Deduplicate entry freshness filesystem work

docs/wasm-runtime-architecture.md

@@ -209,6 +209,9 @@ The same snapshot divides pre-dispatch WASM readiness into entry normalization,
 mutation freshness, artifact stat, complete dependency freshness, and worker
 availability. `ready_check_count` distinguishes the warm one-check path from an
 on-demand compile and recheck; repeated snapshot reads retain the initial values.
+Freshness still stats every distinct source on every entry check. Exact repeated
+load paths are deduplicated before canonicalization, while distinct aliases are
+resolved independently so symlink retargets remain immediately visible.
 When a current serialized module exists, the worker scans wasm section headers
 and reads only `dylink.0` and `uce.abi`; it does not fault the multi-megabyte code
 and data bodies into every new worker. A missing/stale/invalid serialized module

scripts/test_dependency_invalidation.sh

@@ -83,6 +83,16 @@ assert_marker symlink-parent symlink-marker-a
 ln -sfn "symlink-target-b.uce" "$source_dir/symlink-child.uce"
 assert_marker symlink-parent symlink-marker-b
 
+# A repeated exact transitive load is signature-deduplicated before path
+# canonicalization, but the shared dependency must still invalidate the parent.
+printf '%s\n' '#ifndef UCE_DIAMOND_COMMON' '#define UCE_DIAMOND_COMMON' 'String diamond_marker() { return("diamond-a"); }' '#endif' >"$source_dir/diamond-common.uce"
+printf '%s\n' '#load "diamond-common.uce"' >"$source_dir/diamond-left.uce"
+printf '%s\n' '#load "diamond-common.uce"' >"$source_dir/diamond-right.uce"
+printf '%s\n' '#load "diamond-left.uce"' '#load "diamond-right.uce"' 'CLI(Request& context) { print(diamond_marker()); }' >"$source_dir/diamond-parent.uce"
+assert_marker diamond-parent diamond-a
+sed -i 's/diamond-a/diamond-b/' "$source_dir/diamond-common.uce"
+assert_marker diamond-parent diamond-b
+
 # HTTP entry units can resolve route/components dynamically. A changed dynamic
 # component must enter the demand-priority queue just like a changed entry unit;
 # otherwise a common dependency rebuild can leave the requested page stale for

src/lib/compiler.cpp

@@ -216,6 +216,11 @@ UnitSourceSignatureEntry compiler_unit_source_entry(String file_name, bool allow
 
 void compiler_append_unit_source_signature(String file_name, std::set<String>& visited, String& signature, bool allow_recent_stat)
 {
+	// The signature already deduplicates canonical files. Skip repeated exact
+	// load paths before resolving them again; distinct aliases are still
+	// canonicalized so symlink retargets remain visible on the next check.
+	if(visited.find(file_name) != visited.end())
+		return;
 	String normalized = file_name;
 	UnitSourceSignatureEntry entry;
 	bool recent = allow_recent_stat && compiler_recent_unit_source_entry(normalized, entry);
@@ -226,7 +231,11 @@ void compiler_append_unit_source_signature(String file_name, std::set<String>& v
 			normalized = file_name;
 	}
 	if(visited.find(normalized) != visited.end())
+	{
+		visited.insert(file_name);
 		return;
+	}
+	visited.insert(file_name);
 	visited.insert(normalized);
 
 	if(!recent)
@@ -244,9 +253,9 @@ String compiler_unit_source_signature(String file_name, bool allow_recent_stat =
 	return(gen_sha1(signature));
 }
 
-String compiler_unit_input_signature(Request* context, SharedUnit* su, bool allow_recent_source_stat = false)
+String compiler_unit_input_signature(Request* context, SharedUnit* su, bool allow_recent_source_stat = false, bool source_exists_known = false)
 {
-	if(!context || !su || !file_exists(su->file_name))
+	if(!context || !su || (!source_exists_known && !file_exists(su->file_name)))
 		return("");
 
 	String setup_template = context->server->config["COMPILER_SYS_PATH"] + "/" + context->server->config["SETUP_TEMPLATE"];
@@ -509,9 +518,12 @@ bool compiler_has_known_unit_cached(Request* context, String file_name)
 SharedUnitFilesystemState inspect_shared_unit_filesystem(Request* context, SharedUnit* su, bool allow_recent_source_stat = false)
 {
 	SharedUnitFilesystemState state;
-	state.source_exists = file_exists(su->file_name);
-	if(state.source_exists)
-		state.source_time = file_mtime(su->file_name);
+	struct stat source_stat;
+	if(stat(su->file_name.c_str(), &source_stat) == 0)
+	{
+		state.source_exists = true;
+		state.source_time = source_stat.st_mtime;
+	}
 	state.setup_template_time = file_mtime(
 		context->server->config["COMPILER_SYS_PATH"] + "/" +
 		context->server->config["SETUP_TEMPLATE"]
@@ -519,7 +531,7 @@ SharedUnitFilesystemState inspect_shared_unit_filesystem(Request* context, Share
 	state.compiler_abi_time = compiler_runtime_abi_time(context);
 	state.metadata_time = file_mtime(su->meta_file_name);
 	state.compile_output_time = file_mtime(su->compile_output_file_name);
-	state.current_input_signature = compiler_unit_input_signature(context, su, allow_recent_source_stat);
+	state.current_input_signature = compiler_unit_input_signature(context, su, allow_recent_source_stat, state.source_exists);
 	state.metadata_exists = (state.metadata_time != 0);
 	state.compile_output_exists = (state.compile_output_time != 0);
 	if(state.metadata_exists)

src/wasm/backend.cpp

@@ -129,7 +129,7 @@ static bool wasm_artifact_exists(Request* context, const String& entry_unit)
 	// metadata mismatches, which can leave stale wasm with old imports.
 	bool source_missing = false;
 	phase_started = time_precise();
-	if(compiler_unit_needs_recompile(context, entry_unit, &source_missing))
+	if(compiler_unit_needs_recompile(context, entry_unit, &source_missing, false, true))
 	{
 		context->stats.wasm_ready_freshness_us += (u64)((time_precise() - phase_started) * 1000000.0);
 		compiler_prioritize_unit(context, entry_unit);