Changelog for skylighting-0.4.2

Revision history for skylighting

HTML output: remove outer div. This prevented margin collapsing between the pre and surrounding block elements, and often gave us excess white space around code blocks. See jgm/pandoc#3996.

Updated XML definitions from KDE repository. Changed: actionscript, ada, agda, alert, alert_indent, asn1, awk, bash, boo, c, changelog, clojure, cmake, coldfusion, commonlisp, cpp, cs, css, curry, d, dockerfile, dot, doxygen, doxygenlua, eiffel, email, erlang, fortran, fsharp, gcc, haskell, haxe, isocpp, java, javascript, jsp, julia, lilypond, lua, m4, makefile, matlab, maxima, mips, modelines, modula-2, monobasic, objectivec, objectivecpp, ocaml, octave, opencl, pascal, perl, php, pike, postscript, prolog, purebasic, python, r, relaxng, relaxngcompact, rest, rhtml, ruby, rust, scala, scheme, sci, sql-mysql, sql-postgresql, sql, tcl, tcsh, verilog, vhdl, xml, xslt, xul, zsh.
Added support for powershell, using definition from KDE repository.
Stop terminating long builds on TravisCI (Kyle Ondy).

Removed ToJSON/FromJSON instances for KeywordAttr, WordSet, Matcher, Context, ContextSwitch, Rule, Syntax (added in 0.3.5). Creating these increased the memory requirements for compiling skylighting to a degree not justified by the usefulness of these instances.

Added ToJSON/FromJSON instances for all basic types.
Added some background colors to 'kate' style, matching default.theme, and made FromJSON for Style sensitive to background-color.

HTML formatting changes (David Baynard). Note that these changes may require changes in hard-coded CSS that is used with skylighting's HTML output.
- Wrap lines of source code in a div with display set to inline-block. The divs make per-line processing easier. They cannot be set as display: block as that introduces extra new lines when copying and pasting.
- Render line numbers in html using CSS pseudo elements rather than a table. The line numbers are always produced, as data-line-number attributes, and css to display them as ::before elements are always produced. The option to switch on line numbering only toggles a class; this means it is possible to toggle line numbering without re-running skylighting.
- If the linkAnchors option is set, wrap with an a element rather than a div, set so that clicking the line number (and only the line number) will jump to that line.
- Code wraps by default when html is printed, with wrapped lines indented.

Revert change from 0.3.2; now entities from a DTD are resolved if the DTD element is included. This is needed for agda.xml and some others. If you want to use a custom xml definition without needing language.dtd, just remove the reference to language.dtd in your xml file.
Added case keyword for ats (Hanwen (Steinway) Wu).

Parse xml definitions without resolving entities from language.dtd. This allows xml definitions to be used even when language.dtd isn't available. Existing definitions don't rely on the two or three entities definied in language.dtd, so this is harmless.
Small optimizations.

Use handwritten parser for Float instead of regex. Fixes jgm/highlighting-kate#57 (again).
Added float parsing tests.
Use parsec parsers, not regex, for CStringChar and CChar.
Rewrote Int, Hex, Oct, CStringChar, CCHar parsers with parsec instead of regex. This speeds things up.
Don't include leading + in Int, Hex, Oct parsers. That gives bad results for things like i+1.

Store literal binary-encoded representations of the Syntax structure in Skylighting.Syntax.*.hs, instead of the structure itself. This works around a problem ghc has compiling modules with large structures. This indirect method produces faster compile times and avoids massive memory usage by ghc (especially in profiling builds). For background see http://stackoverflow.com/questions/16348340/. Closes #7.
Types: Internals of 'WordSet' are no longer exposed.

Restore Data, Typeable, Generic instances for basic types. To do this, we remove reCompiled field from RE (API change). Instead, we have the tokenizer compile regexes the first time it encounters them, memoizing the results. Performance is not significantly worse.
Skylighting.Syntax.*: use string representation of the Syntax, which is then 'read', rather than including the code for the data structure directly (#7). This indirect method produces faster compile times and avoids massive memory usage by ghc (especially in profiling builds). For background see http://stackoverflow.com/questions/16348340/compiling-very-large-constants-with-ghc
Use -fprof-auto-exported for profiling builds.
Added benchmark for xml syntax definition parsing.
Patched perl.xml (improperly escaped regex) (#8). This fixes a perl regex compilation error for regex patterns with backslash as delimiter, e.g. m\a.
Add a flag to build with system pcre (Felix Yan).

Avoid depending on a PrintfArg instance for Text (#5). This isn't provided in some older versions of the text library. This change should allow the package to build on older platforms.

Properly escape characters in subDynamic. This fixes a problem that caused failures with
```
echo -e "s\0b\0c" | skylighting -s perl
```
More efficient rewrite of char escaping for regexes.

Increase test timeout to 6s.
Avoid double {{}} in latex macros.
Fixed problem compiling Format.LaTeX on older ghc versions (7.8.3) that can't take a Text as PrintfArg.

Added much more extensive testing to ensure that tokenizers don't hang or drop input.
Fixed some issues with line-end and fallthrough context handling.
Fixed a bug in WordDetect handling which caused it to drop input (#2).

Added breezeDark style, from breeze-dark.theme.
Fixed performance bug in regex application (#1). This gives a significant speedup, especially for inputs with long lines.