Discount

Older versions of Discount

• version 2.0.9 puts in a workaround for a Linux security misfeature; some of the commercial hosting services set up /tmp so that it’s mounted noexec, which means that when configure.sh attempts to run a little test program to find scalar sizes, it can’t do it and ends up failing. So configure.sh has been patched to put temporary executables that need to be executed into the working directory instead.

• version 2.0.8 introduces a small collection of patches, bugfixes, and performance enhancements:

  • Table processing was modified to make it hew closer to the markdown extra implementation
  • Optimized the code that’s used to look for tables by scanning for pipe symbols when I’m reading data into the input lines array.
  • Documentation patches from David Banks
  • argument prototype tweaks for Solaris, from Allen Otis
  • --with-github-tags configuration option to to include - and _ in the acceptable characters for maybe_tag_or_link() (code from Ryan Tomayko’s discount SCCS on github.)
  • in the markdown program, -f? gives a list of the known flags.
  • also, -F? gives a list of the known flag bits.
  • --with-id-anchor configuration options to have the table of contents code emit id= instead of the dummy <a name= anchor I replaced it with (name= is depreciated in xhtml, and may vanish.)

• version 2.0.7 finishes support for php markdown extra-style [footnotes] by documenting them, fixing one small initialization bug, and by creating the new public function mkd_ref_prefix() which changes the name= part of the forest of tags from fn:N and fnref:N to {your prefix}:N and {your prefix}ref:N (allows someone to footnote individual articles on a weblog where each article is rendered individually.)

  A few manpage typos are also corrected thanks to a patch from Alessandro Ghedini.

• version 2.0.6 adds preliminary (and undocumented, because I’m still peering at the published interface and internal data structures) support for php markdown extra-style [footnotes] (disabled by default; you enable them with the flag MKD_EXTRA_FOOTNOTES.)

• version 2.0.5 cleans up a long-standing compile-time warning, lightly tweaks the table of contents generator (-ftoc) to use <a name= anchors instead of <hXid= anchors, tweaks the librarian.sh generator so it doesn’t require symbolic links to generate static libraries, and expands the characters allowed in css class names to include dashes and underscores.

• version 2.0.4 corrects a few bugs, adds some chrome, and adds a few features.

  1. add shared library support (--shared in configure.sh; FreeBSD, Linux, and MacOSX only for now.)
  2. correct some typos
  3. Add manpages for the example programs makepage(1) and mkd2html(1)
  4. Initialize the random number generator and tag list separately, just in case someone really wants to do multiple initialize/deallocate cycles.
  5. Add a library destructor function (mkd_shlib_destructor) to discard the kw array before unloading a shared library.
  6. Put in a bunch of useless code to make gcc STFU.
  7. Be more picky about what comes before a ^ if we’re superscripting
  8. Unclosed html/style blocks revert to being regular old markup. True, the markup will contain the opening tag (and most of the other html from that block) but it will be regular old markup with additional garbage added in.
  9. Null-terminate the buffers generated by mkd_line(), mkd_css() and mkd_toc().
  10. Flip the order of arguments to mkd_generatecss(); should be 1,size, not size,1.
  11. It turns out that non-toplevel paragraphs consume adjacent code blocks :-(
  12. Rework (again) indentation inside list items. Indent should never be > 4 for enumerated lists; this means that I have to test and cut back large indents and change my test cases to account for leading whitespace on deeply (2 spaces + enumerated marker + 1 space) indented items.
  13. Add a raft of testcases to cover the new features + bugfixes.
  14. Add markdown extra style definition list support
      • Add a configuration flag (--with-dl={discount|extra|both} to select which way to do definition lists. Put a DL= into version.c so the user can see how this is configured.
      • If dl is turned off at compile time, mark it off in version.c

• version 2.0.3 updates the woefully out of date documentation describing the various MKD_ flags available in the published interface. As a little bit of chrome, it also modified mkd_document() to put a trailing null at the end of the buffer containing the generated document.

• version 2.0.2 fixes a couple of defects, adds more testscripts, and adds and updates a couple of features:

  1. Fix a glitch in new-style [] processing that would collapse two sequential new-style [link]s into one no matter how much space was between them. The reference dingus only coalesces adjacent new-style [link]s together if they are physically adjacent or have one space between them.
  2. Modify superscript grabbing so that it grabs parenthetical and alphanumeric blocks
  3. Rework (again!) the guts of linkylinky to support markdown 1.0.2b8-style <..> links: these are like the original format I implemented before I rolled them out while chasing compatability in that a url starting with ‘<’ contains all characters until an unescaped ‘>’ is encountered.
  4. Change the autolink prefix to closer fit to the reference dingus; if a <url> has a known protocol prefix, it does not need a leading // after the prefix.
  5. Add a couple of functions that write the current runtime configuration (either a flags value (mkd_flags_are) or of an MMIOT (mkd_mmiot_flags)) to a FILE. markdown -VV prints out the flags for now, and the functions are there for other people to use as they desire (see below.)
  6. In theme, add the pseudo-tag <?theme config?>, which writes a listing of all of the runtime settings to the output file. If <?theme config?> is called from the header section of page.theme, it writes the runtime settings in markdown -VV format, otherwise it writes a two-column html table (regrettably this will not work from inside generated content; theme is kind of stupid that way.)
  7. Add test cases for <> links (regular and markdown 1 compat), split the superscript tests out from smarty.t

• version 2.0.1 fixes a newly introduced bug in _ handling (I flaked out and reversed the test for relaxed emphasis, but hadn’t updated the testcase to account for the new runtime configuration for relaxed emphasis,) plus adds a pair of test cases for table of contents support.

