module main import os import time import term import flag const scan_timeout_s = get_scan_timeout_seconds() const max_v_cycles = 1000 const scan_frequency_hz = 4 const scan_period_ms = 1000 / scan_frequency_hz const max_scan_cycles = scan_timeout_s * scan_frequency_hz fn get_scan_timeout_seconds() int { env_vw_timeout := os.getenv('VWATCH_TIMEOUT').int() if env_vw_timeout == 0 { $if gcboehm ? { return 35000000 // over 1 year } $else { return 5 * 60 } } return env_vw_timeout } // // Implements `v watch file.v` , `v watch run file.v` etc. // With this command, V will collect all .v files that are needed for the // compilation, then it will enter an infinite loop, monitoring them for // changes. // // When a change is detected, it will stop the current process, if it is // still running, then rerun/recompile/etc. // // In effect, this makes it easy to have an editor session and a separate // terminal, running just `v watch run file.v`, and you will see your // changes right after you save your .v file in your editor. // // // Since -gc boehm is not available on all platforms yet, // and this program leaks ~8MB/minute without it, the implementation here // is done similarly to vfmt in 2 modes, in the same executable: // // a) A parent/manager process that only manages a single worker // process. The parent process does mostly nothing except restarting // workers, thus it does not leak much. // // b) A worker process, doing the actual monitoring/polling. // NB: *workers are started with the --vwatchworker option* // // Worker processes will run for a limited number of iterations, then // they will do exit(255), and then the parent will start a new worker. // Exiting by any other code will cause the parent to also exit with the // same error code. This limits the potential leak that a worker process // can do, even without using the garbage collection mode. // struct VFileStat { path string mtime int } [unsafe] fn (mut vfs VFileStat) free() { unsafe { vfs.path.free() } } enum RerunCommand { restart quit } struct Context { mut: pid int // the pid of the current process; useful while debugging manager/worker interactions is_worker bool // true in the workers, false in the manager process check_period_ms int = scan_period_ms vexe string affected_paths []string vfiles []VFileStat opts []string rerun_channel chan RerunCommand child_process &os.Process is_exiting bool // set by SIGINT/Ctrl-C v_cycles int // how many times the worker has restarted the V compiler scan_cycles int // how many times the worker has scanned for source file changes clear_terminal bool // whether to clear the terminal before each re-run silent bool // when true, watch will not print a timestamp line before each re-run add_files []string // path to additional files that have to be watched for changes ignore_exts []string // extensions of files that will be ignored, even if they change (useful for sqlite.db files for example) cmd_before_run string // a command to run before each re-run cmd_after_run string // a command to run after each re-run } [if debug_vwatch ?] fn (mut context Context) elog(msg string) { eprintln('> vwatch $context.pid, $msg') } fn (context &Context) str() string { return 'Context{ pid: $context.pid, is_worker: $context.is_worker, check_period_ms: $context.check_period_ms, vexe: $context.vexe, opts: $context.opts, is_exiting: $context.is_exiting, vfiles: $context.vfiles' } fn (mut context Context) get_stats_for_affected_vfiles() []VFileStat { if context.affected_paths.len == 0 { mut apaths := map[string]bool{} // The next command will make V parse the program, and print all .v files, // needed for its compilation, without actually compiling it. copts := context.opts.join(' ') cmd := '"$context.vexe" -silent -print-v-files $copts' // context.elog('> cmd: $cmd') mut paths := []string{} if context.add_files.len > 0 && context.add_files[0] != '' { paths << context.add_files } vfiles := os.execute(cmd) if vfiles.exit_code == 0 { paths_trimmed := vfiles.output.trim_space() paths << paths_trimmed.split('\n') } for vf in paths { apaths[os.real_path(os.dir(vf))] = true } context.affected_paths = apaths.keys() // context.elog('vfiles paths to be scanned: $context.affected_paths') } // scan all files in the found folders mut newstats := []VFileStat{} for path in context.affected_paths { mut files := os.ls(path) or { []string{} } for pf in files { pf_ext := os.file_ext(pf).to_lower() if pf_ext in ['', '.bak', '.exe', '.dll', '.so', '.def'] { continue } if pf_ext in context.ignore_exts { continue } if pf.starts_with('.#') { continue } if pf.ends_with('~') { continue } f := os.join_path(path, pf) fullpath := os.real_path(f) mtime := os.file_last_mod_unix(fullpath) newstats << VFileStat{fullpath, mtime} } } // always add the v compiler itself, so that if it is recompiled with `v self` // the watcher will rerun the compilation too newstats << VFileStat{context.vexe, os.file_last_mod_unix(context.vexe)} return newstats } fn (mut context Context) get_changed_vfiles() int { mut changed := 0 newfiles := context.get_stats_for_affected_vfiles() for vfs in newfiles { mut found := false for existing_vfs in context.vfiles { if existing_vfs.path == vfs.path { found = true if existing_vfs.mtime != vfs.mtime { context.elog('> new updates for file: $vfs') changed++ } break } } if !found { changed++ continue } } context.vfiles = newfiles if changed > 0 { context.elog('> get_changed_vfiles: $changed') } return changed } fn change_detection_loop(ocontext &Context) { mut context := unsafe { ocontext } for { if context.v_cycles >= max_v_cycles || context.scan_cycles >= max_scan_cycles { context.is_exiting = true context.kill_pgroup() time.sleep(50 * time.millisecond) exit(255) } if context.is_exiting { return } changes := context.get_changed_vfiles() if changes > 0 { context.rerun_channel <- RerunCommand.restart } time.sleep(context.check_period_ms * time.millisecond) context.scan_cycles++ } } fn (mut context Context) kill_pgroup() { if context.child_process == 0 { return } if context.child_process.is_alive() { context.child_process.signal_pgkill() } context.child_process.wait() } fn (mut context Context) run_before_cmd() { if context.cmd_before_run != '' { context.elog('> run_before_cmd: "$context.cmd_before_run"') os.system(context.cmd_before_run) } } fn (mut context Context) run_after_cmd() { if context.cmd_after_run != '' { context.elog('> run_after_cmd: "$context.cmd_after_run"') os.system(context.cmd_after_run) } } fn (mut context Context) compilation_runner_loop() { cmd := '"$context.vexe" ${context.opts.join(' ')}' _ := <-context.rerun_channel for { context.elog('>> loop: v_cycles: $context.v_cycles') if context.clear_terminal { term.clear() } context.run_before_cmd() timestamp := time.now().format_ss_milli() context.child_process = os.new_process(context.vexe) context.child_process.use_pgroup = true context.child_process.set_args(context.opts) context.child_process.run() if !context.silent { eprintln('$timestamp: $cmd | pid: ${context.child_process.pid:7d} | reload cycle: ${context.v_cycles:5d}') } for { mut notalive_count := 0 mut cmds := []RerunCommand{} for { if context.is_exiting { return } if !context.child_process.is_alive() { context.child_process.wait() notalive_count++ if notalive_count == 1 { // a short lived process finished, do cleanup: context.run_after_cmd() } } select { action := <-context.rerun_channel { cmds << action if action == .quit { context.kill_pgroup() return } } 100 * time.millisecond { should_restart := RerunCommand.restart in cmds cmds = [] if should_restart { // context.elog('>>>>>>>> KILLING $context.child_process.pid') context.kill_pgroup() break } } } } if !context.child_process.is_alive() { context.child_process.wait() context.child_process.close() if notalive_count == 0 { // a long running process was killed, do cleanup: context.run_after_cmd() } break } } context.v_cycles++ } } const ccontext = Context{ child_process: 0 } fn main() { dump(scan_timeout_s) mut context := unsafe { &Context(voidptr(&ccontext)) } context.pid = os.getpid() context.vexe = os.getenv('VEXE') mut fp := flag.new_flag_parser(os.args[1..]) fp.application('v watch') if os.args[1] == 'watch' { fp.skip_executable() } fp.version('0.0.2') fp.description('Collect all .v files needed for a compilation, then re-run the compilation when any of the source changes.') fp.arguments_description('[--silent] [--clear] [--ignore .db] [--add /path/to/a/file.v] [run] program.v') fp.allow_unknown_args() fp.limit_free_args_to_at_least(1) ? context.is_worker = fp.bool('vwatchworker', 0, false, 'Internal flag. Used to distinguish vwatch manager and worker processes.') context.silent = fp.bool('silent', `s`, false, 'Be more silent; do not print the watch timestamp before each re-run.') context.clear_terminal = fp.bool('clear', `c`, false, 'Clears the terminal before each re-run.') context.add_files = fp.string('add', `a`, '', 'Add more files to be watched. Useful with `v watch -add=/tmp/feature.v run cmd/v /tmp/feature.v`, if you change *both* the compiler, and the feature.v file.').split(',') context.ignore_exts = fp.string('ignore', `i`, '', 'Ignore files having these extensions. Useful with `v watch -ignore=.db run server.v`, if your server writes to an sqlite.db file in the same folder.').split(',') show_help := fp.bool('help', `h`, false, 'Show this help screen.') context.cmd_before_run = fp.string('before', 0, '', 'A command to execute *before* each re-run.') context.cmd_after_run = fp.string('after', 0, '', 'A command to execute *after* each re-run.') if show_help { println(fp.usage()) exit(0) } remaining_options := fp.finalize() or { eprintln('Error: $err') exit(1) } context.opts = remaining_options context.elog('>>> context.pid: $context.pid') context.elog('>>> context.vexe: $context.vexe') context.elog('>>> context.opts: $context.opts') context.elog('>>> context.is_worker: $context.is_worker') context.elog('>>> context.clear_terminal: $context.clear_terminal') context.elog('>>> context.add_files: $context.add_files') context.elog('>>> context.ignore_exts: $context.ignore_exts') if context.is_worker { context.worker_main() } else { context.manager_main() } } fn (mut context Context) manager_main() { myexecutable := os.executable() mut worker_opts := ['--vwatchworker'] worker_opts << os.args[2..] for { mut worker_process := os.new_process(myexecutable) worker_process.set_args(worker_opts) worker_process.run() for { if !worker_process.is_alive() { worker_process.wait() break } time.sleep(200 * time.millisecond) } if !(worker_process.code == 255 && worker_process.status == .exited) { worker_process.close() break } worker_process.close() } } fn (mut context Context) worker_main() { context.rerun_channel = chan RerunCommand{cap: 10} os.signal_opt(.int, fn (_ os.Signal) { mut context := unsafe { &Context(voidptr(&ccontext)) } context.is_exiting = true context.kill_pgroup() }) or { panic(err) } go context.compilation_runner_loop() change_detection_loop(context) }