% Discount -- a C implementation of the Markdown markup language
% Jessica L. Parsons (orc@pell.portland.or.us)
% 

# Discount

## Version 2 Downloads

> To build discount, untar your selected tarball, cd into the directory
> it creates, then do `configure.sh` to generate your Makefiles. After
> doing this, a `make` should give you a functional stack of programs
> and libraries.
> 
> Discount builds, for me, on MacOS 10.12, FreeBSD 4.8, NetBSD 8,
> Minix 3, and Debian Linux (dunno which version, but it's a `systemd`
> joint that’s running a 3.10 kernel.) It may build on SLS Linux and
> Windows with mingw, but I'm not sure about that.

*   [Version 2.2.7d](discount-2.2.7c.tar.bz2) One more maintenance release to fix a few more bugs:
    
    *   markdown extra footnotes were slightly broken -- two adjacent
	footnotes (`[^1][^2]`) were being treated as a regular old hyperlink
	because the code I put in originally was a huge old monster bodge.
    *   change the description for -fstrict.
    *   add a fistful of test cases for the new! improved! Markdown.pl compatible `<tags>`
    *   rework the tag handler to more closely adhere to Markdown.pl's
	observed behavior; also treat incomplete tags as actual tags in
	compatability mode (-fstrict)
    *   fix weird behavior on freebsd 4.8 w/ gcc 2.95.4; if an #ifdef
	... #else ... #endif wraps the end of a `if ()` it pukes up an
	error about a syntax error in a macro; pull those tests out and
	conditionally #define a macro instead
    *   the behavior of tags-starting-with-alpha was not like that
	of the reference implementation; fixed (in a very script-kiddy leaky
	fashion which I can't resolve with a runtime flag thanks to the 32
	bit flags field in the 2.x series :-(

*   [version 2.2.7b](discount-2.2.7b.tar.bz2) *Another* maintenance
    release to fix a bug in the Makefile; I didn't have the proper
    dependencies for the pandoc\_headers tool, so if a parallel make was
    tried it would fall over complaining about missing objects when it tried
    to link everything together.
    
*   [version 2.2.7](discount-2.2.7.tar.bz2) A maintenance release to
    clean up a buffer overflow when generating label names (backported from
    the new v3 under (slow) development)
    
*   [version 2.2.6](discount-2.2.6.tar.bz2) should have been released
    last fall, but I was too busy working on breaking the published interface
    so I could expand the flag structure to permit more than 32 feature
    flags. *\*sigh\**
    
    It's not a huge release; it fixes a few bugs, tweaks a few things,
    and adds a couple of features:
    
    bugs
    
    *   deal with cached text before handling fenced code blocks (because
	otherwise it would accumulate the contents of things between the
	blocks and spit them out afterwards.)
    *   don't give a code block class if it's just an empty string
    *   mkd2html wasn't passing pgm to hoptusage()
    *   mkd2html wasn't properly handling two arguments for input &
	output files. While I was at it, add the -G option to use gfm\_in()
	to read the input file in (and preserve newlines)
    *   push the fencedcode block handler up to the toplevel scanner so
	it will handle blank lines inside the fence (it's splitting the code
	block up into paragraphs and treating them as text.)
    
    features
    
    *   add an anchor format callback, plus add the -x option to
	the markdown program to squash toc anchors to (close to) github
	compatability
    *   add an external code block formatter to the markdown program
    *   clean up all references to flags; define them EVERYWHERE as `mkd_flag_t`
    *   support external code formatters
    
    tweaks
    
    *   configure to
        1.  use the modern standard uintXX\_t types for DWORD, WORD, BYTE if possible and
        2.  if a program is defined via an environment variable (like
	    CC) try to use that variable instead of doing a path search
	    in AC\_PROG
    *   make a test program to examine pandoc header elements
    *   patch the cmakefile so that has an option to only generate a library
    *   set the Plan9 CFLAGS to the two extensions (instead of having
	them be part of the CC macro) and wipe out any CFLAGS (like -T, which
	tells the Plan9 posix C compiler to be stupid) that already exists
    *   the Makefile macros a bit so people can wedge in their own
	compilers (BUILD) and linkers (LINK) during a build
    *   when dumping the parse tree, say which header is being dumped -- h1..h6 vs just header
    
