v3: snapshot generated-C report sources

Author: medvednikov

cmd/v/macos_v3_dispatch.c.v

@@ -37,6 +37,7 @@ const macos_v3_c_error_compiler_file = 'compiler'
 const macos_v3_c_error_output_file = 'output'
 const macos_v3_c_error_source_name_file = 'source_name'
 const macos_v3_c_error_v_sources_file = 'v_sources'
+const macos_v3_c_error_v_source_digests_file = 'v_source_digests'
 // optional marker file inside a staged report dir; when it holds
 // `macos_v3_compiler_error_fallback` the report describes a V3 internal compiler
 // error (V source only) instead of a generated-C compilation error.
@@ -314,7 +315,8 @@ fn retry_macos_v3_with_old_compiler(caller_environment map[string]string, fallba
 		// filing a report. Still forward a notice-only marker so that once the stable
 		// build succeeds the user sees the documented fallback notice (doc/docs.md) rather
 		// than a silent switch that is indistinguishable from a direct V3 success.
-		export_macos_v3_report_content(macos_v3_inline_asm_fallback, 'v3', '', '', [])
+		export_macos_v3_report_content(macos_v3_inline_asm_fallback, 'v3', '', '',
+			map[string]string{})
 		os.rmdir_all(c_error_dir) or {}
 		if should_report {
 			eprintln('V3 requested the compatibility compiler for inline assembly')
@@ -349,7 +351,7 @@ fn take_macos_v3_report_content() ?MacosV3CErrorReport {
 // export_macos_v3_report_content bounds the fallback source in THIS process — which staged
 // the report and therefore trusts `c_file` — and forwards only that content to the V1
 // retry through the environment.
-fn export_macos_v3_report_content(kind string, ccompiler string, c_output string, c_file string, v_sources []string) {
+fn export_macos_v3_report_content(kind string, ccompiler string, c_output string, c_file string, v_sources map[string]string) {
 	v_file, v_source := builder.bounded_v3_fallback_source(kind, c_output, c_file, v_sources)
 	export_macos_v3_bounded_report_content(kind, ccompiler, c_output, v_file, v_source)
 }
@@ -426,7 +428,7 @@ struct MacosV3StagedReport {
 	ccompiler  string
 	c_output   string
 	c_file     string
-	v_sources  []string
+	v_sources  map[string]string
 	report_dir string
 }
 
@@ -462,7 +464,21 @@ fn read_macos_v3_c_error_report(report_dir string) ?MacosV3StagedReport {
 	v_sources_text := os.read_file(os.join_path(report_dir, macos_v3_c_error_v_sources_file)) or {
 		return none
 	}
-	v_sources := v_sources_text.split('\x00').filter(it != '')
+	v_source_digests_text := os.read_file(os.join_path(report_dir,
+		macos_v3_c_error_v_source_digests_file)) or { return none }
+	v_source_paths := v_sources_text.split('\x00').filter(it != '')
+	v_source_digests := v_source_digests_text.split('\x00').filter(it != '')
+	if v_source_paths.len != v_source_digests.len {
+		return none
+	}
+	mut v_sources := map[string]string{}
+	for i, path in v_source_paths {
+		digest := v_source_digests[i]
+		if digest.len != sha256.size * 2 {
+			return none
+		}
+		v_sources[path] = digest
+	}
 	return MacosV3StagedReport{
 		kind:       kind
 		ccompiler:  ccompiler.trim_space()