• version 2.0 fixes one bug in table-of-contents handling (it would dump core if a header item contained a link), fixes the automated labeller so that it creates valid xhtml id attributes, and (by the simple expedient of breaking the published interface by expanding the flags field from an int to an at least 32 bit scalar (to fit the 22-and-counting option flags that discount now has)) restructures the code so that the only things that are compile-time options are memory allocation debugging and proper tab settings.

  It’s a .0 release, of course, so there may be some horrible bugs just waiting to fall out in the places I’ve not written test cases for yet, but there are many advantages to having the vast majority of the features be settable at runtime and I decided that this was a good time to make the change.

• version 1.6.8 finishes the backtick handler cleanup started in version 1.6.7, so that it works like the reference implementation does, and adds in strikethrough support.

  The strikethrough support is a common extension format; you enclose the text you wish to strikethrough with string of (two or more) tildes, like ~~this~~. The syntax for strikethrough is, with the exception of needing two or more tildes, identical to the syntax for backticks; you need to have at least as many opening tildes as you do closing (extra opening tildes are put inside the struck-through code, so ~~~orc~~ becomes ~orc.) If you don’t want strikethrough, the flags MKD_STRICT or MKD_NOSTRIKETHROUGH disable it (I am shifting away from using configure-time options to enable/disable extensions; If I compile them in all the time, it makes it easier to do regression and crash testing inside my automated test suites.)

  In addition, 1.6.8 cleans up the html generated by the ->…<- extension; instead of creating <center>…</center> (not allowed in xhtml, alas) it creates <p style="text-align:center;">…</p>, which appears to work when centering images and text.

  In addition in addition, I’ve added the <object> tag to the list of html blocks that discount will not attempt to process, and I’ve attempted to update the manual pages to reflect some of the recent additions to the codebase.

• version 1.6.7 repairs one defect in backtick handling, where if a code span was closed by more backticks than it was opened with discount would consume the starting characters within the code span ( so ``foo``` would become \oo\ instead of \foo\`.) Ooops. This was a simple case of mishandling the tick matching so that if the # of opening ticks was different from the # of closing ticks I would set the # of ticks to consume to the # of closing ticks no matter what. 1.6.7 corrects this feature.

• version 1.6.6 repairs two defects, one in the markdown compiler and one in theme:

  1. In theme, I needed to take into account the source filename might not have an extension when I’m making the .html filename. The old behavior was to look for a dot and put the .html after that, but I didn’t check to see if there was actually a dot there before appending the .html!

     This did not work out too well if there was no dot.

  2. In the markdown parser, I wasn’t handling escapes of the open square bracket inside a []() construct. So a link like [foo\[and\]bar](does not work properly!) would not parse because my square bracket matcher would look for an additional ] to match the \[ inside the [] part.

• version 1.6.5 repairs six defects and adds two new features.

  The bugfixes are:

  1. properly handle html comments w/o spaces
  2. fix nested list handling with unindented list items.
  3. Add config.h to the dependency lists for cols & echo, so those functions will properly rebuild after a reconfig.
  4. More plan9 fixes from Josh Wood; this one adds the flag _C99_SNPRINTF_EXTENSION to tell the APE cc that, yes, I do really want to be using snprintf() and I know that C99 has it return the size it wants to populate.
  5. don’t describe <style> blocks as mystery node during markdown -d

  6. There is one race condition if you are using discount in a threaded environment; the code that initializes the blocktag list (for html blocks) has an internal test then set variable, but it’s not atomic, and it’s possible to either (a) have two threads go into that initializer at the same time and both attempt to initialize, or (b) have a second thread attempt to mark down some code while the first thread is still initializing. Either way it would produce hours of debugging fun.

     To correct this defect, I’ve created the new function mkd_initialize(), which does all of the initialization. This function is called from within the markdown parser (so existing code will still work) but if you’re doing markdown from a threaded environment, you must call mkd_initialize() before you start spinning off threads.

  And the features:

  1. add a VALID_DOCUMENT magic flag to the MMIOT structure, so that if someone attempts to double-free it, the second attempt will just not work instead of causing a core dump.
  2. restructure the blocktag[] array into a STRING(), so users can add additional block tags as they desire.

  3. in relation to this, add the library function mkd_add_html5_tags(), which adds (globally, and non-removably) a handful of new tags for html5 support.

     There are actually two new functions here – mkd_add_html5_tags() is a additional library function that will not be linked in unless you demand it (or build the library as a dynamic library), and mkd_define_tag(char * tag, int selfclose) is a function that allows you to add html tags as you wish (so you could, if you wished, so something like mkd_define_tag("html", 0) to treat the contents of a html document as html) and which is used by mkd_initialize() to define the standard markdown html block tags, mkd_add_html5_tags() to add the narsty new html5 tags, or any user-written function to add the tags you want.

     It’s not threadsafe, so you need to define your tags before spinning off threads.

• version 1.6.4 repairs a single defect (in the handling of html blocks), adds a pair of features (both via github,) and adds a couple of test cases to further excercise the code block handler changes from version 1.6.3.

