CLI: fix logic for sending output file path

via the compiler protocol
This commit is contained in:
Andrew Kelley 2023-12-24 18:30:38 -07:00
parent e12e296306
commit 9141e1a29c
2 changed files with 68 additions and 24 deletions

View File

@ -1198,16 +1198,6 @@ pub fn create(gpa: Allocator, options: InitOptions) !*Compilation {
const use_llvm = options.config.use_llvm;
// TODO: once we support incremental compilation for the LLVM backend via
// saving the LLVM module into a bitcode file and restoring it, along with
// compiler state, the second clause here can be removed so that incremental
// cache mode is used for LLVM backend too. We need some fuzz testing before
// that can be enabled.
const cache_mode = if ((use_llvm or !have_zcu) and !options.disable_lld_caching)
CacheMode.whole
else
options.cache_mode;
const any_unwind_tables = options.config.any_unwind_tables;
const link_eh_frame_hdr = options.link_eh_frame_hdr or any_unwind_tables;
@ -1560,7 +1550,7 @@ pub fn create(gpa: Allocator, options: InitOptions) !*Compilation {
.compatibility_version = options.compatibility_version,
.each_lib_rpath = each_lib_rpath,
.build_id = build_id,
.disable_lld_caching = options.disable_lld_caching or cache_mode == .whole,
.disable_lld_caching = options.disable_lld_caching or options.cache_mode == .whole,
.subsystem = options.subsystem,
.hash_style = options.hash_style,
.enable_link_snapshots = options.enable_link_snapshots,
@ -1576,7 +1566,7 @@ pub fn create(gpa: Allocator, options: InitOptions) !*Compilation {
.entry_addr = null, // CLI does not expose this option (yet?)
};
switch (cache_mode) {
switch (options.cache_mode) {
.incremental => {
// Options that are specific to zig source files, that cannot be
// modified between incremental updates.

View File

@ -3110,6 +3110,21 @@ fn buildOutputType(
else => false,
};
const disable_lld_caching = !output_to_cache;
const cache_mode: Compilation.CacheMode = b: {
if (disable_lld_caching) break :b .incremental;
if (!create_module.resolved_options.have_zcu) break :b .whole;
// TODO: once we support incremental compilation for the LLVM backend
// via saving the LLVM module into a bitcode file and restoring it,
// along with compiler state, this clause can be removed so that
// incremental cache mode is used for LLVM backend too.
if (create_module.resolved_options.use_llvm) break :b .whole;
break :b .incremental;
};
gimmeMoreOfThoseSweetSweetFileDescriptors();
const comp = Compilation.create(gpa, .{
@ -3211,7 +3226,8 @@ fn buildOutputType(
.test_filter = test_filter,
.test_name_prefix = test_name_prefix,
.test_runner_path = test_runner_path,
.disable_lld_caching = !output_to_cache,
.disable_lld_caching = disable_lld_caching,
.cache_mode = cache_mode,
.subsystem = subsystem,
.debug_compile_errors = debug_compile_errors,
.enable_link_snapshots = enable_link_snapshots,
@ -3942,7 +3958,7 @@ fn serve(
const hdr = try server.receiveMessage();
switch (hdr.tag) {
.exit => return,
.exit => return cleanExit(),
.update => {
assert(main_progress_node.recently_updated_child == null);
tracy.frameMark();
@ -4104,25 +4120,63 @@ fn serveUpdateResults(s: *Server, comp: *Compilation) !void {
try s.serveErrorBundle(error_bundle);
return;
}
// This logic is a bit counter-intuitive because the protocol implies that
// each emitted artifact could possibly be in a different location, when in
// reality, there is only one artifact output directory, and the build
// system depends on that fact. So, until the protocol is changed to
// reflect this, this logic only needs to ensure that emit_bin_path is
// emitted for at least one thing, if there are any artifacts.
if (comp.bin_file) |lf| {
const emit = lf.emit;
// This logic is counter-intuitive because the protocol accounts for each
// emitted artifact possibly being in a different location, which correctly
// matches the behavior of the compiler, however, the build system
// currently always passes flags that makes all build artifacts output to
// the same local cache directory, and relies on them all being in the same
// directory.
//
// So, until the build system and protocol are changed to reflect this,
// this logic must ensure that emit_bin_path is emitted for at least one
// thing, if there are any artifacts.
switch (comp.cache_use) {
.incremental => if (comp.bin_file) |lf| {
const full_path = try lf.emit.directory.join(gpa, &.{lf.emit.sub_path});
defer gpa.free(full_path);
try s.serveEmitBinPath(full_path, .{
.flags = .{ .cache_hit = comp.last_update_was_cache_hit },
});
return;
},
.whole => |whole| if (whole.bin_sub_path) |sub_path| {
const full_path = try comp.local_cache_directory.join(gpa, &.{sub_path});
defer gpa.free(full_path);
try s.serveEmitBinPath(full_path, .{
.flags = .{ .cache_hit = comp.last_update_was_cache_hit },
});
return;
},
}
for ([_]?Compilation.Emit{
comp.docs_emit,
comp.implib_emit,
}) |opt_emit| {
const emit = opt_emit orelse continue;
const full_path = try emit.directory.join(gpa, &.{emit.sub_path});
defer gpa.free(full_path);
try s.serveEmitBinPath(full_path, .{
.flags = .{ .cache_hit = comp.last_update_was_cache_hit },
});
} else if (comp.docs_emit) |emit| {
const full_path = try emit.directory.join(gpa, &.{emit.sub_path});
return;
}
for ([_]?Compilation.EmitLoc{
comp.emit_asm,
comp.emit_llvm_ir,
comp.emit_llvm_bc,
}) |opt_emit_loc| {
const emit_loc = opt_emit_loc orelse continue;
const directory = emit_loc.directory orelse continue;
const full_path = try directory.join(gpa, &.{emit_loc.basename});
defer gpa.free(full_path);
try s.serveEmitBinPath(full_path, .{
.flags = .{ .cache_hit = comp.last_update_was_cache_hit },
});
return;
}
}