Le 17 mars 2010 à 20:34, Joel E. Denny a écrit : >> Yes, but I'm no big fan of having several outputs here. Joel might have >> another opinion. Maybe gnulib could support a factored means to support >> both outputs. But the xsl-based approach you recently proposed seems a >> more acceptable path. > > We plan for the XML output to be a stable user-visible interface to the > parser tables, so XSLT should be easier to maintain than internal Bison > code to support VCG. > > More generally, I see no reason to reject any decent XSLT contribution > that might be useful to Bison users. Maybe there should be some sort of > data/xslt/contrib for XSLT that is not considered part of the core Bison > feature set. For any contributed XSLT, there must be test cases to make > sure we don't break it as the XML output evolves. Maybe > tests/xslt-contrib.at.
That sounds perfect.