  • The defect was that text following the closing tag of a html block would cause following content to be misprocessed (besides, of course, the existing defect that that text would not be processed.) The bugfix is that I now split following content into a new logical line, so input like

    <p>blah blah</p>foo
    

    which used to output

    <p>blah blah</p>foo
    

    now outputs

    <p>blah blah blah</p>
    <p>foo</p>
    

  • The (creeping) features are

    1. Ryan Tomayko@github modified rdiscount to do smartypants with 're, 'll, 've', 'm', and 'd so I sucked those changes into the baseline code.
    2. earthboundkid@github suggested the new pseudo-protocol lang:<whatever> which would expand into <span lang=<whatever>…</span>

  • The additional testcases are to catch mishandled backslashes in code sections (most recently documented at http://code.reddit.com/ticket/695.) This defect was corrected in version 1.6.3, and I testcased it then, but more tests are always good.

• version 1.6.3 repairs code section handling by the simple expedient of ripping most of the existing handlers out and replacing them with newer (and more dingus-compliant) code.

• version 1.6.2 cleans up a couple of defects, adds manpages for the new callback functions introduced in version 1.6.0, and rewrites the emphasis generator (again) to make it work in a more xhtml-compliant manner when fed pathological emphasis.

  The bugfixes are:

  1. fix a bug in autolink that made a hanging prefix generate a bogus <a tag.
  2. don’t emphasise _ or * when they’re in the middle of whitespace.
  3. unescape trailing ‘ ’s when they’re in the middle of a code section or []() link.
  4. restrict the depth of a ETX header to <h6>
  5. tighten the parser for footnote links so that a construct like []-<]: won’t erroneously become a footnote.

• version 1.6.1 fixes two scripting exploits, corrects three edge cases to make them work more like the reference implementation, and lightly modifies list handling to make it work a little more like the reference implementation (but, hopefully, without triggering the awful failure case that the reference implementation has.)

  The two security holes that were fixed were:

  • if explicit html was forbidden, but someone tried to put in a tag anyways, I would replace the opening tag with a < then just copy the rest of the tag to the output until I reached a ‘>’. So if someone did <hi<script>attack<hi</script>, the <script> would be dumped to the output just as prettily as you please.
  • I was being overly greedy about escaping (with \) " and ', so a malformed title string could contain arbitrary text, up to and including scripts.

  The three compatability changes are:

  • lines at the end of a paragraph never get <br/>‘ed,
  • blockquotes can be indented up to 3 spaces,
  • SETEXT-style header lines can have trailing spaces,

  And the tweak is to change how list items absorb new paragraphs. Discount traditionally would absorb new paragraphs if they were indented one level (4 space), and punt them back to the parent level if they weren’t indented that much. The reference implementation, on the other hand, captures new paragraphs into the preceding list if, apparently, they are indented at all. The people at github, who, apparently, are using Discount, have a bug report against list handling where

   * list indented 3 spaces
   * list indented 3 spaces
  
     paragraph indented 3 spaces
  

  doesn’t absorb “paragraph …” into the second list item. I have modified the indent handler to absorb items indented as deeply as the text on the preceding list item instead of requiring a 4 space indent. It doesn’t appear to break any of my test cases (markdowntest 1.0, markdowntest 1.0.3, and any of my flock of tests), and it allows for slightly more readable lists.

• version 1.6.0 adds one new feature that’s large enough to require a version bump; I’ve now added callback routines so that a program that uses discount can

  1. modify the urls found in []() and <link> constructs, as well as
  2. supply extra arguments to the tags generated by []() and <link> constructs.

  As an example, if you wanted to ensure that no urls were relative, you could set up a basename callback to add a prefix to every relative url (this was hardwired into discount prior to 1.6.0, but has been newly made into a callback.)

• version 1.5.8 fixes two table bugs, and adds a marginally useful feature.

  The bug fixes (one reported by Matthew Kennard, the other discovered when fixing that bug) corrected

  • lines with |’s in them and followed by a blank line were incorrectly treated as tables, and
  • tables without data didn’t display anything.

  The new feature takes advantage of a parsing glitch to allow image scaling; you can specify the height or width of the image by itself by giving a size of “=xW” (to set the width) or “=Hx” (to st the height.) This was suggested by Val Schmidt at UNH, and it’s a tiny change for adding new functionality, so I couldn’t resist putting it in.

• version 1.5.7 fixes one bug that caused discount to crash. It turns out that if you had an html block with an unclosed closing tag, the html parser would run up to the end of input and then attempt to return the next line. Which there wasn’t. So off we’d go into null pointer reference land, which meant segmentation violations and discount lying dead on the floor. This was discovered by someone who was using the iPhone app Trunk Notes, and was then reported to me by Matthew Kennard, (the developer and maintainer of that app.) The fix was fairly simple, but it still requires a release to sweep the nasty bug away.

• version 1.5.6 makes four small changes;

  1. A feature fix to make malformed tags behave more like the way the reference implementation does. This was discovered when I found that an unmatched \<word would cause the entire rest of the paragraph to be treated as a html tag, and when I fixed that I made malformed tags produce the same output that they do in the reference dingus.
  2. Add a configure.sh test for bzero() or memset()
  3. and remove a place where I used bzero()
  4. Documentation changes contributed by Josh Wood

• version 1.5.5 corrects a defect in the new table code where I was expanding an internal array instead of truncating it, and thus ending up with a table full of bogus numbers, which, if the proper build options were used, would make discount dump core.