cmd/v/macos_v3_test.v

@@ -1,6 +1,7 @@
 module main
 
 import os
+import crypto.sha256
 import v.pref
 import v.builder
 
@@ -1545,7 +1546,10 @@ fn test_macos_v3_reads_c_error_fallback_report() {
 		os.write_file(os.join_path(root, macos_v3_c_error_compiler_file), 'clang')!
 		os.write_file(os.join_path(root, macos_v3_c_error_output_file),
 			'src.c:2:1: error: generated failure')!
-		os.write_file(os.join_path(root, macos_v3_c_error_v_sources_file), '')!
+		parsed_source := os.join_path(root, 'parsed.v')
+		parsed_digest := 'a'.repeat(sha256.size * 2)
+		os.write_file(os.join_path(root, macos_v3_c_error_v_sources_file), parsed_source)!
+		os.write_file(os.join_path(root, macos_v3_c_error_v_source_digests_file), parsed_digest)!
 		os.write_file(os.join_path(root, 'src.c'), 'int main(void) { return missing; }\n')!
 		report := read_macos_v3_c_error_report(root) or {
 			assert false
@@ -1554,6 +1558,7 @@ fn test_macos_v3_reads_c_error_fallback_report() {
 		assert report.ccompiler == 'clang'
 		assert report.c_output.contains('generated failure')
 		assert report.c_file == os.join_path(root, 'src.c')
+		assert report.v_sources[parsed_source] == parsed_digest
 		assert report.report_dir == root
 	}
 }
@@ -2014,7 +2019,7 @@ fn test_take_macos_v3_report_content_carries_no_path() {
 		// A forwarded content report round-trips as content only.
 		compiler_error := macos_v3_compiler_error_message('type specialization')
 		export_macos_v3_report_content(macos_v3_compiler_error_fallback, 'v3', compiler_error, '',
-			[])
+			map[string]string{})
 		report := take_macos_v3_report_content() or {
 			assert false, 'the forwarded content report must be returned'
 			return

doc/docs.md

@@ -117,10 +117,10 @@ source excerpt is included it is always a **bounded strict subset** of the faili
 file — a window around the failure, or a head+tail window — and the whole file is
 never uploaded.
 An internal V3 compiler error on a short program (and any directory build such as
-`v .`) submits metadata only. If a single-file input changes while V3 is compiling
-it, the fallback also submits metadata only rather than source V3 did not parse. A
-generated-C error, however, maps to a specific V file, so it can still upload a
-strict-subset excerpt of that file plus a few lines of context around the failing
+`v .`) submits metadata only. If the input selected for a source excerpt changes
+after V3 parses it, that report also submits metadata only rather than source V3
+did not parse. An unchanged input mapped from a generated-C error can still upload
+a strict-subset excerpt of that file plus a few lines of context around the failing
 line, even when the file is short. Inline-assembly
 fallbacks are notice-only and do not submit a report; reporting is also skipped
 for test compilations and to the default endpoint in GitHub CI. A custom fallback

vlib/v/builder/c_error_report.v

@@ -2,6 +2,7 @@ module builder
 
 import os
 import strings
+import crypto.sha256
 import v.pref
 import v.gen.c as cgen
 import v.util.version
@@ -298,11 +299,12 @@ pub fn export_external_v3_report_to_env(report ExternalCErrorBugReport) {
 // error. Internal compiler errors deliberately return no source here: their input must
 // be snapshotted before V3 starts and passed to bounded_v3_internal_fallback_source, so
 // a later editor/build-watcher rewrite cannot be uploaded as if V3 had parsed it.
-// `allowed_v_sources` is the set of source files that V3 actually parsed; generated-C
-// mappings outside that set are rejected before their contents are read. The returned
-// snippet is always a bounded strict subset (never a whole file); ('', '') means no
-// source is available, so the report stays metadata-only.
-pub fn bounded_v3_fallback_source(kind string, c_output string, c_file string, allowed_v_sources []string) (string, string) {
+// `allowed_v_sources` maps every source file V3 actually parsed to the SHA-256 digest
+// captured from those exact parser bytes. Generated-C mappings outside that set, or
+// files whose current contents no longer match the captured digest, contribute no
+// source. The returned snippet is always a bounded strict subset (never a whole file);
+// ('', '') means no source is available, so the report stays metadata-only.
+pub fn bounded_v3_fallback_source(kind string, c_output string, c_file string, allowed_v_sources map[string]string) (string, string) {
 	if kind == external_v3_compiler_error_kind {
 		return '', ''
 	}
@@ -326,7 +328,7 @@ pub fn bounded_v3_internal_fallback_source(source_name string, source string) (s
 // source line it came from (via the #line directives in the trusted staged C) and returns
 // a bounded window of that V file. The generated C was staged by this process's own V3
 // run, while its mapped V path is trusted only when it matches a parsed compiler input.
-fn bounded_v_source_for_generated_c(c_output string, generated_c_file string, allowed_v_sources []string) (string, string) {
+fn bounded_v_source_for_generated_c(c_output string, generated_c_file string, allowed_v_sources map[string]string) (string, string) {
 	c_source := os.read_file(generated_c_file) or { return '', '' }
 	c_lines := c_source.split_into_lines()
 	mut v_file := ''
@@ -340,10 +342,25 @@ fn bounded_v_source_for_generated_c(c_output string, generated_c_file string, al
 		v_file = source_loc.file
 		v_line = source_loc.line
 	}
-	if v_file == '' || !allowed_v_sources.any(same_path(it, v_file)) || !os.is_file(v_file) {
+	if v_file == '' || !os.is_file(v_file) {
+		return '', ''
+	}
+	mut parsed_digest := ''
+	for parsed_file, digest in allowed_v_sources {
+		if same_path(parsed_file, v_file) {
+			parsed_digest = digest
+			break
+		}
+	}
+	if parsed_digest == '' {
 		return '', ''
 	}
 	mapped_source := os.read_file(v_file) or { return '', '' }
+	if sha256.hexhash(mapped_source) != parsed_digest {
+		// An editor/build watcher rewrote this input after V3 parsed it. Keep the
+		// report metadata-only rather than uploading bytes unrelated to the failure.
+		return os.base(v_file), ''
+	}
 	mapped_lines := mapped_source.split_into_lines()
 	chunk := selected_v_source(v_file, mapped_lines, v_line)
 	mut v_source := bounded_v_source(chunk.text, c_error_bug_report_max_v_source_bytes, chunk.focus)

