coverage: support output in LCOV tracefile format (#30381)

Also ensure that `julia_cmd` is forwarding all desirable options,
and provide a general framework for specifying formatted filenames
for similar such options.

Author: vtjnash

base/options.jl

@@ -39,6 +39,7 @@ struct JLOptions
     outputjitbc::Ptr{UInt8}
     outputo::Ptr{UInt8}
     outputji::Ptr{UInt8}
+    output_code_coverage::Ptr{UInt8}
     incremental::Int8
 end
 

base/util.jl

@@ -399,42 +399,79 @@ printstyled(io::IO, msg...; bold::Bool=false, color::Union{Int,Symbol}=:normal)
     with_output_color(print, color, io, msg...; bold=bold)
 printstyled(msg...; bold::Bool=false, color::Union{Int,Symbol}=:normal) =
     printstyled(stdout, msg...; bold=bold, color=color)
+
 """
     Base.julia_cmd(juliapath=joinpath(Sys.BINDIR::String, julia_exename()))
 
 Return a julia command similar to the one of the running process.
-Propagates the `--cpu-target`, `--sysimage`, --compile `, `--depwarn`
-and `--inline` command line arguments.
+Propagates any of the `--cpu-target`, `--sysimage`, `--compile`, `--sysimage-native-code`,
+`--compiled-modules`, `--inline`, `--check-bounds`, `--optimize`, `-g`,
+`--code-coverage`, and `--depwarn`
+command line arguments that are not at their default values.
+
+Among others, `--math-mode`, `--warn-overwrite`, and `--trace-compile` are notably not propagated currently.
 
 !!! compat "Julia 1.1"
-    The `--inline` flag is only propagated in Julia 1.1 and later.
+    Only the `--cpu-target`, `--sysimage`, `--depwarn`, `--compile` and `--check-bounds` flags were propagated before Julia 1.1.
 """
 function julia_cmd(julia=joinpath(Sys.BINDIR::String, julia_exename()))
     opts = JLOptions()
     cpu_target = unsafe_string(opts.cpu_target)
     image_file = unsafe_string(opts.image_file)
-    compile = if opts.compile_enabled == 0
-                  "no"
-              elseif opts.compile_enabled == 2
-                  "all"
-              elseif opts.compile_enabled == 3
-                  "min"
-              else
-                  "yes"
-              end
-    depwarn = if opts.depwarn == 0
-                  "no"
-              elseif opts.depwarn == 2
-                  "error"
-              else
-                  "yes"
-              end
-    inline  = if opts.can_inline == 0
-                  "no"
-              else
-                  "yes"
-              end
-    `$julia -C$cpu_target -J$image_file --compile=$compile --depwarn=$depwarn --inline=$inline`
+    addflags = String[]
+    let compile = if opts.compile_enabled == 0
+                      "no"
+                  elseif opts.compile_enabled == 2
+                      "all"
+                  elseif opts.compile_enabled == 3
+                      "min"
+                  else
+                      "" # default = "yes"
+                  end
+        isempty(compile) || push!(addflags, "--compile=$compile")
+    end
+    let depwarn = if opts.depwarn == 0
+                      "no"
+                  elseif opts.depwarn == 2
+                      "error"
+                  else
+                      "" # default = "yes"
+                  end
+        isempty(depwarn) || push!(addflags, "--depwarn=$depwarn")
+    end
+    let check_bounds = if opts.check_bounds == 1
+                      "yes" # on
+                  elseif opts.check_bounds == 2
+                      "no" # off
+                  else
+                      "" # "default"
+                  end
+        isempty(check_bounds) || push!(addflags, "--check-bounds=$check_bounds")
+    end
+    opts.can_inline == 0 && push!(addflags, "--inline=no")
+    opts.use_compiled_modules == 0 && push!(addflags, "--compiled-modules=no")
+    opts.opt_level == 2 || push!(addflags, "-O$(opts.opt_level)")
+    push!(addflags, "-g$(opts.debug_level)")
+    if opts.code_coverage != 0
+        # Forward the code-coverage flag only if applicable (if the filename is pid-dependent)
+        coverage_file = (opts.output_code_coverage != C_NULL) ?  unsafe_string(opts.output_code_coverage) : ""
+        if isempty(coverage_file) || occursin("%p", coverage_file)
+            if opts.code_coverage == 1
+                push!(addflags, "--code-coverage=user")
+            elseif opts.code_coverage == 2
+                push!(addflags, "--code-coverage=all")
+            end
+            isempty(coverage_file) || push!(addflags, "--code-coverage=$coverage_file")
+        end
+    end
+    if opts.malloc_log != 0
+        if opts.malloc_log == 1
+            push!(addflags, "--track-allocation=user")
+        elseif opts.malloc_log == 2
+            push!(addflags, "--track-allocation=all")
+        end
+    end
+    return `$julia -C$cpu_target -J$image_file $addflags`
 end
 
