Multiple rhs (#25)

* notes

* more hacking

* more hacking

* more argh

* hackikng away

* stuff

* start rewriting all pattern matching

* pattern match rewrite almost done

* pattern match rewrite passes the first test

* codegen for multiple rhs passes first test

* note another bug

* note codegen after parse unit

* getting closer

* next multidef example works

* make later defs local to main def

to avoid top level mutual recursion

* multiple defs test case three works

* add some tests

* detect redefinition across parse units

* case three works

* fix up filemodel merge

* fix access to nested structs

* more debugging

* omg fixed pattern defs

* more fixes

* fix local multiple defs

* tiny

* fix Vector ++ Image

* fix Image op Vector

* add more tests for multiple RHS

* ready for 9.0.14 release

* fix date

Author: jcupitt

CHANGELOG.md

@@ -1,5 +1,11 @@
 ## master
 
+- pattern matching rewrite
+- add multiple defs and parameter pattern match to compiler
+- add filemodel::new_from_file, rework filemodel_open
+- reloading a toolkit now deletes all defs, then does load ... adding a new
+  def to a func across a parse unit is now an error
+
 ## 9.0.14 2025/10/25
 
 - fix load cancel

TODO

@@ -1,38 +1,54 @@
-- should allow vips_image ++ Image
+- nicholas's bug report
 
-- remove remove prefs workspace stuff?
+- lcomps leave stray unreffed symbols in "list unresolved"
 
-    things like max heap size should be settable ... maybe pres are useful?
+    Image_transform_item.Resize_item.Size_within_item.action.within ($SAVEDIR/start/Image.def:509)  refers to undefined symbol h
+    Image_transform_item.Resize_item.Size_within_item.action.within ($SAVEDIR/start/Image.def:509)  refers to undefined symbol w
 
-- add multiple definitions, finish function argument destructuring
+- rework pattern matching in lcomps
 
-    sym has a field for "next definition", initially NULL
+- what does the programming window do for multiple defs?
 
-`   add_defining looks for an existing sym with this name, if it finds
-    one, add a new sym called "{name}-$$4" or whatever
+    probably need to follow next_def and concatenate them all
 
-    chase to the end of "next definition" on the existing sym, append our
-    new sym
+- dir should hide $$vars
 
-    during compile, generate code like
+- programming window is showing generated syms, like $$pattern_lhs0?
 
-        fred a b c
-            = destructured_fred, args_match
-            = fred-$$1 a b c
-        {
-            destructured_fred = rhs of fred
-            args_match = a_matches && b_matches && c_matches
-        }
+- ban patterns in class parameters
+
+    or could we allow multiple class defs?
+
+        Fred (Image x) = class { ... };
 
-    if a func has many RHS and the last RHS uses destrucuring, generate a
-    final def with
+    probably only useful for trivial classes
+
+- what about
+
+    fred (list x) = ...;
+    fred (complex x) = ...;
+
+  ie. allow names of builtin types as well as class names
+
+  maybe:
 
-        fred a b c
-            = error "no def of fred matches args a b c";
+    fred (is_list x) = ...;
 
-    - error if more than one def of fred has no destructuring
-    - error if defs don't all have the same number of args
-    - error if a def with no destructuring isn't the last def
+  ie. a predicate before the arg, as well as a class name?
+
+- do we allow eg.
+
+    fred [a, b ..] = a + b;
+
+  equivalent to
+
+    fred a:b:x = a + b;
+
+
+
+- remove remove prefs workspace stuff?
+
+    things like max heap size should be settable ... maybe pres are useful?
 
 - try < > in the image titlebar
 

meson.build

@@ -1,6 +1,6 @@
 project('nip4', 'c',
   # ie. a major after nip2 8.9 for workspace save file versioning
-  version: '9.0.14',
+  version: '9.0.15',
   license: 'GPL',
   meson_version: '>=0.64',
   default_options: [