  Fixed, I hope, and I’ve done a brief code review of the unpleasantly complicated table code to try and verify that it won’t happen again.

• version 1.5.4 corrects one more defect; when I fixed the obscure, but long-standing [footnote]: link defect in version 1.4.2, I ended up rearranging a bunch of internal data structures. This broke the table of contents code. It’s been fixed now, and I’ve added a test case for tables of contents to ensure that it stays fixed.

• version 1.5.3 corrects one long-standing, but obscure, defect: I was treating [text]whitespace(text) as a link, when the standard implementation treats that as text. []() is, apparently, only valid if there is no whitespace between the ] and the (, and 1.5.3 corrects that misfeature.

  Thanks to Bruce D’Arcus for reporting this defect.

• version 1.5.2 corrects two long-standing defects, both of which are fairly obscure but which are still wrong.

  The first defect was that you could not have consecutive >%div% blocks, because the parser would treat the second block as a blockquote continuation. 1.5.2 corrects that, and now

  >%class:A%
  >this is a class A div block
  
  >%class:B%
  >this is a class B div block
  

  properly generates

  <div class="A"><p>this is a class A div block</p></div>
  
  <div class="B"><p>this is a class B div block</p></div>
  

  The second defect dates back to the beginning of time, discount-style; I had assumed that [footnote]: links were only valid if they were in a block by themselves, and this is most certainly not the case. I had to rewrite some of the parsing in markdown.c (splitting the compile() function into compile_document() and compile(),) and now code like

  This is a paragraph with
  [footnote]: invisible!
  a surprise inside.
  

  generates

  <p>This is a paragraph with
  a surprise inside.</p>
  

  while tucking footnote away for later use. 1.5.2 does not completely correct this defect – the reference implementation will not treat [footnote]: as a link if the url part contains any whitespace, while mine does no matter what – but aside from that, it makes my implementation work more like the reference at the cost of surprisingly few cpu cycles.

• version 1.5.1 corrects a typo in the way I was generating tables (I had transposed the </tbody> and </table>), briefly documents the this feature, and adds the new flag MKD_NOTABLES (-fnotables) to disable table processing.

• version 1.5 includes support for PHP Markdown Extra-style tables, plus has a couple of minor tweaks to fix some bugs in the way list indentation is handled and list contents are blocked.

• version 1.4.5 has some documentation cleanups (courtesy of Josh Wood) and has a few minor tweaks to the list handling to better handle indentation and paragraph wrapping of multiparagraph list items.

• version 1.4.4 does some small feature cleanups, removes a few bug cases that crept in without causing test cases to fail, and rewrites all of the test cases so that they don’t rely on grep (and regular expressions) to work. No new features have crept in, but some of the existing new features have been reworked again to make them more stable (and work.)

• version 1.4.3 goes back into the code that handles []() and rewrites it so that it hews closer to the behavior of the reference dingus on some bizarre edge cases. And, as an extra bonus, it includes a Plan 9 support directory that contains a mkfile and copies of the markdown.1 and markdown.6 manpages rewritten in the USG style.

  The bizarre edge cases are ones that you don’t see in the standard test suite. It turns out that if you have a []() construct like [ouch!](http://MY 'BRAIN' HURTS), the url should be http://MY 'BRAIN' HURTS. This was amazingly difficult to make work correctly, and it involved several iterations of writing sample code and throwing it away before I figured out (on a 50 mile bicycle ride) the proper way of doing it (using regular-expression-style backtracking, which is fairly expensive on pathological edge cases, but which is fortunately quite cheap on well-formed constructs.)

• version 1.4.2 fixes a handful of bugs, rewrites some kludge-bespotted code, and adds a couple of features.

  The features are:

  1. Add the new abbr: pseudo-protocol (the html <abbr> tag)
  2. Rework the way []() links are parsed so you can have URLs that contain spaces and (properly escaped) ‘)’s

  The bugfixes are:

  1. If -fautolink was turned on, a naked @ became a mailto: link. Fixed by only triggering maybe_tag_or_link() on alphabetic characters.
  2. If -fnohtml was set, forced linebreaks became &lt;br/>. Fixed by filtering out nonprinting nonwhitespace characters on the input and using ^C as the <br/> generator character instead of converting the forced linebreak directly to <br/>.
  3. inline links (via []()) didn’t allow spaces or escaped ) characters. Fixed by rewriting linkyurl() to absorb input until a terminator instead of until whitespace or a terminator, and allowing \ to escape ), =, or “.
  4. Changed the Qchar() function so that it take an integer argument instead of a char. (reported by Josh Wood from the plan9 port.)
  5. Flag the raw: pseudo-protocol so that it will be disabled when DENY_HTML is set.
  6. Rework maybe_tag_or_link() so it better rejects random stuff that begins with a <.

  The kludge-bespotted code was the contents of linkylinky() (see feature #2 and bug #3) which had reached the point where every bugfix or feature addition would spin off another bug or two. A complete rewrite later and it doesn’t appear to be spinning off bugs, though I need to re-revisit some of the linkylinky() logic to make it hew more closely to the reference implementation.