 function julia_exename()

src/codegen.cpp

@@ -1696,7 +1696,7 @@ static void write_log_data(logdata_t &logData, const char *extension)
             std::ifstream inf(filename.c_str());
             if (inf.is_open()) {
                 std::string outfile = filename + extension;
-                std::ofstream outf(outfile.c_str(), std::ofstream::trunc | std::ofstream::out);
+                std::ofstream outf(outfile.c_str(), std::ofstream::trunc | std::ofstream::out | std::ofstream::binary);
                 char line[1024];
                 int l = 1;
                 unsigned block = 0;
@@ -1722,7 +1722,7 @@ static void write_log_data(logdata_t &logData, const char *extension)
                     else
                         outf << (value - 1);
                     outf.width(0);
-                    outf << " " << line << std::endl;
+                    outf << " " << line << '\n';
                 }
                 outf.close();
                 inf.close();
@@ -1732,20 +1732,68 @@ static void write_log_data(logdata_t &logData, const char *extension)
 }
 
 extern "C" int jl_getpid();
-extern "C" void jl_write_coverage_data(void)
+
+static void write_lcov_data(logdata_t &logData, const std::string &outfile)
 {
-    std::ostringstream stm;
-    stm << jl_getpid();
-    std::string outf = "." + stm.str() + ".cov";
-    write_log_data(coverageData, outf.c_str());
+    std::ofstream outf(outfile.c_str(), std::ofstream::ate | std::ofstream::out | std::ofstream::binary);
+    //std::string base = std::string(jl_options.julia_bindir);
+    //base = base + "/../share/julia/base/";
+    logdata_t::iterator it = logData.begin();
+    for (; it != logData.end(); it++) {
+        const std::string &filename = it->first();
+        const std::vector<logdata_block*> &values = it->second;
+        if (!values.empty()) {
+            //if (!isabspath(filename.c_str()))
+            //    filename = base + filename;
+            outf << "SF:" << filename << '\n';
+            size_t n_covered = 0;
+            size_t n_instrumented = 0;
+            size_t lno = 0;
+            for (auto &itv : values) {
+                if (itv) {
+                    logdata_block &data = *itv;
+                    for (int i = 0; i < logdata_blocksize; i++) {
+                        auto cov = data[i];
+                        if (cov > 0) {
+                            n_instrumented++;
+                            if (cov > 1)
+                                n_covered++;
+                            outf << "DA:" << lno << ',' << (cov - 1) << '\n';
+                        }
+                        lno++;
+                    }
+                }
+                else {
+                    lno += logdata_blocksize;
+                }
+            }
+            outf << "LH:" << n_covered << '\n';
+            outf << "LF:" << n_instrumented << '\n';
+            outf << "end_of_record\n";
+        }
+    }
+    outf.close();
+}
+
+extern "C" void jl_write_coverage_data(const char *output)
+{
+    if (output) {
+        StringRef output_pattern(output);
+        if (output_pattern.endswith(".info"))
+            write_lcov_data(coverageData, jl_format_filename(output_pattern));
+    }
+    else {
+        std::ostringstream stm;
+        stm << "." << jl_getpid() << ".cov";
+        write_log_data(coverageData, stm.str().c_str());
+    }
 }
 
 extern "C" void jl_write_malloc_log(void)
 {
     std::ostringstream stm;
-    stm << jl_getpid();
-    std::string outf = "." + stm.str() + ".mem";
-    write_log_data(mallocData, outf.c_str());
+    stm << "." << jl_getpid() << ".mem";
+    write_log_data(mallocData, stm.str().c_str());
 }
 
 // --- constant determination ---
@@ -7368,7 +7416,7 @@ extern "C" void *jl_init_llvm(void)
     cl::ParseEnvironmentOptions("Julia", "JULIA_LLVM_ARGS");
 
     jl_page_size = jl_getpagesize();
-    imaging_mode = jl_generating_output();
+    imaging_mode = jl_generating_output() && !jl_options.incremental;
     jl_init_debuginfo();
 