*   [version 2.2.4](discount-2.2.4.tar.bz2) fixes a couple of small
    buglets, adds a couple of new features, and tweaks the html5 support
    module to chase the html5 standard:
    
    1.  (**bugfix**) when splitting a line, null terminate the new line;
	also don't inherit the dle from the previous line
    2.  (**bugfix**) fix the broken footnote code; allow multi-paragraph
	footnotes, not just one line of footnote.
    3.  (**feature**) support github checkbox list items (static configuration option)
        1.  `--github-checkbox` does the checkbox with html check entities
        2.  `--github-checkbox=input` does them with html `<input>`
	    elements (set to disabled so that people can't check and uncheck
	    willy-nilly)
    4.  (**feature**) on systems that support it, use the "destructor"
	attribute on `mkd_shlib_destructor()` so it will run automatically
	when the library unloads.
    5.  (**feature**) add `--cxx-binding` option to `configure.sh` to
	generate a `mkdio.h` that’s got an `extern "C"` wrapper around it.
    6.  (**tweak**)delete `HGROUP` from the list of html5
	elements, because the w3c working group decided to [punt
	it](http://lists.w3.org/Archives/Public/public-html-admin/2013Apr/0003.html)
    
    The bugfix for not null-terminating a split line was reported by
    Github user [fCorleone](https://github.com/fCorleone), who was running
    a input mangler program ([afl](http://lcamtuf.coredump.cx/afl/))
    to stress discount by feeding random garbage to it. It saw that
    the `splitline()` function (used when breaking out embedded chunks
    of html) was not null-terminating the split line chunks, and was
    throwing errors on that. (this is something that `clang` on modern
    macos catches if you use the `-fsanitize=address` option.)
    
    I never actually use footnotes, so I'd been running for about a decade
    assuming that footnotes were one or maybe two lines, but github user
    [somasis](https://github.com/somasis) pointed out that this was wrong
    wrong wrong! Ooops; at least it's fixable w/o much pain and suffering.
    
    `HGROUP` was caught by (no longer registered) github user
    Crypto-Anarchist in their own branch of discount, so I cherry-picked
    that changeset and pulled it back into the mainline discount.
    
*   [version 2.2.3a](discount-2.2.3a.tar.bz2) 2.2.3 has a configuration
    glitch (not properly testing for the existance of S\_ISSOCK, et seq) which
    2.2.3a corrects. I also pulled the plug on the single use of `alloca()` in
    `theme.c` (Windows MinGW doesn't support it properly(?) and in this one
    case I'm better off just `malloc()`ing the offending thing and letting
    it be garbage collected when the program finishes.)
    
*   [version 2.2.3](discount-2.2.3.tar.bz2) I turn around and blink,
    and suddenly 13 months have gone by while I slowly tested various beta
    versions of 2.2.3, and now it's 2018 and a small collection of cosmetic,
    portability, and build tweaks, plus some and actually bad bugs have been
    cleaned up:
    
    1.  Have `tools/branch` be a no-op if there’s no git that I can
	use to check for my SCCS status.
        
    2.  Add user-contributed cmake support
        
    3.  check for non-null T(link) in the safelink function; rename
	paranoia.t to safelink.t, add a couple more tests for safelinks
        
    4.  tweak the safelink detection code to allow more types of url fragments
        
    5.  when finding the installer, check that -s works (doesn't work on Minix 3?)
        
    6.  Correct makepage & theme to use the new `set_flag()` return scheme
	(0==success, !0==pointer to bogus flag)
        
    7.  if an unknown flag was in the middle of a comma-delimited flag
	string (like -flatex,bogus,footnote), the markdown program would
	incorrectly report the first flag as unknown (because set\_flag
	returned 0 on error, 1 on success and the strtok() of the flag string
	had already replaced the commas up to that point with nulls.)
        
        Change it so that set\_flag returns null on successful processing
        and a pointer to the offending flag on an unknown one.
        
    8.  Tweak the install rule to install a GNU-style .pc file iff
	@MK\_PKGCONFIG@ is not defined as #
        
    9.  Hand-resolve a conflict in the msvc Makefile
        
    10.  Add in paranoid list splitting \[EXPLICITLIST\] (the default
	behavior many many versions ago before I realized it wasn't the
	standard) (and take the last of the 32-bit flag mask for it) between
	adjacent ordered and unordered lists.
         