• version 1.4.1 adds two new features suggested by Mike Schiraldi;

  1. “autolinks” – expand unbracketed urls into links (flag MKD_AUTOLINK, -fautolink in the markdown program)
  2. “safe links” – don’t linkify footnotes where the link protocol isn’t a well-known protocol (currently http: https:,ftp:,news:) (flag MKD_SAFELINK, -fsafelink in the markdown program)

• version 1.4.0 fixes a few bugs in the way I was (mis)handling utf-8 characters, plus reworks the published interface to add more methods to get data out in non FILE* form.

  The style I am trying to follow is that output functions that write to FILE*s are named mkd_generatex, while functions that allocate and populate a string are named mkd_x. I didn’t do a very good job of this in the first place, and ended up having rename a bunch of functions as I added the new ones.

  1. To access embedded style sections, I supplemented the old mkd_style (write style sections to a file) function with the new functions mkd_generatecss (== mkd_style) and mkd_css (allocate a buffer and populate it with style sections.) mkd_style remains in the published interface, but as a #define to mkd_generatecss

  2. To get xml-style output, I added the functions mkd_xml (allocate a buffer and populate it with the xml version of the input string) and mkd_generatexml (write the xml version to a FILE*.) These supplement the MKD_CDATA flag, which makes the markdown() function write xml.

  3. mkd_text (parse a single block of text, write it to a FILE*) is renamed to mkd_generateline (with mkd_text remaining as a #define) and mkd_line does the expected buffer allocation and population.

  There are some documentation cleanups as well; many versions ago, I had a MKD_QUOT flag, which vanished but floated around in the documents until now, and I’ve beefed up the description of some of the other build and operational flags.

• version 1.3.6 changes the way I parse raw html from the fairly rudimentary (but still test-passing) pattern matcher I had to a slightly more sophisticated blocking scanner. This means that the html block

  <div><pre>
  </pre></div>
  

  (which used to capture all input up to the next line that contained a solitary </div>) now properly finishes the html block at the expected place.

• version 1.3.5 fixes one bug (I’d messed up the code that handles >%class% blocks, so that an implicit >%class-title% prefix would cause core dumps on machines with sufficiently paranoid memory allocation, and adds one enhancement to the build by prefixing all the install paths with $(DESTDIR) (so you can configure the code to install into {destination}, but then make install it into a sandbox and then tar up the contents of that sandbox and extract it directly into {destination}.)

• version 1.3.4 has a couple of code changes it it; one to make it work better as a library, and the other to make it work better with a more recent version of the official markdown test suite.

  1. The library enhancement is the addition of the new function mkd_toc() which writes the table-of-contents xml into a string instead of to a FILE*; This is so Ryan Tomakyo’s rdiscount ruby binding can include a table-of-contents method on systems that don’t have the nonportable fopencookie() [GNU] or funopen() [BSD] stdio functions.
  2. The test suite enhancement removes the MarkdownTest_1.0 bodge of removing whitespace from the end of the first line in a code block, and adds support for what I call implicit reference links, which are regular old reference links ([link][]) without the second (empty) [] on them. I have added the new flag MKD_1_COMPAT, which disables implicit reference links and puts back the MarkdownTest_1.0 trailing whitespace bodge, so now both the 1.0 and 1.0.3 test suites work (with appropriate function flags.)

  (Discount 1.3.3, which is now withdrawn, broke the “two trailing spaces makes a <br/> rule. Neither the two versions of MarkdownTest nor my own test suite tested for these. Ooops on me!)

• version 1.3.2 has no code changes, just configuration ones – On Solaris, the tr in /usr/xpg4/bin does not like the character ranges [a-z] or [A-Z], and I can’t just use [:upper:] and [:lower:] because it’s declasse to do tr -dc [:upper:] when you could instead run all the way around Robin Hood’s barn to do the same thing with multiple commands.

  So I’ve changed configure to try the dumbest possible character ranges, and fail if enumerated sets, ranges, or bracketed ranges don’t work. And now it works on Solaris. Today.

• version 1.3.1 modifies the puturl() function (used inside [] links) so that it doesn’t url-encode printable characters aside from &, <, and >

• version 1.3.0 documents the 1.3.0 changes, so that they aren’t simply little magic cookies left in the code as obstacles for the unwary.

• version 1.3.0pre4 corrects a bunch of edge cases and misfeatures, and adds in a small raft of new experimental features, such as:

  • table of contents code by Jjgod Jiang
  • mkd_xhtmlpage() function suggested by Tim Channon. mkd_xhtmlpage() takes a markdown document and attempts to build a complete xhtml page out of it, inserting the contents of any headers and style blocks in the correct places.
  • many boundary condition bugs discovered by Petite Abeille
  • class blocks.
  • superscripts.
  • alphabetic ordered lists.

• version 1.2.10 corrects a tiny bug in [title](url) and ![title](image) links, where if the [title] was empty the tag wouldn’t be expanded. Admittedly, it’s a good idea to actually put a meaningful title in a link or image, but there’s no reason to actually require it and I’ve cleaned up the code to reflect that.

• version 1.2.9, which includes:

  1. Add the new flag MKD_NOHTML, which blocks all embedded html and forces the user to use the markup language for doing markup.
  2. Rearrange RELAXED_EMPHASIS so that it only applies to underscores in the middle of words. This breaks the emphasis rules for things like a_b (or, if you’re describing code, thing like size_t or time_t or …) which may not be a problem for you if you’re writing about code.
  3. Prune leading spaces off atx headers, so # FOO won’t generate <h1> FOO</h1>
  4. Tweak some prototypes to reduce the # of complaints that a certain anal-compulsive pseudo-C compiler spits out when building the package.

