Commit 9a92f3d290694bfefbc7d71b5ba1823edb6c547f | zig

author	Luuk de Gram <luuk@degram.dev> 2022-08-29 20:13:55+02:00
committer	Luuk de Gram <luuk@degram.dev> 2022-09-07 18:53:12+02:00
log	9a92f3d290694bfefbc7d71b5ba1823edb6c547f
tree	a2b6759748a3c0d561d0bccdcbe43f14e7d20c4b
parent	924679abc46deeaae9284ab6ce928aaddb0fae95
signature	Commit is signed but in an unrecognized format.

wasm/Object: parse debug sections into reloc data

Rather than storing the name of a debug section into the structure
`RelocatableData`, we use the `index` field as an offset into the
debug names table. This means we do not have to store an extra 16 bytes
for non-debug sections which can be massive for object files where each
data symbol has its own data section. The name of a debug section
can then be retrieved again when needed by using the offset and
then reading until the 0-delimiter.

2 files changed, 45 insertions(+), 8 deletions(-)

src
- link
  - Wasm.zig
  - Wasm
    Object.zig

src/link/Wasm.zig+1-1

...	...	@@ -1782,7 +1782,7 @@ pub fn getMatchingSegment(self: *Wasm, object_index: u16, relocatable_index: u32
1782	1782	});
1783	1783	break :blk index;
1784	1784	},
1785		.custom => return error.@"TODO: Custom section relocations for wasm",
	1785	.debug => return error.@"TODO: Custom section relocations for wasm",
1786	1786	}
1787	1787	}
1788	1788

src/link/Wasm/Object.zig+44-7

...	...	@@ -63,16 +63,21 @@ relocatable_data: []const RelocatableData = &.{},
63	63	/// import name, module name and export names. Each string will be deduplicated
64	64	/// and returns an offset into the table.
65	65	string_table: Wasm.StringTable = .{},
	66	/// All the names of each debug section found in the current object file.
	67	/// Each name is terminated by a null-terminator. The name can be found,
	68	/// from the `index` offset within the `RelocatableData`.
	69	debug_names: [:0]const u8,
66	70
67	71	/// Represents a single item within a section (depending on its `type`)
68	72	const RelocatableData = struct {
69	73	/// The type of the relocatable data
70		type: enum { data, code, custom },
	74	type: enum { data, code, debug },
71	75	/// Pointer to the data of the segment, where its length is written to `size`
72	76	data: [*]u8,
73	77	/// The size in bytes of the data representing the segment within the section
74	78	size: u32,
75		/// The index within the section itself
	79	/// The index within the section itself, or in case of a debug section,
	80	/// the offset within the `debug_names` table.
76	81	index: u32,
77	82	/// The offset within the section where the data starts
78	83	offset: u32,
...	...	@@ -96,7 +101,7 @@ const RelocatableData = struct {
96	101	return switch (self.type) {
97	102	.data => .data,
98	103	.code => .function,
99		.custom => .section,
	104	.debug => unreachable, // illegal, debug sections are not represented by a symbol
100	105	};
101	106	}
102	107	};
...	...	@@ -111,6 +116,7 @@ pub fn create(gpa: Allocator, file: std.fs.File, name: []const u8, maybe_max_siz
111	116	var object: Object = .{
112	117	.file = file,
113	118	.name = try gpa.dupe(u8, name),
	119	.debug_names = &.{},
114	120	};
115	121
116	122	var is_object_file: bool = false;
...	...	@@ -197,6 +203,11 @@ pub fn importedCountByKind(self: *const Object, kind: std.wasm.ExternalKind) u32
197	203	} else i;
198	204	}
199	205
	206	/// From a given `RelocatableDate`, find the corresponding debug section name
	207	pub fn getDebugName(self: *const Object, relocatable_data: RelocatableData) []const u8 {
	208	return std.mem.sliceTo(self.debug_names[relocatable_data.index..], 0);
	209	}
	210
200	211	/// Checks if the object file is an MVP version.
201	212	/// When that's the case, we check if there's an import table definiton with its name
202	213	/// set to '__indirect_function_table". When that's also the case,
...	...	@@ -328,10 +339,15 @@ fn Parser(comptime ReaderType: type) type {
328	339
329	340	self.object.version = version;
330	341	var relocatable_data = std.ArrayList(RelocatableData).init(gpa);
331
332		errdefer while (relocatable_data.popOrNull()) \|rel_data\| {
333		gpa.free(rel_data.data[0..rel_data.size]);
334		} else relocatable_data.deinit();
	342	var debug_names = std.ArrayList(u8).init(gpa);
	343
	344	errdefer {
	345	while (relocatable_data.popOrNull()) \|rel_data\| {
	346	gpa.free(rel_data.data[0..rel_data.size]);
	347	} else relocatable_data.deinit();
	348	gpa.free(debug_names.items);
	349	debug_names.deinit();
	350	}
335	351
336	352	var section_index: u32 = 0;
337	353	while (self.reader.reader().readByte()) \|byte\| : (section_index += 1) {
...	...	@@ -352,6 +368,24 @@ fn Parser(comptime ReaderType: type) type {
352	368	try self.parseRelocations(gpa);
353	369	} else if (std.mem.eql(u8, name, "target_features")) {
354	370	try self.parseFeatures(gpa);
	371	} else if (std.mem.startsWith(u8, name, ".debug")) {
	372	const debug_size = @intCast(u32, reader.context.bytes_left);
	373	const debug_content = try gpa.alloc(u8, debug_size);
	374	errdefer gpa.free(debug_content);
	375	try reader.readNoEof(debug_content);
	376
	377	const debug_name_index = @intCast(u32, debug_names.items.len);
	378	try debug_names.ensureUnusedCapacity(name.len + 1);
	379	debug_names.appendSliceAssumeCapacity(try gpa.dupe(u8, name));
	380	debug_names.appendAssumeCapacity(0);
	381	try relocatable_data.append(.{
	382	.type = .debug,
	383	.data = debug_content.ptr,
	384	.size = debug_size,
	385	.index = debug_name_index,
	386	.offset = len - debug_size,
	387	.section_index = section_index,
	388	});
355	389	} else {
356	390	try reader.skipBytes(reader.context.bytes_left, .{});
357	391	}
...	...	@@ -517,6 +551,9 @@ fn Parser(comptime ReaderType: type) type {
517	551	else => \|e\| return e,
518	552	}
519	553	self.object.relocatable_data = relocatable_data.toOwnedSlice();
	554
	555	const names = debug_names.toOwnedSlice();
	556	self.object.debug_names = names[0 .. names.len - 1 :0];
520	557	}
521	558
522	559	/// Based on the "features" custom section, parses it into a list of