 #ifdef USE_POLLY

src/init.c

@@ -163,7 +163,7 @@ static void jl_uv_exitcleanup_walk(uv_handle_t *handle, void *arg)
     jl_uv_exitcleanup_add(handle, (struct uv_shutdown_queue*)arg);
 }
 
-void jl_write_coverage_data(void);
+void jl_write_coverage_data(const char*);
 void jl_write_malloc_log(void);
 void jl_write_compiler_output(void);
 
@@ -222,7 +222,7 @@ JL_DLLEXPORT void jl_atexit_hook(int exitcode)
         jl_write_compiler_output();
     jl_print_gc_stats(JL_STDERR);
     if (jl_options.code_coverage)
-        jl_write_coverage_data();
+        jl_write_coverage_data(jl_options.output_code_coverage);
     if (jl_options.malloc_log)
         jl_write_malloc_log();
     if (jl_base_module) {
@@ -462,7 +462,7 @@ int isabspath(const char *in)
 }
 
 static char *abspath(const char *in, int nprefix)
-{ // compute an absolute path location, so that chdir doesn't change the file reference
+{ // compute an absolute realpath location, so that chdir doesn't change the file reference
   // ignores (copies directly over) nprefix characters at the start of abspath
 #ifndef _OS_WINDOWS_
     char *out = realpath(in + nprefix, NULL);
@@ -495,6 +495,8 @@ static char *abspath(const char *in, int nprefix)
                 jl_error("fatal error: unexpected error while retrieving current working directory");
             }
             out = (char*)malloc(path_size + 1 + sz + nprefix);
+            if (!out)
+                jl_errorf("fatal error: failed to allocate memory: %s", strerror(errno));
             memcpy(out, in, nprefix);
             memcpy(out + nprefix, path, path_size);
             out[nprefix + path_size] = PATHSEPSTRING[0];
@@ -508,6 +510,8 @@ static char *abspath(const char *in, int nprefix)
         jl_error("fatal error: jl_options.image_file path too long or GetFullPathName failed");
     }
     char *out = (char*)malloc(n + nprefix);
+    if (!out)
+        jl_errorf("fatal error: failed to allocate memory: %s", strerror(errno));
     DWORD m = GetFullPathName(in + nprefix, n, out + nprefix, NULL);
     if (n != m + 1) {
         jl_error("fatal error: jl_options.image_file path too long or GetFullPathName failed");
@@ -517,6 +521,38 @@ static char *abspath(const char *in, int nprefix)
     return out;
 }
 
