gg/brep.c

/*
 * brep - minimal grep-like utility for basic text search.
 *
 * Usage: brep SEARCH_TERM [FILE_OR_DIRECTORY]
 *
 * Recursively walks FILE_OR_DIRECTORY (default: current working directory),
 * reads every regular file, searches it for the case-sensitive SEARCH_TERM,
 * and prints the paths of all files containing at least one occurrence
 * (one per line).
 *
 * Implementation notes
 * --------------------
 * Scanning is split into two phases so that the expensive work parallelises
 * well:
 *
 *   1. Walk: a single thread enumerates every regular file in the tree into a
 *      growable array of paths. The walk uses d_type from readdir and only
 *      falls back to lstat for DT_UNKNOWN entries; symlinks are not followed
 *      (cycle-safe). This phase is cheap (it touches metadata, not file data).
 *
 *   2. Scan: N worker threads grab files from a single atomic counter and scan
 *      each one independently. Files are read with read() into a small
 *      per-thread buffer (avoids the mmap/munmap setup and the page-fault
 *      storm mmap incurs across a large tree, while keeping RAM bounded to one
 *      buffer per worker). Chunks overlap by needle_len-1 bytes so a match
 *      straddling a chunk boundary is still found.
 *
 * Binary files are skipped, matching ripgrep's default behaviour: a file that
 * contains a NUL byte anywhere is treated as binary and is never reported,
 * even if it contains the search term. Detection is cheap: we search for the
 * needle with memmem (a single pass over the data); only when a needle is
 * found do we probe the already-read chunk for a NUL byte, so the common
 * non-matching path pays no extra scan.
 *
 * Matching uses POSIX memmem(); the scan phase runs across the online CPUs.
 *
 * Exit codes: 0 - at least one file matched,
 *             1 - no matches,
 *             2 - usage or input error.
 *
 * No third-party dependencies, no external commands.
 */
#ifndef _POSIX_C_SOURCE
#define _POSIX_C_SOURCE 200809L
#endif
#ifndef _DEFAULT_SOURCE
#define _DEFAULT_SOURCE
#endif

#include <dirent.h>
#include <errno.h>
#include <fcntl.h>
#include <pthread.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <unistd.h>

#ifndef PATH_MAX
#define PATH_MAX 4096
#endif

#define IO_BUFSZ (32u * 1024) /* per-thread read buffer */

/* ---------- shared scan state ---------- */
static const char *needle;
static size_t needle_len;
static int found_any = 0;
static pthread_mutex_t found_mx = PTHREAD_MUTEX_INITIALIZER;

/* ---------- collected file paths ---------- */
static char **paths = NULL;
static size_t npaths = 0, path_cap = 0;
static pthread_mutex_t paths_mx = PTHREAD_MUTEX_INITIALIZER;
static size_t scan_idx = 0; /* atomic work counter */

static void die(const char *msg) {
  perror(msg);
  exit(2);
}

static void add_path(const char *p) {
  pthread_mutex_lock(&paths_mx);
  if (npaths == path_cap) {
    path_cap = path_cap ? path_cap * 2 : 1 << 16;
    paths = realloc(paths, path_cap * sizeof *paths);
    if (paths == NULL)
      die("realloc");
  }
  paths[npaths++] = strdup(p);
  if (paths[npaths - 1] == NULL)
    die("strdup");
  pthread_mutex_unlock(&paths_mx);
}

/* ---------- walk: collect regular files ---------- */
static void walk(const char *dir) {
  DIR *d = opendir(dir);
  if (d == NULL)
    return;
  struct dirent *ent;
  char child[PATH_MAX];
  while ((ent = readdir(d)) != NULL) {
    if (strcmp(ent->d_name, ".") == 0 ||
        strcmp(ent->d_name, "..") == 0)
      continue;
    int n = snprintf(child, sizeof child, "%s/%s", dir, ent->d_name);
    if (n < 0 || (size_t)n >= sizeof child)
      continue; /* path too long */

    if (ent->d_type == DT_REG) {
      add_path(child);
    } else if (ent->d_type == DT_DIR) {
      walk(child);
    } else if (ent->d_type == DT_LNK) {
      continue; /* do not follow symlinks: avoids cycles */
    } else {
      struct stat st;
      if (lstat(child, &st) != 0)
        continue;
      if (S_ISDIR(st.st_mode))
        walk(child);
      else if (S_ISREG(st.st_mode))
        add_path(child);
    }
  }
  closedir(d);
}

