Implement smart linking, take 3

contrib/bash_compl/_rgblink.bash

@@ -20,6 +20,7 @@ _rgblink_completions() {
 		[O]="overlay:glob-*.gb *.gbc *.sgb"
 		[o]="output:glob-*.gb *.gbc *.sgb"
 		[p]="pad:unk"
+		[s]="smart:unk"
 	)
 	# Parse command-line up to current word
 	local opt_ena=true

contrib/zsh_compl/_rgblink

@@ -18,6 +18,7 @@ local args=(
 	'(-o --output)'{-o,--output}"+[Write ROM image to this file]:rom file:_files -g '*.{gb,sgb,gbc}'"
 	'(-p --pad-value)'{-p,--pad-value}'+[Set padding byte]:padding byte:'
 	'(-S --scramble)'{-s,--scramble}'+[Activate scrambling]:scramble spec'
+	'(-s --smart)'{-s,--smart}'+[Perform smart linking from this section]:section name:'
 
 	'*'":object files:_files -g '*.o'"
 )

include/link/patch.hpp

@@ -3,6 +3,12 @@
 #ifndef RGBDS_LINK_PATCH_HPP
 #define RGBDS_LINK_PATCH_HPP
 
+#include <vector>
+
+struct Patch;
+struct Section;
+struct Symbol;
+
 /*
  * Checks all assertions
  * @return true if assertion failed
@@ -14,4 +20,12 @@ void patch_CheckAssertions();
  */
 void patch_ApplyPatches();
 
+/**
+ * Executes a callback on all sections referenced by a patch's expression
+ * @param patch The patch to scan the expression of
+ */
+void patch_FindReferencedSections(
+    Patch const &patch, void (*callback)(Section &), std::vector<Symbol> const &fileSymbols
+);
+
 #endif // RGBDS_LINK_PATCH_HPP

include/link/section.hpp

@@ -54,6 +54,7 @@ struct Section {
 	std::vector<Symbol> *fileSymbols;
 	std::vector<Symbol *> symbols;
 	std::unique_ptr<Section> nextu; // The next "component" of this unionized sect
+	bool smartLinked = false;       // Set to true if kept by smart linking
 };
 
 struct Assertion {
@@ -90,4 +91,11 @@ Section *sect_GetSection(std::string const &name);
  */
 void sect_DoSanityChecks();
 
+/**
+ * Adds a section as a new "root" of the smart link graph
+ */
+void sect_AddSmartSection(std::string const &name);
+
+void sect_PerformSmartLink();
+
 #endif // RGBDS_LINK_SECTION_HPP

include/link/symbol.hpp

@@ -48,6 +48,7 @@ void sym_AddSymbol(Symbol &symbol);
  */
 Symbol *sym_GetSymbol(std::string const &name);
 
+void sym_RemoveSymbol(std::string const &name);
 void sym_DumpLocalAliasedSymbols(std::string const &name);
 
