but it's not very useful since the editor is paralyzed until all math is rendered.

currently Tab inserts a hard tab.

MathJax does all typesetting in-order (TODO: could be because I'm queueing one by one, check with one Typeset call on long string).
Which means any extra extension to be loaded delays the whole process.

We could save the list of mathjax extensions actually needed for a given document in Firebase, and pre-load them all at start. Could then give up the -full config.

The basic need is straightforward, the question is what kind of login do I want (cf #41 for the other major use case for logging-in users).

• Simplest: login required to remember visited documents.
  Should start with this, easiest to implement.
• Better (a-la writeLaTeX): when anonymous, still remember documents in localStorage (or firebase with anon auth cached in cookies/localStorage?)
• Fancy partial login: first just ask for email, send secret URL for accessing your doc list.
  This effectively validates email, but asynchronously — link only needed for 2nd login.
  If link is lost, can resend it — similar to password reset.
  After using secret link, can optionally set a password?

Also, should allow search.
Firepad indexing is hard, but something as dumb as loading ALL docs into the doc list page for browser search to work might be acceptable if sufficiently lightweight (just <pre> with no highlighting, load snapshot instead of firepad reconstruction...)

Should probably expose doc title in firebase, without reconstructing firepad text.

I created a new document, which was surprisingly empty.
Typed several lines.
Then saw them disappear and the default # Untitled ... text replaced them :-(
Luckily, Ctrl+Z twice brought my text back.

Anat also said it has happened to her.

Probably a race condition, may be related to slow loading (cold cache + bittorrent running).

> quote
# Header 0123456789

Renderend now as a header inside a continued quote.

• That's a CM bug. Should be considered outside quote:
  http://johnmacfarlane.net/babelmark2/?text=%3E+lazy%0Aquote%0A%23+Header%3F
  Or, if requiring empty line before header, render as continued quote but not highlighted as header.

(The non-portability here is one of these things where having a "markdown lint" would help.)

Made worse by 2 bugs => only Header is colored green as a quote.

• The # is bright-bluish - wrong CSS cascading?
• 0123456789 is blue, somehow due to GFM styling it as github SHA link and my kludge unstyling it.

Should be

But here things are

Find and recommend some way to write theorems, lemmas etc. in MD and get them converted to \begin{theorem}... in latex.

Pandoc supposedly added arbitrary <div> and <span> but I haven't been able to reproduce yet — probably need a newer version.

Github pages doesn't serve HTTPS with a valid certificate.

• register mathdown.{com,net}
• get a cert (StartSSL offer free ones, and cheap recursive ones — which might come in handy for staging new versions / forks etc.)
• find free/cheap hosting with easy deploy from github.

P.S. Cloud9 workspaces like https://c9.io/cben/codemirror-mathjax/workspace/demo.html support HTML. Great for editing / sending others a demo, not good enough for the main site.

It's tempting to use #xyz (as in Firebase demo) instead of ?doc=xyz,
so that secret document ids are not sent in the plain (they'd only be sent to Firebase over https).
But this doesn't really make HTTP hosting OK — a man in the middle can easily inject malicious JS that would leak the document id.
A stronger point is that ?doc=xyz would be stored on server's log, even with HTTPS. Not sure it's evil, collecting URL-level statistics is standard and useful.

What's more important, I want ?doc=xyz to preserve disting documents having distinct URLs from search engine POV. I'd use /xyz but then I'd need .htaccess or similar.
And I want to use #heading for linking to # Heading inside doc (#5).

Per-user undo seems to work nicely (thanks to firepad).

But once you refresh the page you can't undo older changes, probably because you got a new random user id.

• When I implement any kind of login, make sure to assign a fixed firepad user id.
• Sticky user ID in a cookie. (Is there any anonymity/tracking concern?).
  This is useful even after login is implemented

Additionally, look into other ways to expose the firepad history (e.g. etherpad-like slider?). It's all there, just no way to access it.

gfm.js supports it, but only for modes that have been loaded.
(gfm ... actually works ;-)

I don't want to pre-load all modes up front, so need to integrate loadmode addon.

It seems someone wants AsciiMath support, as evidenced by this attempt on the front page:

Suppose $$ax^2+b x+c=0$$ and `a!=0`. We first divide by `a` to get `x^2+b/a x+c/a=0`. 

Then we complete the square and obtain `x^2+b/a x+(b/(2a))^2-(b/(2a))^2+c/a=0`. 
The first three terms factor to give `(x+b/(2a))^2=(b^2)/(4a^2)-c/a`.
Now we take square roots on both sides and get `x+b/(2a)=+-sqrt((b^2)/(4a^2)-c/a)`.

Finally we subtract `$$b/(2a)$$d` from both sides and simplify to get the two solutions: 
`x_(1,2)=(-b+-sqrt(b^2 - 4a c))/(2a)`

1. I could easily enable AsciiMathML in mathjax, but not with backtick syntax - it's already taken by literals in markdown.
2. Markdown with AsciiMathML wouldn't be very useful; the only converter I know that understands it was MultiMarkdown 2, replaced by LaTeX math since version 3.
   This does suggest the syntax should be MMD2-compatible: << e^(i pi) + 1 = 0>>.

I see how AsciiMath is attractive for easier input (e.g. a/b compared to \frac{a}{b}) and some degree of Do-What-I-Mean magic for beginners.
In the long run, I believe LaTeX math is the right syntax for stored documents, for power and interoperability.
The benefits of AsciiMath could be realized if you could type a/b and Mathdown would suggest to replace it with $\frac{a}{b}$ (cf. #36). But these are very far-off ideas, whereas enabling AsciiMath is doable now.

