Start a changelog

CHANGES.rst

@@ -0,0 +1,106 @@
+CHANGELOG
+=========
+
+2.0.0
+~~~~~
+
+Date: 2015-05-09
+
+- Rewrite the parser engine. A few comments:
+
+  The new code base uses C++11 to do more work in much fewer lines of code.
+  The number of lines of code has decreased from about 1600 to 900 while at
+  the same time increasing the amount of guiding comments.
+
+  Previously, the parser code was essentially a hard-to-grap state machine
+  (which b.t.w. could also have been implemented using only flex). This has
+  been remedied. The code complexity has decreased by a lot and the process
+  is severed into distinct parsing phases. This makes the code a lot easier
+  to understand and modify.
+
+  Downside: the code base is now fully dependent on C++11 which requires a
+  recent compiler.
+
+- Remove some arbitrary restrictions in the language:
+
+  - You can now use the shorthand notation 'A / B,C / D' to define markov
+    chain with a link being a list of variables. The same limitation was
+    overcome for the function-of syntax 'A:B' and the mutual independence
+    statement 'A.B.C'.
+
+  - You can now use inequalities as constraints
+
+  - You can now use information expressions with arbitrary constants such
+    as 'H(X) >= 1'. This is probably not very useful, but then again, there
+    was no real reason not to allow it (or I didn't think of it, yet).
+
+- Don't not create temporary files anymore. Flex supports parsing in-memory
+  data which is a clear winner compared to needless filesystem access.
+
+- use the lexer to split the input stream into tokens.
+
+  Previously, the lexer was only used to do some obscur preprocessing the
+  purpose of which is still completely unclear to me. In fact, I find it
+  likely that this separate usage of the lexer could lead to inconsistent
+  behaviour in some cases.
+
+- Variable names cannot contain white spaces any longer. This notation is
+  now treated as a syntax error.
+
+  In fact, in the face of the grammar definition, I believe this was an
+  unintended side effect the hand-written yylex function just stripping
+  away all whitespace. For example, 'H(X Y)' would be equivalent to 'H(XY)'
+  but not to 'H(X,Y)', while their was a grammar rule trying to achieve the
+  inverse.
+
+- Always perform stack unwinding in the main function
+
+- Slightly improve behaviour in the face of some errors
+
+- I am currently missing (at least?) one optimization that was present in
+  Xitip: collapsing sets of variables that only appear together.
+
+
+1.1.0
+~~~~~
+
+Date: 2015-04-18
+
+- Port to GLPK
+- Fix behaviour when checking for basic information measure (the
+  mis-feature was inherited from Xitip)
+- obsessive style fixes
+- Reenable 'A:B' syntax (was broken due to my ignorance)
+
+
+1.0.0 (now Citip)
+~~~~~~~~~~~~~~~~~
+
+Date: 2015-04-01
+
+- Replace the GTK frontend by a simple CLI.
+- Thus eliminate build time and run time dependencies on GTK, etc.
+- Start a proper README
+
+
+Xitip (unspecified version)
+~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Date: 2008-07-03
+
+This is the version of Xitip as could be downloaded in a .tar.gz archive
+from their website in early 2015.
+
+Unfortunately, I couldn't get into contact with any of the Xitip project
+authors (3 dead emails!), so I don't have access to their version
+history nor development branches or version number.
+
+The release date guess (03 July 2008) is based on the most recent mtime
+among the extracted source files.
+
+I excluded the binary and temporary files and the .svn metadata folder
+contained in the upstream tarball for obvious reasons.
+
+Furthermore, I excluded the qsopt files. These should be downloaded by
+the user separately (which is on 64bit distributions necessary anyway).
+Though, the main reason not to include these files are the license issues.