shithub: pprolog

--- a/TODO

+++ b/TODO

@@ -1,10 +1,5 @@

 * Stop comparing strings all the time

 * Stop copying the entire goal stack into every choicepoint

-* Stop creating choicepoints when it is not needed

-* How to implement builtins nicely?

-* Right now we copy and allocate a lot, but almost never free stuff.

-* Many builtins should really throw an error, but they just fail for now.

-* Exceptions (throw, catch)

 * Modules (I try to do something like SWI prolog for now, but I know there is also an iso standard)

 The following builtins:

@@ -12,4 +7,24 @@

 unify_with_occurs_check/2: not implemented

 '\\='/2: handle occurs check

-* Handle overflow and uderflow in arithmetic

\ No newline at end of file

+* Handle overflow and uderflow in arithmetic

+* Variables should be compared and printed in a way that makes sense. Right now, their number is changed when a clause is used,

+  but their name remains the same, so prints will be bad.

+* Write a loader in prolog, that calls builtins to activate the C functions needed. This means we can do some syntax transformations

+  and handle directives in prolog code, instead of C. The "bootstrap" modules will have to be written in a way the original C loader

+  understands, but everything after that can use the full feature set.

+  Steps:

+	1) Change the lexer and parser so they carry around their state instead of global variables.

+	2) Make it possible to evaluate a query using 'evalquery' even when another query is running. This

+	   means no more global variables there either.

+	3) Create a "loading context" which is indexed by a prolog term and consists of:

+		* Lexer + parser state

+		* module loading state

+	4) Loading then becomes:

+		1) Create loading context

+		2) Open file

+		3) Repeatedly read terms using read_term

+		4) Inspect the term (clause or directive?) and act on it, using builtints and reference to context

+		5) Close the file and context