Hard to reproduce — only after a lot of writing, when firepad tries to checkpoint.

/checkpoint writes are conditioned on "author of checkpoint is the same as the author of the revision they're checkpointing."

A couple times seen with these network errors:

Failed to load resource https://s-dal5-nss-8.firebaseio.com/.lp?start=t&ser=14891542&cb=547&v=5&ns=mathdown
Failed to load resource https://mathdown.firebaseio.com/.lp?start=t&ser=19417576&cb=548&v=5

Today seen with "undefined has no .from property." exceptions.

Using similar tech to what CodeMirror-MathJax does.

CM prevents browser's spell check.

Needed for comments, chat, threaded responses, everything.

An obvious direction is Etherpad-like magic coloring of all text by author. But I'd like to steer cleer of magic text properties and keep the document model be plain text [until proven that it doesn't work].
And there are really 2 scenarious:

1. Collaborative writing, where the end result should be unattributed (though a special history/blame mode is occassionally useful). In Etherpad, you nuke all coloring when it gets annoying.
2. Discussion (including comments on writing), where who wrote what should be part of the text.

Ideas:

• <user> text attributed quotation a-la IRC logs.
• @username and/or email@domain. I'm leaning towards email with compact rendering as a gravatar and autocomplete for fast typing.
  • Potential conflict between @user notification ("to") semantics and attribution ("from") semantics I need here.
  • Conflict between @ and pandoc citations.
  • Email makes more portable documents. [Unless I adopt 1:1 login-with-twitter usernames.]

Getting lists right is tricky for markdown newcomers.

• Enter should probably continue list item; second Enter would start new item; third Enter would close list.
• List items should auto-indent so text (not bullet) is aligned to multiple of 4 spaces, so continuation lines can line up nicely. Reindent when typing - ("electric"), on Enter, or both?
• Tab/Shift+Tab when notheng is selected should affect whole paragraph / list item (incl. sub-items) etc.

Steal ideas from Stack Exchange?

http://meta.stackexchange.com/a/56430/204832
Actually Shift+Enter is a continue formatting command: I.e. it doesn't continue extra indenting within code, but it does add the first 4 needed spaces if you're already in code. It also adds the next numeric value for a numbered list, a copy of the first mark for an unnumbered list, and for both of the latter it "fixes" the other numbers/marks to be sequential/the same. It also continues > blockquotes. – Mark Hurd Jul 10 '10 at 3:16

Only consider doing all that on unshifted Enter?

• Unseen activity indication in favicon. This is easy.
• Some kind of transient highlighting of what changed since you last looked? This is hard — when exactly have you seen something? Track your viewport? Cursor motion perhaps is good middle ground (but then if 2 people are doing append-only chat, are their messages never considered read)?

This may overlap with #41 but I think I do want in-text names while side-channel magic for unread activity is more acceptable as a projection of the history of the text?

Firepad uses markText with className to color selections by other users.
Turns out widgets don't get any classes applied, even when fully covered by a marker. Marijn says it's intentional: https://groups.google.com/d/msg/codemirror/L7VPuMrul78/l_iicpP97N4J

So should apply selection styles automatically to math widgets.
Should (1) apply existing selection classes to new widgets, (2) apply new selections to existing widgets.
Both should probably use recently added markerAdded, markerCleared events.
The place to do it is probably firepad, but with these events also can do it myself.
Or maybe CM addon?

Initializing Firepad takes over 40s for long documents!
This is not result of MathJax, slow even if disabled.
It seems that Firepad fails to create snapshots due to security rules. Perhaps I must use simple login even for anonymous users?

The Firebase backend is proprietary so you can't run your own instance of the full stack.
At this stage, I would not trust myself with running a secure & reliable backend, so using Firebase is better for Mathdown end-users.

OTOH, there is little need to change anything about Firebase and being able to develop as a purely client-side code makes forking and testing much easier - a user can try any fork without exporting/importing user data.
"Running" a fork is as simple as loading the static HTML from anywhere - local checkout, github pages, cloud9 workspace etc.

Any free solution should preserve the ability to fork and use without migrating data.

Best free-backend alternatives to Firepad:

• https://github.com/Operational-Transformation/ot.js/ is mature, well documented, has CodeMirror binding.
  Includes node.js server code; no persistence layer?!
• https://github.com/share/ShareJS is simple and has (currently incomplete - josephg/ShareJS#124)) CM binding.
  Includes node.js server code, needs MongoDB.

For other firebase-like functionality (if we add any), meteor.com looks most similar in spirit, and even provides free hosting for server-side part of the app (arbitrary node.js?).

Shows a scrollbar instead of full text.

Should offer autocomplete when typing \foo.

• Autocomplete dollars when outside math: \al -> $\alpha$ (with cursor before closing $ if possible).
• Bag-of-words search & spelling, ideally \arow should offer \rightarrow etc.
• Display symbols in completions list.

Big question: should I autocomplete to unicode chars (e.g. \beta -> β) to encourage Mathdown's "the source is the document" ethos? It's seems to be supported well enough by MathJax(?). Should limit to only chars that work with \usepackage[utf8]{inputenc} on pdflatex. Anyway there must be an option to export back to ascii.

• Integrate (or just link to) Detexify? Or for unicode, http://shapecatcher.com/. Keep an eye on Detexify's client-side successor https://github.com/kirel/sketch-a-char (not yet trained).

# С Днем Рождения! => http://mathdown.net/?doc=IuNRlPn7J7U#

Do all/most modern browsers display human-friendly UTF8 in URLs, specifically in fragment?
What id do pandoc and github generate from unicode headings?

http://webdevchecklist.com/