  Version 1.2.9 was originally version 1.2.8, except that 1.2.8 was incorrectly published after git didn’t push the 1.2.8 changes back to my master repository. It pushed the VERSION file, which was something, but not the actual code changes.

• version 1.2.7 contains two bugfixes and one new feature:

  1. Corrects a bug with strget() not properly returning EOF because it was returning a char or unsigned int instead of an int (should be a signed int, but that way lies C99 madness.) This bug was reported by A.S.Bradbury, who discovered it when trying to build markdown with a 64 bit version of gcc.
  2. Corrects a boundary condition bug with ![]() and/or [](), where the content parser stops parsing 4 (for IMG) or 5 (for HTML) characters before the end of the embedded content. This bug was reported by Christian Herenz, who has been finding many many boundary condition bugs in the way ![]() and []() (don’t) work.
  3. The new configuration option --enable-superscript turns on fancypants substitutions that translate word^word into word<sup>word</sup>. I’m not sure if it’s the most useful thing in the world, but I have found cases where I’d like to be able to get superscripts generated just by themselves. (and since I don’t code in PASCAL any more, I don’t need to worry about tripping over a dereferencing ^ as much as I used to worry about it.)

• version 1.2.6 contains patches to work around some of the broken features in Xcode on macos 10.5. The AC_SUB function in configure.inc has been reworked to not do echo -n (because that one spits out ^? when I echo an empty string,) and I’ve had to write my own echo because the builtin echo that the Xcode /bin/sh (actually the gnu clone shell) uses no longer accepts -n as the “do not put a newline at the end of this line, k?” switch.

  There’s also another tiny bugfix: I discovered that if I had a line of the form [link][] text text [link][]: ..., it would get treated like a footnote despite obviously not being one. This release corrects the isfootnote() function in markdown.c so it won’t do that anymore.

• version 1.2.5 fixes a small bug inside <url>’s, where the url would be parsed as if it was text instead of being treated as code.

• version 1.2.4 fixes a bug which caused builds to fail if --enable-dl-tag was not set; I was combining code that required WITH_DL_TAG and code that didn’t, so a isdefinition() call crept into unDL_TAG territory.

  To fix that bug, I moved the test for WITH_DL_TAG into isdefinition() and worked the code so that isdefinition() always returns 0 if the code was configured without --enable-dl-tag.

• version 1.2.3 fixes two bugs in the markdown program, fixes one bug inside the markdown library, adds support for testing mkd_text() inside the markdown program (-t text processes text with mkd_text(),) and adds support for sequential <dt> blocks with sequential =tag= labels.

• version 1.2.2 fixes two bugs;

  1. it now properly blocks toplevel paragraphs when they are immediately followed by a non-paragraph block.
  2. footnotes are no longer forgotten inside [] sections.

• version 1.2.1 cleans up a few more memory leaks, adds a new function (mkd_document(), which returns a pointer to the compiled document,) adds the new configuration option --enable-amalloc, which builds against a debugging memory allocator, and, finally, it rips into the guts of the second-pass of the compiler and rearranges how * and ** are compiled to (hopefully) guarantee that they will never generate incorrect XML.

  The “rip into the guts” part adds a half-pass to the second pass of the compiler; htmlify() now writes into a linked list of text and emphasis tokens, which are then flattened into the output document by the new function emblock() which attempts to match emphasis blocks up properly instead of the old-style “generate <strong> and <em> blindly as we roll on down the line.”)

• version 1.2.0 has two subtle output change, plus a bugfix that steps on a memory leak in codeblock().

  The output changes (which caused the minor# to go from 1 to 2) are

  1. The contents of a <html tag> are now parsed as code, not as text. This keeps <a href="linky"> from being written as <a href=&ldquo;linky&rdquo;>, and may thus confuse source that was written with \"’s to get around the smartypants extensions.
  2. When printing <email@address> automatic links, ensure that the human-readable part of the link doesn’t include a mailto:, just the mail address.

• version 1.1.5 contains assorted cleanups:

  1. First, try to make it more portable to not-quite-standard systems (like Windows + MSVC) by not using alloca anywhere and by having configure look for strcasecmp vs stricmp and strncasecmp vs strnicmp, then
  2. Clean up some horrid interactions with bsearch by converting blocktags[] into an array of struct kws and having isopentag() construct a struct kw and use that for searching instead of malloc()ing a buffer, making a null-terminated string, then comparing against that, and, finally
  3. Catch a horrible parsing bug where I didn’t null-terminate html blocks during the blocking pass of the compiler, so at cleanup time I ended up double-freeing blocks and causing the program to dump core if I used anything other than alloca() inside isopentag().

• version 1.1.4 adds some build tweaks to (hopefully) finally make the sample programs build on Windows.

• version 1.1.3 adds a README and an INSTALL file, plus tweaks the build process to not use bzero() [because it broke the gcc on macos] nor the revised version of cstring.h [because, surprise, it broke the gcc on macos.]

