Pavel Stehule wrote: > > I used your proposed way based on Restarget
Thanks. Some more tweaking to go yet before I consider this committable, but it's much better now. Here's v28. I changed a few things:
- make expression evaluation code more orthodox: 1. avoid PG_TRY, use a ExprContext shutdown callback instead 2. use a "Fast" evaluator, for calls past the first one 3. don't look up fmgrinfos until execution time 4. don't duplicate get_expr_result_type - make parser accept DEFAULT namespace. Only xml implementation barfs. (this means we lost the errposition pointer, but I don't really care. We could fix it if we cared) - clean up parse analysis code a little bit - move decls/struct defs to better locations in source code - remove leftover "namespaces" in TableExprState - pgindent the whole mess.
I checked the changes and looks correct - although for some I had not courage :) - like dynamic change of exprstate->evalfunc
I fixed test, and append forgotten header file
I don't like the xml.c code and the "evalcols" flag. That's next on my list to fix.
You need some flag to specify if column paths are valid or not.
I don't think to_xmlstr() is necessary, considering xml_text2xmlChar. We could just apply a cast of the source cstring to xmlChar.