/* ---------- scan one open fd (chunked read + memmem, binary-skipping) ---------- */
typedef struct {
  char *buf;
  char *carry;
} ScanBufs;

static int scan_fd(int fd, ScanBufs *sb) {
  char *buf = sb->buf;
  char *carry = sb->carry;
  size_t carryn = 0;
  int hit = 0;

  for (;;) {
    ssize_t r = read(fd, buf + carryn, IO_BUFSZ);
    if (r < 0) {
      if (errno == EINTR)
        continue;
      break;
    }
    if (r == 0)
      break; /* EOF */
    size_t wn = carryn + (size_t)r;

    if (memmem(buf, wn, needle, needle_len) != NULL) {
      /* Candidate match. A NUL byte anywhere in the data read marks a binary
       * file, which we skip (ripgrep's default). The probe is bounded to the
       * chunk already in memory, so it is cheap and only runs on the rare
       * files that actually contain the needle. */
      if (memchr(buf, '\0', wn) == NULL) {
        hit = 1;
        break;
      }
      break; /* binary file that happens to contain the needle */
    }

    if (wn >= needle_len)
      carryn = needle_len - 1;
    else
      carryn = wn;
    memcpy(carry, buf + wn - carryn, carryn);
    memcpy(buf, carry, carryn);
  }

  return hit;
}

static void report_match(const char *path) {
  pthread_mutex_lock(&found_mx);
  found_any = 1;
  puts(path);
  pthread_mutex_unlock(&found_mx);
}

static void scan_file(const char *path, ScanBufs *sb) {
  int fd = open(path, O_RDONLY | O_CLOEXEC);
  if (fd < 0)
    return; /* unreadable entries are silently skipped */
  if (scan_fd(fd, sb))
    report_match(path);
  close(fd);
}

/* ---------- worker: grab files by atomic index, scan ---------- */
static void *worker(void *arg) {
  (void)arg;
  ScanBufs sb;
  sb.buf = malloc(IO_BUFSZ + needle_len);
  sb.carry = malloc(needle_len);
  if (sb.buf == NULL || sb.carry == NULL) {
    free(sb.buf);
    free(sb.carry);
    return NULL;
  }
  for (;;) {
    size_t i = __sync_fetch_and_add(&scan_idx, 1);
    if (i >= npaths)
      break;
    scan_file(paths[i], &sb);
  }
  free(sb.buf);
  free(sb.carry);
  return NULL;
}

int main(int argc, char **argv) {
  if (argc < 2 || argc > 3) {
    fprintf(stderr, "usage: %s SEARCH_TERM [FILE_OR_DIRECTORY]\n", argv[0]);
    return 2;
  }

  needle = argv[1];
  needle_len = strlen(needle);
  if (needle_len == 0) {
    fprintf(stderr, "brep: empty search term\n");
    return 2;
  }

  char cwd[PATH_MAX];
  const char *root;
  if (argc == 3) {
    struct stat st;
    if (stat(argv[2], &st) != 0) {
      fprintf(stderr, "brep: %s: %s\n", argv[2], strerror(errno));
      return 2;
    }
    if (!S_ISDIR(st.st_mode) && !S_ISREG(st.st_mode)) {
      fprintf(stderr, "brep: %s: not a regular file or directory\n", argv[2]);
      return 2;
    }
    root = argv[2];
  } else {
    if (getcwd(cwd, sizeof cwd) == NULL)
      die("getcwd");
    root = cwd;
  }

  struct stat st;
  if (stat(root, &st) != 0) {
    fprintf(stderr, "brep: %s: %s\n", root, strerror(errno));
    return 2;
  }

  /* Phase 1: collect files (cheap metadata walk). */
  if (S_ISREG(st.st_mode)) {
    add_path(root);
  } else {
    walk(root);
  }

  if (npaths == 0)
    return 1; /* nothing to scan */

  /* Phase 2: parallel scan. */
  long nproc = sysconf(_SC_NPROCESSORS_ONLN);
  if (nproc < 1)
    nproc = 1;
  int nth = (int)nproc;
  if (nth > 256)
    nth = 256;

  pthread_t t[256];
  int started = 0;
  for (int i = 0; i < nth; i++) {
    if (pthread_create(&t[i], NULL, worker, NULL) != 0)
      break;
    started++;
  }
  for (int i = 0; i < started; i++)
    pthread_join(t[i], NULL);

  return found_any ? 0 : 1;
}