+// create an absolute-path copy of the input path format string
+// formed as `joinpath(replace(pwd(), "%" => "%%"), in)`
+// unless `in` starts with `%`
+static const char *absformat(const char *in)
+{
+    if (in[0] == '%' || isabspath(in))
+        return in;
+    // get an escaped copy of cwd
+    size_t path_size = PATH_MAX;
+    char path[PATH_MAX];
+    if (uv_cwd(path, &path_size)) {
+        jl_error("fatal error: unexpected error while retrieving current working directory");
+    }
+    size_t sz = strlen(in) + 1;
+    size_t i, fmt_size = 0;
+    for (i = 0; i < path_size; i++)
+        fmt_size += (path[i] == '%' ? 2 : 1);
+    char *out = (char*)malloc(fmt_size + 1 + sz);
+    if (!out)
+        jl_errorf("fatal error: failed to allocate memory: %s", strerror(errno));
+    fmt_size = 0;
+    for (i = 0; i < path_size; i++) { // copy-replace pwd portion
+        char c = path[i];
+        out[fmt_size++] = c;
+        if (c == '%')
+            out[fmt_size++] = '%';
+    }
+    out[fmt_size++] = PATHSEPSTRING[0]; // path sep
+    memcpy(out + fmt_size, in, sz); // copy over format, including nul
+    return out;
+}
+
 static void jl_resolve_sysimg_location(JL_IMAGE_SEARCH rel)
 {   // this function resolves the paths in jl_options to absolute file locations as needed
     // and it replaces the pointers to `julia_bindir`, `julia_bin`, `image_file`, and output file paths
@@ -527,13 +563,17 @@ static void jl_resolve_sysimg_location(JL_IMAGE_SEARCH rel)
     // calling `julia_init()`
     char *free_path = (char*)malloc(PATH_MAX);
     size_t path_size = PATH_MAX;
+    if (!free_path)
+        jl_errorf("fatal error: failed to allocate memory: %s", strerror(errno));
     if (uv_exepath(free_path, &path_size)) {
         jl_error("fatal error: unexpected error while retrieving exepath");
     }
     if (path_size >= PATH_MAX) {
         jl_error("fatal error: jl_options.julia_bin path too long");
     }
     jl_options.julia_bin = (char*)malloc(path_size+1);
+    if (!jl_options.julia_bin)
+        jl_errorf("fatal error: failed to allocate memory: %s", strerror(errno));
     memcpy((char*)jl_options.julia_bin, free_path, path_size);
     ((char*)jl_options.julia_bin)[path_size] = '\0';
     if (!jl_options.julia_bindir) {
@@ -550,6 +590,8 @@ static void jl_resolve_sysimg_location(JL_IMAGE_SEARCH rel)
         if (rel == JL_IMAGE_JULIA_HOME && !isabspath(jl_options.image_file)) {
             // build time path, relative to JULIA_BINDIR
             free_path = (char*)malloc(PATH_MAX);
+            if (!free_path)
+                jl_errorf("fatal error: failed to allocate memory: %s", strerror(errno));
             int n = snprintf(free_path, PATH_MAX, "%s" PATHSEPSTRING "%s",
                              jl_options.julia_bindir, jl_options.image_file);
             if (n >= PATH_MAX || n < 0) {
@@ -572,8 +614,13 @@ static void jl_resolve_sysimg_location(JL_IMAGE_SEARCH rel)
         jl_options.outputbc = abspath(jl_options.outputbc, 0);
     if (jl_options.machine_file)
         jl_options.machine_file = abspath(jl_options.machine_file, 0);
-    if (jl_options.project && strncmp(jl_options.project, "@.", strlen(jl_options.project)) != 0)
+    if (jl_options.project
+            && strcmp(jl_options.project, "@.") != 0
+            && strcmp(jl_options.project, "@") != 0
+            && strcmp(jl_options.project, "") != 0)
         jl_options.project = abspath(jl_options.project, 0);
+    if (jl_options.output_code_coverage)
+        jl_options.output_code_coverage = absformat(jl_options.output_code_coverage);
 
     const char **cmdp = jl_options.cmds;
     if (cmdp) {
@@ -600,6 +647,7 @@ void _julia_init(JL_IMAGE_SEARCH rel)
     jl_get_ptls_states_getter();
 #endif
     jl_ptls_t ptls = jl_get_ptls_states();
+    (void)ptls; assert(ptls); // make sure early that we have initialized ptls
     jl_safepoint_init();
     libsupport_init();
     htable_new(&jl_current_modules, 0);

src/jloptions.c

@@ -68,6 +68,7 @@ jl_options_t jl_options = { 0,    // quiet
                             NULL, // output-jit-bc
                             NULL, // output-o
                             NULL, // output-ji
+                            NULL,    // output-code_coverage
                             0, // incremental
                             0 // image_file_specified
 };
@@ -129,6 +130,9 @@ static const char opts[]  =
     // instrumentation options
     " --code-coverage={none|user|all}, --code-coverage\n"
     "                           Count executions of source lines (omitting setting is equivalent to \"user\")\n"
+    " --code-coverage=tracefile.info\n"
+    "                           Append coverage information to the LCOV tracefile (filename supports format tokens).\n"
+// TODO: These TOKENS are defined in `runtime_ccall.cpp`. A more verbose `--help` should include that list here.
     " --track-allocation={none|user|all}, --track-allocation\n"
     "                           Count bytes allocated by each source line\n\n"
 
@@ -437,12 +441,18 @@ JL_DLLEXPORT void jl_parse_opts(int *argcp, char ***argvp)
             break;
         case opt_code_coverage:
             if (optarg != NULL) {
-                if (!strcmp(optarg,"user"))
+                size_t endof = strlen(optarg);
+                if (!strcmp(optarg, "user"))
                     codecov = JL_LOG_USER;
-                else if (!strcmp(optarg,"all"))
+                else if (!strcmp(optarg, "all"))
                     codecov = JL_LOG_ALL;
-                else if (!strcmp(optarg,"none"))
+                else if (!strcmp(optarg, "none"))
                     codecov = JL_LOG_NONE;
+                else if (endof > 5 && !strcmp(optarg + endof - 5, ".info")) {
+                    if (codecov == JL_LOG_NONE)
+                        codecov = JL_LOG_ALL;
+                    jl_options.output_code_coverage = optarg;
+                }
                 else
                     jl_errorf("julia: invalid argument to --code-coverage (%s)", optarg);
                 break;