• version 1.1.2 adds support for quoting link titles with (…) (as well as the regular " and ' quoting formats), plus continues the possibly-doomed project to make it portable to MS-DOS+mingw systems:

  1. ifndef bzero, define bzero(s,n) memset(s,0,n)
  2. ifdef srandom, define SEEDRNG(x) srandom((unsigned int)x)
     eldef srand, define SEEDRNG(x) srand((unsigned int)x)
     else define SEEDRNG(x) (void)1
  3. ifdef random, define COINTOSS() (random()&1)
     eldef rand, define COINTOSS() (rand()&1)
     else define COINTOSS() 1

• version 1.1.1 adds NO NEW FEATURES, but simply adds an checks for pwd.h and getpwuid() to configure.sh, to make it slightly easier to build the code on (mingw‘ed) MS-DOS systems.

• version 1.1.0 adds a new output flag (MKD_CDATA) so that it can generate output suitable for use as data in an xml document. This feature was added in because it appears that the definitions for rss and atom newsfeeds have shifted and I can no longer protect the contents of a field by wrapping it in ![CDATA[…]]. In addition, I’ve renamed the version field from version[] to markdown_version[], thus breaking the library interface but in a way that actually makes the variable more useful.

• version 1.0.0 is the first “yes, I’ve seriously nailed down the published interface and have been using the code enough to call it stable!” release (which will probably turn out to have some horrible flaw somewhere in it.)

  There are not very many new features in 1.0.0; I’ve written a manpage for theme(1), changed MMIOT in mkdio.h from a void* to a void so it will more closely resemble the traditional FILE* that stdio provides, and collapsed a couple of annoying buglets. I’ve increased support for styles by handling style blocks (<style>…</style>) by parsing them into a separate bin so they can be retrieved with the mkd_style() function. But the big thing is that I’ve officially nailed down the interface under the hopes I don’t find Yet Another Thing that will cause the whole shebang to go obsolete.

• version 0.9.7 has support for the raw:, id:, and class: pseudo-protocols, plus has a theme clone added as an example program.

  0.9.7 may end up being 1.0 after I document it. I’ve used the current version of markdown to re-markup my homepage and I’ve got it in the copy of annotations that’s running on the weblog server, and it seems stable.

• version 0.9.6 is where I finally snapped and found a better workaround for the nasty tabstop feature in the standard. I’ve added the configuration option --with-tabstop which tells markdown to properly expand tabs to eight spaces, then added in another input flag (MKD_TABSTOP [0x0200]) that can be set when you have to use markdown on input that’s written using the standard (but still wrong) 4 space tab.

• version 0.9.5

  1. lowercase tag names for better compatability with anal-retentive newer web standards.
  2. clean up the usage comment and example in the header.
  3. • In mkd2html, don’t #include <libgen.h> unless the system has it.
     • Also in mkd2html, define a basename() if the system doesn’t have it.
  4. Trim back leading spaces when returning fields from the document header.
  5. #include <string.h> in cstring.h so we can get the prototype for memmove()
  6. Update the mkd2html test program.
  7. Make the <img tag more xhtml-friendly by ending it with />.
  8. . . . now maps to …, just like ... does.
  9. Move the line trimming into codeblock() – we need to be able to do trailing spaces on regular paragraphs.

• version 0.9.3

  The festival of final cleanups and bugfixes continues.

  1. Put most of the smartypants entity replacement strings into a table and do one table lookup instead of a bunch of individual tests.
  2. I reversed the order of height and width in the =widthxheight processing part of a ![][] tag. Sheeesh!

• version 0.9.2

  • the particular type of whitespace is not important when matching footnote tags ( “this\nis” matches “this is”, for example.)
  • handle both <orc@pell> and <mailto:orc@pell> automatic links.
  • mkd_dump() USES the output FILE* now.
  • quoted blocks are no longer parsed as toplevel blocks now that toplevel blocks have different paragraph handling.
  • smartypants() converts 1/4 to ¼, 1/2 to ½, and 3/4 to ¾. It also knows about ¼th and ¾ths.

• version 0.9.1

  1. Remember that I said that the version 0.9 published interface was now a “version 1.0 release candidate?”

     I lied.

     The 0.9 published interface was carrying around a dependency on an internal data structure that was not publically defined, so I had to rearrange the published interface to get that back into the library space where it belonged. And in addition to that I needed to add the accessor functions for the document header fields, so the poor interface was doubly whacked.

     Given any luck, the 0.9.1 published interface will be the version 1.0 release candidate. I certainly don’t want to have to go back and redo all of the other code that calls the existing public functions again.

  2. And, on a slightly less destructive note, version 0.9.1 passes John Gruber’s test suite and I only had to put two horrible hacks in to make it conform to the standard.

• version 0.9 fixes most of the remaining failures when running the test suite. It has a few more features as well:

  1. I now support Pandoc header blocks – enabled during configuration with --enable-pandoc-header, and turnoffable by passing the MKD_NOHEADER flag to mkd_in() or mkd_string().
  2. The published interface is now a version 1.0 release candidate. I’ve expanded it to allow finer-grained access to the compilation process: =mkd_compile()= compiles the output of mkd_in() or mkd_string(). It is a boolean function that returns yes/no, depending on whether it compiled the code. =mkd_generatehtml()= generates html from the compiled input doc. =mkd_cleanup()= cleans up all of the data structures associated with the input doc.

  3. Various paranoia and edge-condition patches.

  4. version 0.8.2 fixes two bugs that I found when I compiled the MacOS TextMate editor manual.