vlib/v/builder/c_error_report_test.v

@@ -1,6 +1,7 @@
 module builder
 
 import os
+import crypto.sha256
 import v.pref
 
 fn restore_env_var(name string, old_value ?string) {
@@ -329,7 +330,7 @@ fn test_external_v3_report_env_round_trip() {
 	assert v_source.contains(c_error_v_source_truncation_notice)
 	// The path-based extractor must never reopen an internal-error input after V3 fails.
 	late_file, late_source := bounded_v3_fallback_source(external_v3_compiler_error_kind,
-		'error: v3 failed', c_file, [])
+		'error: v3 failed', c_file, map[string]string{})
 	assert late_file == ''
 	assert late_source == ''
 	// ...then forwards only that content; export reads no path and deletes no directory.
@@ -629,7 +630,9 @@ fn test_bounded_v3_fallback_source_maps_generated_c_error() {
 	os.write_file(generated_c, '#line 100 "${v_path}"\nint a = 1;\nint b = missing;\n')!
 	c_output := '${generated_c}:3:9: error: use of undeclared identifier missing'
 	// kind '' selects the generated-C mapping path.
-	v_file, v_source := bounded_v3_fallback_source('', c_output, generated_c, [v_path])
+	v_file, v_source := bounded_v3_fallback_source('', c_output, generated_c, {
+		v_path: sha256.hexhash(whole)
+	})
 	assert v_file == 'source.v'
 	assert v_source != ''
 	// A bounded strict subset centered on the mapped V line (~100), never the whole file.
@@ -654,11 +657,14 @@ fn test_bounded_v3_fallback_source_rejects_nonblank_whole_file_generated_c() {
 		lines << 'fn f${i}() { println(${i}) }'
 	}
 	// 80 substantive lines plus a whitespace-only final line (81 lines total).
-	os.write_file(v_path, lines.join('\n') + '\n   ')!
+	whole := lines.join('\n') + '\n   '
+	os.write_file(v_path, whole)!
 	generated_c := os.join_path(dir, 'program.tmp.c')
 	os.write_file(generated_c, '#line 40 "${v_path}"\nint b = missing;\n')!
 	c_output := '${generated_c}:2:9: error: use of undeclared identifier missing'
-	v_file, v_source := bounded_v3_fallback_source('', c_output, generated_c, [v_path])
+	v_file, v_source := bounded_v3_fallback_source('', c_output, generated_c, {
+		v_path: sha256.hexhash(whole)
+	})
 	assert v_file == 'source.v'
 	// The mapped window exposes every nonblank line, so no source is uploaded.
 	assert v_source == '', v_source
@@ -681,13 +687,40 @@ fn test_bounded_v3_fallback_source_rejects_unparsed_mapped_file() {
 	generated_c := os.join_path(dir, 'program.tmp.c')
 	os.write_file(generated_c, '#line 1 "${unrelated_path}"\nint exposed = missing;\n')!
 	c_output := '${generated_c}:2:15: error: use of undeclared identifier missing'
-	v_file, v_source := bounded_v3_fallback_source('', c_output, generated_c, [
-		parsed_path,
-	])
+	v_file, v_source := bounded_v3_fallback_source('', c_output, generated_c, {
+		parsed_path: sha256.hexhash(os.read_file(parsed_path)!)
+	})
 	assert v_file == ''
 	assert v_source == ''
 }
 