    11.  Create a utility function \[`notspecial()`\] for theme & mkd2html
	-- check a filename to see if it's a special file that doesn't need
	to be deleted or to have a .html suffix added to it. It only works
	on machines that have a stat() system call + the S\_ISFIFO, S\_ISCHR,
	and S\_ISSOCK macros, otherwise it thinks that nothing is special.
         
    12.  Add a README for the utilities in the tools subdirectory
         
    13.  Add the git branch name into the version string if we're not
	on the main branch.
         
    14.  Add support for NMAKE and Visual C++ toolset. (courtesy of
	Martin Hofmann (tin-pot)'s fork of Discount).
         
    15.  Add a "dirty" flag to the Document structure & set it whenever
	a callback changes (github issue #136) so that the next mkd\_compile
	will regenerate the document.
         
    16.  Tweak superscript handling to be able to superscript something
	wrapped in html
         
    17.  Tweak the `show_flags()` function so that if it's called
	verbosely it will show synonyms for named flags. (Calling it verbosely
	is done by the `V` option, which is overloaded for verbosity when
	listing flags.)
         
    18.  Process html blocks in compile (as well as in
	compile\_document(); `compile_document()` needs to handle `<style>`
	blocks and `compile()` needs to handle html blocks that are nested
	inside blockquotes, tables, lists, &tc.
         
    19.  Add `-h1-title` as an option to configure.sh; this enables
	code that uses the first h1 in a document as the document title
	(in mkd2html & theme)
         
    20.  add `gethopt()` -- a q&d getopt clone that accepts both full-word
	& single character options -- instead of doing kludgy bespoke argument
	parsing in mkd2html & theme.
         
*   [version 2.2.2](discount-2.2.2.tar.bz2) A few months worth of
    bugfixes, mainly for annoying bugs but one for a serious one:
    
    *   In `mkd_line()` I was doing a sneaky hack to take the rendered
	text out of the MMIOT it's in which was fine if I was building
	discount w/o `--with-amalloc`, but horribly wrong `--with-amalloc`;
	the return from `mkd_line()` is a freeable string, so my sneak (I
	was pulling the contents of the rendered string out, then zeroing
	the pointers in the MMIOT; if amalloc was active, the rendered
	string was in the middle of a allocated block which could not be
	freed (and would cause core dumps on some systems. Kludges *always*
	come back to bite you in the ass)) was terribly terribly wrong.
        
    *   Redo toc anchor generation to reduce namespace collisions
	(by encoding out of namespace characters to -XX- hex sequences
	(except space, which encodes to just -) in html4, %XX in html5
	(formerly urlencodedanchor)).
        
    *   Add html5anchor as a synonym for urlencodedanchor (urlencoded
	is depreciated, but will stick around like an unwelcome guest until
	the next major release) now that I'm doing html5 encoding there.
        
    *   When generating a TOC, don't anchor-encode the human-readable
	contents of the links .
        
    *   "quote" "everything" "when" "I" "generate" "librarian.sh"
        
    *   Horrifying kludge to deal with destination directories with spaces in them;
        
        1.  quote filename arguments to AC\_SUB (via the \_\_ac\_quote builtin), and
        2.  instead of using the shell to check for sed substitute markers
	    (if I do `echo "$*" | sed -e 's/;/\\;/'` it collapses a layer
	    of backslashes needlessly. It's awful enough to do
	    `echo $* | sed -e 's/ /\\\\ /'`
	    (so sed won't eat the backslashes) but to do `sed -e
	    's/ /\\\\\\\\ /'` to keep sed + the shell from eating the
	    backslashes is just appalling) I generate the little C program
	    `config.sed`, which generates the sed pattern, escaping the
	    `;` and NOT collapsing backslashes.
            
            Ugh. It's all pretty gross, but it looks like it works? At least for discount.
            
    *   Always have `mkd_xmlpage()` generate a title element, even if it's empty
        
    *   Add `<form>` … `</form>` to the set of known block tags
        
    *   Use `\r` instead of `^C` for the internal hard end of line marker
	(which means if I don't unescape it when rendering it won't leave
	`^C` poop in the output.)
        
    *   Bring the installation instructions a little closer to up to date.
        
    *   When there's either `pkg-config` on the system or if the configure
	option `--pkg-config` is used, generate a pkgconfig `.pc` file.
        