 #endif // RGBDS_LINK_SYMBOL_HPP

man/rgblink.1

@@ -16,6 +16,7 @@
 .Op Fl o Ar out_file
 .Op Fl p Ar pad_value
 .Op Fl S Ar spec
+.Op Fl s Ar symbol
 .Ar
 .Sh DESCRIPTION
 The
@@ -104,6 +105,16 @@ See
 .Sx Scrambling algorithm
 below for an explanation and a description of
 .Ar spec .
+.It Fl s Ar sect_name , Fl Fl smart Ar sect_name
+This option specifies the name of a section that will be used as a starting point for smart linking; it may appear several times per
+.Nm
+invocation.
+See
+.Sx SMART LINKING
+below.
+If no section with that name is found,
+.Nm
+will error out.
 .It Fl t , Fl \-tiny
 Expand the ROM0 section size from 16 KiB to the full 32 KiB assigned to ROM.
 ROMX sections that are fixed to a bank other than 1 become errors, other ROMX sections are treated as ROM0.
@@ -191,6 +202,103 @@ to fix these so that the program will actually run in a Game Boy:
 Here is a more complete example:
 .Pp
 .Dl $ rgblink -o bin/game.gb -n bin/game.sym -p 0xFF obj/title.o obj/engine.o
+.Sh SMART LINKING
+Smart linking is a feature of
+.Nm
+that allows "trimming the fat" off of a ROM.
+It is enabled only if at least one
+.Fl s
+option is given on the command line.
+.Pp
+The smart linking process begins by finding all the
+.Ql referenced
+sections.
+A section is referenced if its name is specified using a
+.Fl s
+option, or if it is referred to by a referenced section.
+This definition applies recursively, so if section
+.Ql A
+is specified using
+.Fl s Va A ,
+section
+.Ql A
+references section
+.Ql B ,
+and section
+.Ql B
+references section
+.Ql C ,
+then all three sections
+.Ql A ,
+.Ql B ,
+and
+.Ql C
+are referenced.
+.Pp
+Sections refer to each other through expressions. For example:
+.Bd -literal -offset indent
+SECTION "A", ROM0
+    db Someplace
+    db BANK("C")
+SECTION "B", ROM0
+Someplace:
+SECTION "C", ROMX
+    db 42
+.Ed
+Here, section
+.Ql A
+references section
+.Ql B
+via the label
+.Ql Someplace ,
+and references section
+.Ql C
+via its name in
+.Ql BANK("C") .
+.Pp
+After all the referenced sections are found, all sections that were not referenced are deleted, and the linking process continues as normal.
+.Sy This should not cause any symbols not to be found, please report a bug (see Sx BUGS Ns Sy ) if this occurs.
+.Pp
+This is useful to detect
+.Dq unused
+sections, i.e. sections that contain data not used by anything.
+Typically, the section containing the header, including the entry point at
+.Ad $00:0100
+will be one of the starting sections; more exotic use cases may require more starting sections.
+It may be a good idea to start with the header as the only root, and if needed, add more root sections.
+.Pp
+Be careful, as numeric expressions do
+.Sy not
+cause references:
+.Bd -literal -offset indent
+DEF BASE_ADDR EQU $4000
+SECTION "A", ROM0
+    dw BASE_ADDR
+SECTION "B", ROMX[BASE_ADDR]
+    db 42
+.Ed
+Section
+.Ql A
+does
+.Sy not
+reference section
+.Ql B ,
+since
+.Va BASE_ADDR
+is a constant, and thus does not belong to section
+.Ql B .
+.Pp
+Finally, be careful that
+.Xr rgbasm 1
+tries to fill in data by itself to speed up
+.Nm Ap s
+work, which may cause
+.Nm
+not to see references to sections whose bank and/or address are fixed.
+It may be advisable to avoid fixing those (notably, to enable
+.Nm
+to improve section placement), but they can still be manually referenced using
+.Fl s .
 .Sh BUGS
 Please report bugs on
 .Lk https://github.com/gbdev/rgbds/issues GitHub .

src/link/assign.cpp

@@ -367,6 +367,9 @@ void assign_AssignSections() {
 
 	initFreeSpace();
 
+	// Check if we need to do smart linking and discard any sections
+	sect_PerformSmartLink();
+
 	// Generate linked lists of sections to assign
 	nbSectionsToAssign = 0;
 	sect_ForEach(categorizeSection);

src/link/main.cpp

@@ -130,7 +130,7 @@ void argErr(char flag, char const *fmt, ...) {
 }
 
 // Short options
-static char const *optstring = "dl:m:Mn:O:o:p:S:tVvWwx";
+static char const *optstring = "dl:m:Mn:O:o:p:S:s:tVvWwx";
 
 /*
  * Equivalent long options
@@ -152,6 +152,7 @@ static option const longopts[] = {
     {"output",        required_argument, nullptr, 'o'},
     {"pad",           required_argument, nullptr, 'p'},
     {"scramble",      required_argument, nullptr, 'S'},
+    {"smart",         required_argument, nullptr, 's'},
     {"tiny",          no_argument,       nullptr, 't'},
     {"version",       no_argument,       nullptr, 'V'},
     {"verbose",       no_argument,       nullptr, 'v'},
@@ -164,7 +165,7 @@ static void printUsage() {
 	fputs(
 	    "Usage: rgblink [-dMtVvwx] [-l script] [-m map_file] [-n sym_file]\n"
 	    "               [-O overlay_file] [-o out_file] [-p pad_value]\n"
-	    "               [-S spec] <file> ...\n"
+	    "               [-S spec] [-s symbol] <file> ...\n"
 	    "Useful options:\n"
 	    "    -l, --linkerscript <path>  set the input linker script\n"
 	    "    -m, --map <path>           set the output map file\n"
@@ -374,6 +375,9 @@ int main(int argc, char *argv[]) {
 		case 'S':
 			parseScrambleSpec(musl_optarg);
 			break;
+		case 's':
+			sect_AddSmartSection(musl_optarg);
+			break;
 		case 't':
 			is32kMode = true;
 			break;