+fn test_bounded_v3_fallback_source_rejects_changed_parsed_source() {
+	dir := os.join_path(os.vtmp_dir(), 'v3_gen_c_changed_${os.getpid()}')
+	os.rmdir_all(dir) or {}
+	os.mkdir_all(dir) or { panic(err) }
+	defer {
+		os.rmdir_all(dir) or {}
+	}
+	v_path := os.join_path(dir, 'source.v')
+	mut original_lines := []string{}
+	for i in 0 .. 200 {
+		original_lines << 'fn original_${i}() { println(${i}) }'
+	}
+	original := original_lines.join('\n')
+	os.write_file(v_path, original)!
+	parsed_digest := sha256.hexhash(original)
+	generated_c := os.join_path(dir, 'program.tmp.c')
+	os.write_file(generated_c, '#line 100 "${v_path}"\nint b = missing;\n')!
+	// Simulate an editor/build watcher replacing the mapped input after V3 parsed it.
+	os.write_file(v_path, original.replace('original_', 'new_private_'))!
+	c_output := '${generated_c}:2:9: error: use of undeclared identifier missing'
+	v_file, v_source := bounded_v3_fallback_source('', c_output, generated_c, {
+		v_path: parsed_digest
+	})
+	assert v_file == 'source.v'
+	assert v_source == ''
+}
+
 fn test_generated_c_reset_line_is_not_reported_as_v_source() {
 	c_lines := [
 		'#line 40 "/tmp/program.tmp.c"',

vlib/v/help/build/build-c.txt

@@ -22,10 +22,11 @@ see also `v help build`.
       be fixed: the V version, target and build options, plus a bounded snippet
       of the failing V source when one is available (always a strict subset,
       never the whole file). An internal V3 error on a short program or a
-      directory build such as `v .` submits metadata only. A single-file input
-      changed while V3 is compiling also submits metadata only, rather than source
-      V3 did not parse. A generated-C error can still include a strict-subset
-      excerpt. Inline-assembly fallbacks
+      directory build such as `v .` submits metadata only. If the input selected
+      for a source excerpt changes after V3 parses it, that report also submits
+      metadata only, rather than source V3 did not parse. An unchanged input mapped
+      from a generated-C error can still include a strict-subset excerpt.
+      Inline-assembly fallbacks
       are notice-only; test compilations and GitHub CI also submit no report to
       the default endpoint. An explicitly configured custom bug report URL remains
       active in CI. See `v help build` for the full description of the default.

vlib/v/help/build/build.txt

@@ -24,10 +24,10 @@ The default compiler:
   https://bugs.vlang.io, plus a bounded snippet of the failing V source when one
   is available. Any snippet is a strict subset and the whole file is never
   uploaded. An internal V3 compiler error on a short program or a directory build
-  such as `v .` submits metadata only. A single-file input changed while V3 is
-  compiling also submits metadata only, rather than source V3 did not parse. A
-  generated-C error maps to a V file and can still upload a strict-subset excerpt
-  of it, even for a short file.
+  such as `v .` submits metadata only. If the input selected for a source excerpt
+  changes after V3 parses it, that report also submits metadata only, rather than
+  source V3 did not parse. An unchanged input mapped from a generated-C error can
+  still upload a strict-subset excerpt of it, even for a short file.
   Inline-assembly fallbacks are notice-only; test compilations and GitHub CI also
   submit no report to the default endpoint. A custom fallback URL set with
   V_C_ERROR_BUG_REPORT_URL remains active in CI. The -bug-report-url option

vlib/v3/driver/driver.v

@@ -5,6 +5,7 @@ import runtime
 import strconv
 import strings
 import time
+import crypto.sha256
 import v3.ansi
 import v3.bench
 import v3.cmdexec
@@ -52,6 +53,7 @@ const macos_v3_c_error_compiler_file = 'compiler'
 const macos_v3_c_error_output_file = 'output'
 const macos_v3_c_error_source_name_file = 'source_name'
 const macos_v3_c_error_v_sources_file = 'v_sources'
+const macos_v3_c_error_v_source_digests_file = 'v_source_digests'
 
 fn configure_selfhost_parallelism(building_v bool) {
 	if !building_v || os.getenv('VJOBS') != '' || os.getenv('V3_NO_SELFHOST_JOB_OVERCOMMIT') != '' {
@@ -6218,7 +6220,7 @@ fn v3_impure_v_diagnostics(a &flat.FlatAst) []parser.Diagnostic {
 	return diagnostics
 }
 