*   [version 2.2.1](discount-2.2.1.tar.bz2) A few months of small
    bugfixes, a few tweaks, and some more conversion of static flags to
    runtime ones.
    
    *   Update the muñoz test case for a text fragment with an 0xff
    *   Kludge `peek()` and `poke()` to not sign extend on machines with
	signed chars, so a 0xff character will not sign extend and become an
	EOF. This breaks a test in muñoz.t (which tripped the 0xff becomes
	EOF bug) so that test needed to be rewritten.
    *   Have `configure.sh` not do the `WinDef.h` short circuit
    *   include `"config.h"` to pick up the definition of
	`DWORD/WORD/BYTE` (windows support) & add a new label field for
	better TOC label generation
    *   Wasn’t updating `config.sub` with {scalar\_type}->{scalar\_type}
	on windows machines.
    *   Tweak `mkd_xmlpage()` so that it only uses the published interface.
    *   Eat one of the two remaining flag bits (`flag_t` structure -- or
	64-bit int, which isn't portable to old Unices – here we come!) to
	make LaTeX support a runtime flag.
    *   Add some missing dependencies to the makefile
    *   In makepage, I was using the wrong argument for the file
	to make a page from. needed to check for `argc > 0` & `argv[0]`
	(after shifting argc/argv by optind) but was checking `argc > 1` &
	`argv[1]`, which was something less than useful.
    *   Strip out `--enable-all-features` from `configure.sh` (obsolete
	now after the mass conversion of features from compile-time to
	run-time.)
    *   Add `config.h` to the includes for amalloc (for the if `#define`
	to make clang stfu)
*   [version 2.2.0](discount-2.2.0.tar.bz2) Many **MANY** tweaks over
    the last year, including---
    
    *   mathjax support (`with-latex` changed to the runtime flag
	`MKD_LATEX` in [2.2.1](#v2.2.1)) (`$$..$$`, `\(..\)`, and `\[..\]`,
	not `$..$`))
    *   make the `amalloc()` paranoia malloc library even more paranoid
	by putting markers at the start *and* the end of each allocated block.
    *   Redo comment block handling; standard markdown only treats
	comments as block html if the start comment marker starts at the
	beginning of a line and the end comment marker is at the end of
	a line.
    *   clean up broken and insufficiently paranoid parts of `configure.inc`
    *   When attempting to match the closing tag of an html block,
	don't advance the match index unless that character actually
	matches. (defect: `<p></>*` was splitting into 2 lines when it should
	have generated `<p><p></>*</p>`)
    *   if `mkd_compile()` is called multiple times, actually recompile
	the document if the flags change.
    *   When processing automatic links, explicitly allow extended utf-8
	characters as part of the url.
    *   Tweak `configure.inc` to quote `__cwd` & `__d` so that a
	`$__cwd` with spaces in the path won't make `configure.sh` (or
	`make install`) puke
    *   messed up the `fwrite()` error check in `mkd_generatehtml()`,
    *   return EOF instead of -1 on error in `mkd_xhtmlpage()`,
    *   if the mkd output fails, exit with nonzero status
    *   Handle error conditions and pass errorcodes out of various
	`mkd_xxx()` output functions (inspired by a patch written by [*Koen
	Punt*](mailto:koen@koenpunt.nl))
    *   in `mkd_document()` don't pad the generated html with a 0 unless
	it's actually generated.
    *   If the C compiler generates .dSYM directories, get rid of them
	during distclean
    *   Change the `configure.sh` message for the `--with-`(foo) variables
	to reflect what it's actually doing with them.
    *   Theme really wants the old behavior of `--with-`(foo), so pass
	those settings into theme in a #define
    *   Have `configure.sh` just state that an option is not supported
	instead of dying.
    *   Convert many configure-time settings to runtime flags
    *   Manpage editing by [*Nathan Phillip Brink*](mailto:binki@gentoo.org)
    *   Update plan9 support.
*   [version 2.1.8a](discount-2.1.8a.tar.bz2) When I put in the patch to ...
    
    *   Change the mail demangler to a debian-specific **always mangle one
	way** hack. (enabled with the configure.sh option `--debian-glitch`)
    
    I messed up the format string and made the mangled email address
    into a fixed bogus string. Sigh. Fixed (thanks to a patch from
    *Alessandro Ghedini*), updated (and I really have to expand the
    runtime configuration flags array to be long enough to fit 64
    settings, but that’s a fix for a different day) and released.
    
