Re: Using a single standalone-backend run in initdb (was Re: Bootstrap DATA is a pita)

For dollar strings, when a $ is found when reading SQL text (not an identifier/keyword), enter SQL_TEXT_DOLLAR. What comes next must be a parameter or the start of a dollar string. If the next char is a digit then it's a parameter so switch to NUMBER, since dollar-quoted string delims follow identifier rules and unquoted identifiers can't start with a number. Otherwise switch to DOLLAR_STRING_START_DELIM and consume until a $ is found or something illegal in an identifier is found. Or of course it could be lone $ which is bogus syntax but as far as we're concerned still just SQL_TEXT. Really, this is just looking for a dollar-quote start and doesn't care what it finds if it isn't a valid dollar-quote start.

If a valid dollar-quote delim is found switch to DOLLAR_STRING and read until we find the matching delim using a similar process, entering DOLLAR_STRING_DOLLAR, looking for param vs end delim, etc. When a full delimiter is read compare to the start delimiter and switch back to SQL_TEXT mode if it matches, otherwise remain in DOLLAR_STRING.

If an invalid dollar string delim was found switch back to SQL_TEXT (since it wasn't a valid beginning of a dollar string) and continue.

For QUOTED_IDENTIFIER_QUOTE and STANDARD_STRING_QUOTE, it found a " or ' while reading a quoted identifier or standard string and transitioned into that state. If the next char doubles the quote it'll return to reading the string; otherwise it'll return to the SQL_TEXT state since the identifier or literal has ended.

Similarly with ESCAPE_STRING_ESCAPE. Having found an escape, consume the next char even if it's a quote and return to the string parsing.

All this ignores psql backslash commands.

Have I missed anything really obvious? Does it seem useful to have more re-usable statement splitting code? Is there any sane justification for doing anything but extracting what psql does verbatim while carefully changing nothing?