-fn request_macos_v3_c_error_fallback(fallback_file string, report_dir string, ccompiler string, c_output string, c_source string, v_sources []string) bool {
+fn request_macos_v3_c_error_fallback(fallback_file string, report_dir string, ccompiler string, c_output string, c_source string, v_sources map[string]string) bool {
 	if fallback_file == '' || report_dir == '' || !os.is_file(c_source) {
 		return false
 	}
@@ -6242,7 +6244,25 @@ fn request_macos_v3_c_error_fallback(fallback_file string, report_dir string, cc
 		os.rmdir_all(report_dir) or {}
 		return false
 	}
-	os.write_file(os.join_path(report_dir, macos_v3_c_error_v_sources_file), v_sources.join('\x00')) or {
+	mut v_source_paths := v_sources.keys()
+	v_source_paths.sort()
+	mut v_source_paths_text := strings.new_builder(v_source_paths.len * 64)
+	mut v_source_digests_text := strings.new_builder(v_source_paths.len * (sha256.size * 2 + 1))
+	for i, path in v_source_paths {
+		if i > 0 {
+			v_source_paths_text.write_u8(0)
+			v_source_digests_text.write_u8(0)
+		}
+		v_source_paths_text.write_string(path)
+		v_source_digests_text.write_string(v_sources[path])
+	}
+	os.write_file(os.join_path(report_dir, macos_v3_c_error_v_sources_file),
+		v_source_paths_text.str()) or {
+		os.rmdir_all(report_dir) or {}
+		return false
+	}
+	os.write_file(os.join_path(report_dir, macos_v3_c_error_v_source_digests_file),
+		v_source_digests_text.str()) or {
 		os.rmdir_all(report_dir) or {}
 		return false
 	}
@@ -6253,7 +6273,7 @@ fn request_macos_v3_c_error_fallback(fallback_file string, report_dir string, cc
 	return true
 }
 
-fn request_macos_v3_c_error_fallback_from_message(fallback_file string, report_dir string, ccompiler string, message string, c_sources []string, v_sources []string) bool {
+fn request_macos_v3_c_error_fallback_from_message(fallback_file string, report_dir string, ccompiler string, message string, c_sources []string, v_sources map[string]string) bool {
 	is_c_error := message.starts_with('failed to build C object ')
 		|| message.starts_with('failed to build cached module object ')
 		|| message.starts_with('failed to build cached program prefix:')
@@ -6270,16 +6290,28 @@ fn request_macos_v3_c_error_fallback_from_message(fallback_file string, report_d
 	return false
 }
 
-fn macos_v3_fallback_report_sources(a &flat.FlatAst) []string {
-	mut sources := map[string]bool{}
+fn macos_v3_fallback_report_sources(a &flat.FlatAst) map[string]string {
+	mut sources := map[string]string{}
+	mut ambiguous := map[string]bool{}
 	for _, file in a.source_files {
-		if file.name.ends_with('.v') || file.name.ends_with('.vv') || file.name.ends_with('.vsh') {
-			sources[os.real_path(file.name)] = true
+		if (file.name.ends_with('.v') || file.name.ends_with('.vv')
+			|| file.name.ends_with('.vsh')) && file.has_source_sha256() {
+			path := os.real_path(file.name)
+			source_digest := file.source_sha256()
+			digest := source_digest[..].hex()
+			if old_digest := sources[path] {
+				if old_digest != digest {
+					ambiguous[path] = true
+				}
+			} else {
+				sources[path] = digest
+			}
 		}
 	}
-	mut paths := sources.keys()
-	paths.sort()
-	return paths
+	for path, _ in ambiguous {
+		sources.delete(path)
+	}
+	return sources
 }
 
 fn input_uses_minimal_literal_output_builtin(input_file string, prefs &pref.Preferences, is_test_command bool, is_checker_fixture bool) bool {