*   [version 2.1.8](discount-2.1.8.tar.bz2) After a year or so of
    letting the code sit and slowly accumulate fixes, a new version which
    fixes a wad of bugs and adds a few new features. Some of this code is
    front other people, and those changes will be marked with their names:
    
    *   FINALLY address the bug where markdown extra-style footnotes
	lose numbering when they show up in nested element; I was not
	carrying the m-e reference# inside the footnotes structure, but was
	instead carrying it in the parent structure and not updating it. So
	I changed the footnotes structure to include the reference + the
	list of footnotes, which made the misnumbering go away on my tests.
    *   Fix makefile distclean to cleanup all the generated files and
	corrected the names of the installed sample program man pages to end
	in .1 (*Mark Pizzolato [mark@infocomm.com](mailto:mark@infocomm.com)*)
    *   Change the mail demangler to a debian-specific **always mangle
	one way* hack. (enabled with the configure.sh option –debian-glitch)
    *   Add –with-unmangled-email compile-time flag to disable mailto:
	mangling
    *   Allow the magic output filename `-`, which means send output to
	stdout instead of to a file.
    *   Fix a bug where autolink + github flavored markdown absorbs the
	^C eoln character into a link at the end of a line.
    *   Tweak install.samples so that the user can supply a SAMPLE\_PFX
	on the command line `SAMPLE_PFX=discount- make install.samples`
	to install the sample programs with a package-specific prefix.
    *   Emit pages in utf-8 instead of us-ascii (simply a
	change to the Content-Type meta) (*Nathan Phillip Brink
	[binki@gentoo.org](mailto:binki@gentoo.org)*)
    *   Patch the horrible list handler to support long numeric list items
	(*George Hartzell [hartzell@alerce.com](mailto:hartzell@alerce.com)*)
    *   Various bugfixes (*Masayoshi Sekimura
	[sekimura@gmail.com](mailto:sekimura@gmail.com)*)
    *   Fix support for `CFLAGS=-m32 ./configure.sh` by using
	CFLAGS for all build invokations of CC. (*Nathan Phillip Brink
	[binki@gentoo.org](mailto:binki@gentoo.org)*)
    *   Github-style language attributes on fenced code blocks (*Loren
	Segal [lsegal@amazon.com](mailto:lsegal@amazon.com)*)
    *   When defining WORD & DWORD, check first for the MS Windows
	WinDef.h file; if found, include it instead of defining WORD &
	DWORD ourselves.
    *   support url-encoded anchor links with the
	`--with-urlencoded-anchor` option (*Daisuke Murase
	[typester@cpan.org](mailto:typester@cpan.org)*)

*   [version 2.1.6](discount-2.1.6.tar.bz2) does nothing except for
    some bugfixes (and ignores some particularly scary ones that I /must/
    fix soon) and adds two small features.
    
    The bugfixes are:
    
    1.  A < at the end of the input is exactly the same as \\<(space)
    2.  Markdown.pl does not appear to escape `\<[nonwhite]` sequences. Sigh.
    3.  Tweak the previous `Markdown does not escape...` commit to simply
	push out the backslash and back up to the start of the `<[nonwhite]`
	sequence, so -fnohtml will continue to work.
    4.  Treat hard `<br/>` (via two spaces) as whitespace.
    5.  Tweak divquote handling so that two adjacent divquotes won't
	die if there is a space between the second > & leading %
    6.  Tweak one of the list tests back to the previous behavior (I've
	put in a hack for list indentation, and accidentally committed the
	changes. Oops!)
    
    The features are that I now use styles for table cell alignment
    instead of `align=`, and that I'm using the 3-clause BSD license
    for this release (because there is one widely used closed-source
    license that claims that you can't dynamically link with code that
    uses the 4-clause license. Fine. I’ll 3-clause this release to
    make the stupid GPL happy.)
    
