swig/swig-4.3.0-Fix-precedence-of-casts.patch

56 lines
1.9 KiB
Diff

From 4e315cdd7c9072fd33ac1df3d208a0990b8231c3 Mon Sep 17 00:00:00 2001
From: Olly Betts <olly@survex.com>
Date: Sun, 27 Oct 2024 10:02:11 +1300
Subject: [PATCH] Fix precedence of casts
Casts should have the same high precedence as unary plus and minus, but
actually had a lower precedence than anything else.
This could lead to the wrong type being deduced in obscure cases, but
also prevented SWIG deducing a type for expressions such as (0)*1+2
which SWIG parses as a cast and then fixes up afterwards.
A bug fixed in 4.3.0 made this latter problem manifest more often
(previously type deduction happened to work for (0)*1+2 due to an
internal field not getting cleared properly).
Fixes #3058
Index: swig-4.3.0/Examples/test-suite/cpp11_auto_variable.i
===================================================================
--- swig-4.3.0.orig/Examples/test-suite/cpp11_auto_variable.i
+++ swig-4.3.0/Examples/test-suite/cpp11_auto_variable.i
@@ -62,3 +62,11 @@ static auto wstring_lit_len2 = sizeof("1
//static auto constexpr greeting = "Hello";
%}
+
+%inline %{
+/* Regression test for #3058 */
+auto CAST_HAD_WRONG_PRECEDENCE1 = (0)*1+2;
+auto CAST_HAD_WRONG_PRECEDENCE2 = (0)&1|2;
+auto CAST_HAD_WRONG_PRECEDENCE3 = (0)-1|2;
+auto CAST_HAD_WRONG_PRECEDENCE4 = (0)+1|2;
+%}
Index: swig-4.3.0/Source/CParse/parser.y
===================================================================
--- swig-4.3.0.orig/Source/CParse/parser.y
+++ swig-4.3.0/Source/CParse/parser.y
@@ -1793,7 +1793,6 @@ static String *add_qualifier_to_declarat
%token <str> DOXYGENSTRING
%token <str> DOXYGENPOSTSTRING
-%precedence CAST
%left QUESTIONMARK
%left LOR
%left LAND
@@ -1809,7 +1808,7 @@ static String *add_qualifier_to_declarat
%left LSHIFT RSHIFT
%left PLUS MINUS
%left STAR SLASH MODULO
-%precedence UMINUS NOT LNOT
+%precedence UMINUS NOT LNOT CAST
%token DCOLON
%type <node> program interface declaration swig_directive ;