     1. I forgot to define <table> as a block html element.
     2. UTF-8-encoded characters with bit 0x80 set were being silently dropped from the input.

  5. version 0.8.1 has been tweaked so that it (mainly) passes the official markdown test suite. There are a bunch of minor tweaks to correct places where 0.8 failed to pass the test suite, and now 0.8.1 passes 16 of 19 tests where 0.8 failed 15 of 19 tests.

     I added a magic enviroment variable to to markdown program to tell it whether to enable or disable smartypants, links, or images. MARKDOWN_FLAGS is a bitmask consisting of:

     #define	bit	what does it do?
     MKD_NOHEADER	(0x0100)	Do not process the document header.
     MKD_NOPANTS	(0x04)	Disable smartypants processing.
     MKD_NOIMAGE	(0x02)	Disable ![][] expansion, convert <IMG tags into &lt;IMG.
     MKD_NOLINKS	(0x01)	Disable [][] expansion, convert <A tags into &lt;A.

     To run the test suite, I set MARKDOWN_FLAGS=0x4 to stop those pesky smart quotes from messing everything up.

  6. version 0.8 adds in a whole bunch of new features, so many that if I had nailed the published interface down this would be the time to roll up to version 1.0. In 0.8, I’ve:

     1. Added markup for definition lists is now supported, if enabled during configuration with the --with-dl-tag option. The new markup is ^=tag=$, with the list item – indented at least 4 spaces – beginning on the very next line.
     2. Start writing public interfaces to get at some of the internals in the code.
     3. A sample “mkd2html” program that demonstrates the (still unfinished) internal interfaces.
     4. Correct the order of <code> and <pre> in code blocks.
     5. Properly inherit from the parent frame in reparse()
     6. Add a parse tree dumper (dumptree()) and modify the markdown program to use it when passed the -d flag.
     7. Rip markdown.c in half; one half is the backend html generator, the other half is the tree builder.
     8. Change mkd_text() and markdown() to return int
     9. Write a manpage for markdown(1).
     10. Completely rework the way whitespace is handled. It is now a paragraph separator, not a terminator.
     11. Configuration cleanups to make it happier on more platforms.

  7. version 0.7.5 bumps up 0.7 after 12 hours because I wasn’t happy with the way that title= and alt= were being encoded, so I switched over to do doing a generalized reparse of those fields instead of having a custom parser (I was trying to encode “Le Gâteau Noir”, and it mangled the â into a literal â. This would not do, so I had to fix it.)

     As a happy coincidence, this let me clean up a few of the places where I had a collection of functions that did almost the same thing and replace them with calls to the one function with a new flag setting to turn off and on the one difference (expanding " into " is important when you’re inside a title= or alt=, but not important anywhere else. I could just have " expand everywhere, but the generated html is not quite so readable that way.)

  8. version 0.7 revisits the way lists are assembled and finally gets them to work closer to the same way that they work in the reference implementation. I have also gone over the output and made it so that text inside embedded urls and images ([] and ![]) has & and < escaped in the output.

  9. version 0.6 cleans up some bugs and includes the start of an automated test suite.

     Version 0.4 had a couple of buglets in it that would cause markdown to dump core if I had a dangling list item or a >-quoted paragraph as the last paragraph in a list. After fixing those bugs, I decided that this was a good time to start writing an automated test suite so I could type “make test” and have all of the hopefully-fixed defects passed through the mill without any further human intervention.

     As an aside, I’ve cleaned up the code for ##headers## so that extraneous <p>’s aren’t wrapped around the <Hn>…</Hn> code. It shouldn’t really matter, but I do look at the generated html and this bothered me.

  10. version 0.4 fills in some more of the requirements for using it as the formatter for annotations – I am now reparsing the description field in [link][fields], so if link contains text that wants to be marked up it will be. I use it for things like

      ->[![dust mite](http:/~orc/dustmite_t.jpg)](http:/~orc/dustmite.jpg)<-
      

      Which formats down to

      <center>
      <a href="/~orc/dustmite.jpg"><img src=http:/~orc/dustmite_t.jpg></a>
      </center>
      

      Without forcing me to go through the sort of quoting gymnastics I had to do with the old home-grown annotations formatter.

      I cleaned up a small wad of boundary conditions when I did a code tidying sweep, and a few others when annotations started dumping core when I would edit a post.

      And, finally, 0.4 includes a nroff-ed translation of John Gruber’s Markdown language definition, which was a fun-filled hour or so tour through the interesting ways that groff can make a Macbook go to 100% cpu usage.

  11. version 0.3 is a cleaned up version of 0.2.

      I didn’t like the version 0.2 published interface, so I discarded it and wrote a new one from scratch. The published interface now consists of markdown(void* blob,FILE*out,int flags), void* mkd_in(FILE*in), mkd_string(char*bfr,int size), and mkd_text(char*bfr,int size,FILE*out,int flags). markdown() is the full-document formatter (the input blob is assembled by mkd_in() or mkd_string), and mkd_text() does simple markup (no code sections, no pure html, no quoting) on the input string.

      Version 0.3 also has a manpage for the published interface, and a little demonstration program (named, intuitively, markdown) that is a formatting pipe filter.

  12. version 0.2 started to define a published interface, so I could bolt it into annotations to see what would fall apart.

  13. version 0.1 is the first public release