build: avoid compiling self-hosted twice

build.zig: add a 'compile' step to compile the self-hosted compiler
without installing it.
Compilation: set cache mode to whole when using the LLVM backend and
--enable-cache is passed.

This makes `zig build` act the same as it does with stage1. Upside is
that a second invocation of `zig build` on an unmodified source tree
will avoid redoing the compilation again. Downside is that it will
proliferate more garbage in the project-local cache (same as stage1).

This can eventually be fixed when Zig's incremental compilation is more
robust; we can go back to having LLVM use CacheMode.incremental and rely
on it detecting no changes and avoiding doing the flush() step.

CMakeLists.txt

@@ -1045,7 +1045,7 @@ elseif(MINGW)
   target_link_libraries(zig2 ntdll)
 endif()
 
-set(ZIG_BUILD_ARGS "build"
+set(ZIG_BUILD_ARGS
     --zig-lib-dir "${CMAKE_SOURCE_DIR}/lib"
     "-Dconfig_h=${ZIG_CONFIG_H_OUT}"
     "-Denable-llvm"
@@ -1060,7 +1060,7 @@ set(ZIG_BUILD_ARGS "build"
 )
 
 add_custom_target(stage3 ALL
-    COMMAND zig2 ${ZIG_BUILD_ARGS}
+    COMMAND zig2 build compile ${ZIG_BUILD_ARGS}
     DEPENDS zig2
     COMMENT STATUS "Building stage3"
     WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"

build.zig

@@ -142,6 +142,10 @@ pub fn build(b: *Builder) !void {
     };
 
     const exe = b.addExecutable("zig", main_file);
+
+    const compile_step = b.step("compile", "Build the self-hosted compiler");
+    compile_step.dependOn(&exe.step);
+
     exe.stack_size = stack_size;
     exe.strip = strip;
     exe.sanitize_thread = sanitize_thread;

cmake/install.cmake

@@ -1,5 +1,8 @@
-set(ZIG_INSTALL_ARGS ${ZIG_BUILD_ARGS} --prefix "${CMAKE_INSTALL_PREFIX}")
-execute_process(COMMAND "${ZIG_EXECUTABLE}" ${ZIG_INSTALL_ARGS} WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}" RESULT_VARIABLE _result)
+set(ZIG_INSTALL_ARGS build ${ZIG_BUILD_ARGS} --prefix "${CMAKE_INSTALL_PREFIX}")
+execute_process(
+  COMMAND "${ZIG_EXECUTABLE}" ${ZIG_INSTALL_ARGS}
+  WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
+  RESULT_VARIABLE _result)
 
 if(_result)
     message("::")

src/Compilation.zig

@@ -1109,11 +1109,6 @@ pub fn create(gpa: Allocator, options: InitOptions) !*Compilation {
 
         const use_stage1 = options.use_stage1 orelse false;
 
-        const cache_mode = if (use_stage1 and !options.disable_lld_caching)
-            CacheMode.whole
-        else
-            options.cache_mode;
-
         // Make a decision on whether to use LLVM or our own backend.
         const use_llvm = build_options.have_llvm and blk: {
             if (options.use_llvm) |explicit|
@@ -1154,6 +1149,14 @@ pub fn create(gpa: Allocator, options: InitOptions) !*Compilation {
             }
         }
 
+        // 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_stage1 and !options.disable_lld_caching) or
+            (use_llvm and !options.disable_lld_caching)) CacheMode.whole else options.cache_mode;
+
         const tsan = options.want_tsan orelse false;
         // TSAN is implemented in C++ so it requires linking libc++.
         const link_libcpp = options.link_libcpp or tsan;

src/link/Elf.zig

@@ -1282,7 +1282,7 @@ fn linkWithLLD(self: *Elf, comp: *Compilation, prog_node: *std.Progress.Node) !v
     // linked are in the hash that namespaces the directory we are outputting to. Therefore,
     // we must hash those now, and the resulting digest will form the "id" of the linking
     // job we are about to perform.
-    // After a successful link, we store the id in the metadata of a symlink named "id.txt" in
+    // After a successful link, we store the id in the metadata of a symlink named "lld.id" in
     // the artifact directory. So, now, we check if this symlink exists, and if it matches
     // our digest. If so, we can skip linking. Otherwise, we proceed with invoking LLD.
     const id_symlink_basename = "lld.id";