*   [version 2.1.5a](discount-2.1.5a.tar.bz2) does *even* more cleanup to
    deal with [clang](http://clang.llvm.org), plus adds a few small features:
    
    *   MKD\_NOSTYLE -- treat `<style>` blocks as regular html.
    *   some github flavored markdown support; gfm\_...() input methods
	that put hardbreaks (== two spaces) at the end of every input line.
    *   support for github flavored markdown backtick-delimited code
	blocks (in addition to tilde-delimited codeblocks)
    *   in the `markdown` program, add
        1.  \-S flag (tell markdown to spit out style sections)
        2.  \-n flag (tell markdown not to output generated text)
	
*   [version 2.1.3](discount-2.1.3.tar.bz2) cleans up a couple of
    bugs that have been here for a while, plus tweaks the build process
    a bit for better compilation with the [LLVM](http://www.llvm.org/)
    [C compiler](http://clang.llvm.org/), [mingw](http://mingw.org/), and
    [cygwin](http://cygwin.com/).
    
    The bugfixes are
    
    1.  Stop tripping over tables with leading `|`s; the first
	implementation of tables treated leading `|`s as the end of an empty
	table cell. 2.1.3 corrects this to properly be merely decoration.
        
        As a side-effect, you now need to have all the rows of the
        table either with a leading `|` or not; you cannot mix and match
        them, sorry.
        
    2.  For some mysterious reason I was treating the `<br>` tag as a
	block-level html tag, so my html blockifier would split paragraphs
	with explicit `<br>`s in them.
        
    3.  The table of contents code was apparently generating bad html
	(something I never noticed, because I never use that feature,
	alas!) but *Stefano D'Angelo* contributed a patch to clean up the
	generated html to make it correct.
        
*   [version 2.1.2](discount-2.1.2.tar.bz2) tweaks table handling so that
    tables with leading `|`s won't end up generating empty false `<td></td>`s,
    and that tables with trailing `|`’s won't end up getting those pipes
    included in the output.
    
*   [version 2.1.1.3](discount-2.1.1.3.tar.bz2) corrects a
    defect in smartypants processing that has been there since the
    beginning of time; I'd managed to misread the [reference smartypants
    documentation](http://daringfireball.net/projects/smartypants/) and
    thought that **1** dash made an `&ndash;` and **2** dashes made an
    `&mdash;` (it's actually 2 and 3 dashes respectively.) *John Foerch*
    read the documentation recently, noticed it's wrong, and sent me a note
    pointing it out. Whoops! But it's fixed now (as is the this page, which
    is, regrettably, the only documentation about smartypants.)
    
*   [version 2.1.1.2](discount-2.1.1.2.tar.bz2) corrects one small
    defect in block handling, plus changes the format of the output for
    failed tests in the test suite.
    
    The defect in block handling is that the reference implementation
    has text block absorbing adjacent code, so
    
	text
	text
	text
	~~~ code ~~~
    
    will generate
    
	<p>text text text code</p>
    
    instead of a paragraph of *text*(s) followed by `code`.
    
    The change in failed test output makes it output first the source
    code of the failed test, and then the *differences* between the
    expected output and the generated output.
    
*   [version 2.1.1.1](discount-2.1.1.1.tar.bz2) implements PHP markdown
    extra-style fenced code sections, where your chunks of code are surrounded
    by `~~~` lines instead of being indented 4 spaces.
    
    Fenced code sections are a *configuration* option, not a
    runtime option, and are enabled by using the `configure.sh` flag
    `--with-fenced-code` (**FENCED-CODE** in the version string).
    
    There are a few optimizations inside `markdown.c` to support
    fenced code detection without slowing the parser down, but those
    optimizations don't cause any of my test cases to fail. Version
    2.1.1 is still a fairly experimental release despite this, so take
    some care with it.
    
*   [version 2.1.0](discount-2.1.0.tar.bz2) cleans up a small collection
    of warts and build bugs, plus adds some fairly small enhancements to the
    `theme` and `makepage` programs:
    
    *   more modifications to `configure.sh` so that it generates *all*
    its test scripts in the build directory.
    *   more MacOS support in `configure.sh`; check to see if `.dSYM`
    folders are created when a test script is compiled, and, if so,
    don't forget to delete them when they’re cleaned up.
    *   `makepage` now accepts markdown option flags a'la the `markdown`
	program (via `-Fxxxx`, `-fname`, or in the `MARKDOWN_FLAGS`
	environment variable.)
    *   strip bitfields out of `opts[]` – I can't initialize a bitfield
	on plan9 cc.
    *   add a `-E` flag to `theme` to ignore context-sensitivity on
	`<?theme xxx?>` substitutions.

[even older versions of the code](older.html) are still available, but there is not
enough room in the margins to list them here.