A new release of CG-3 has been tagged v0.9.9.10800. Performance: - Rules application is now 60% faster than previous release. The reference test ( http://visl.sdu.dk/cg3_performance.html ) went from 390 cohorts/sec to 620, and performs 1/3rd the number of re-allocations. This was accomplished by taking advantage of an internal re-structuring of using sequential set numbers instead of their hashes, which in turn was initially implemented to eliminate some hash collisions. - EXTERNAL processing is now 20 times faster than previous release.
New features: - Binary grammars are now kept forward compatible, to the best of my ability. Grammars compiled all the way back to r10043 will continue to load on latest versions. - Added cmdline option --quiet and alias -v0 - Grammar compiling diagnostics now print what file the message is from, alongside better context - Added STRICT-TAGS to enforce a basic tag set. See http://visl.sdu.dk/cg3/chunked/tags.html#strict-tags - Added OPTIONS to enforce stricter cleaner grammars. See http://visl.sdu.dk/cg3/chunked/grammar.html#grammar-options - Added cg-conv cmdline options for handling weights and sub-readings (--wfactor, --wtag, --sub-delim) - Added REOPEN-MAPPINGS to let a grammar reuse a prefix for chosen tags. See http://visl.sdu.dk/cg3/chunked/grammar.html#reopen-mappings - Added magic set _SAME_BASIC_ to check for identical readings. See http://visl.sdu.dk/cg3/chunked/sets.html#set-same-basic - Added context modifier 'f' to automatically create an alternative branch where numeric tags don't apply. See http://visl.sdu.dk/cg3/chunked/contexts.html#test-branch - Added SUB:* to match any sub-reading. Can currently only be used for Select/Remove-like actions. - Added cmdline flag --split-mappings to keep mapping tags on a reading of their own in the output. Note that you MUST NOT treat this format as different from the merged format - they are logically identical. Changes: - cg-conv now auto-detects Apertium better - Tag prefix ! (inversion) removed in favour of ^ (fail-fast) - Empty composite tag is now an error - Readings that would be considered identical without --trace are now marked deleted under --trace. This is incorrect behavior, but being blessed because it makes things easier for grammarians. - Regex captures are now done on a true per-reading basis, which fixes a lot of quirks - Non-capturing unification is now also done on a true per-reading basis, and ditto - Cmdline flag --prefix must now match the loaded binary grammar's prefix - The Emacs major mode is now named cg.el again - CR and NEL are no longer considered separate newlines, which fixes line numbers for DOS/Windows files Fixed Bugs: - Substitute out-of-order tags no longer crashes - Mapping tags created via varstring now work like mapping tags - Regex captures no longer remain after a failed context - Fixed partial wordform parsing in the input stream - Fixed empty sections that caused --sections to skip last section - Fixed sub-readings remaining active in certain situations - Fixed parsing the set SUB as the SUB: flag - Fixed a minor memory leak regarding regexes - Fixed dependencies looking at removed cohorts and subsequently crashing Main site is http://visl.sdu.dk/cg3.html Google Group is http://groups.google.com/group/constraint-grammar Source snapshots available at http://visl.sdu.dk/download/cg3/ Windows binary is at http://apertium.projectjj.com/win32/nightly/ OS X binary is at http://apertium.projectjj.com/osx/nightly/ RHEL/Fedora/CentOS/OpenSUSE packages are at http://apertium.projectjj.com/rpm/howto.txt Debian/Ubuntu packages are at http://apertium.projectjj.com/apt/howto.txt I strongly recommend everyone uses STRICT-TAGS in their grammars. It greatly reduces the risk of typos. -- Tino Didriksen CG-3 Developer -- You received this message because you are subscribed to the Google Groups "Constraint Grammar" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/constraint-grammar. For more options, visit https://groups.google.com/d/optout.
