ccache.c

ccache 是一个快速的编译器缓存。当您编译一个程序的时候

/*
  a re-implementation of the compilercache scripts in C

  The idea is based on the shell-script compilercache by Erik Thiele <[e-mail]>

   Copyright (C) Andrew Tridgell 2002
   Copyright (C) Martin Pool 2003
   
   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 2 of the License, or
   (at your option) any later version.
   
   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.
   
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/

#include "ccache.h"

/* the base cache directory */
char *cache_dir = NULL;

/* the debug logfile name, if set */
char *cache_logfile = NULL;

/* the argument list after processing */
static ARGS *stripped_args;

/* the original argument list */
static ARGS *orig_args;

/* the output filename being compiled to */
static char *output_file;

/* the source file */
static char *input_file;

/* the name of the file containing the cached object code */
static char *hashname;

/* the extension of the file after pre-processing */
static const char *i_extension;

/* the name of the temporary pre-processor file */
static char *i_tmpfile;

/* are we compiling a .i or .ii file directly? */
static int direct_i_file;

/* the name of the cpp stderr file */
static char *cpp_stderr;

/* the name of the statistics file */
char *stats_file = NULL;

/* can we safely use the unification hashing backend? */
static int enable_unify;

/* a list of supported file extensions, and the equivalent
   extension for code that has been through the pre-processor
*/
static struct {
	char *extension;
	char *i_extension;
} extensions[] = {
	{"c", "i"},
	{"C", "ii"},
	{"m", "mi"},
	{"cc", "ii"},
	{"CC", "ii"},
	{"cpp", "ii"},
	{"CPP", "ii"},
	{"cxx", "ii"},
	{"CXX", "ii"},
	{"c++", "ii"},
	{"C++", "ii"},
	{"i", "i"},
	{"ii", "ii"},
	{NULL, NULL}};

/*
  something went badly wrong - just execute the real compiler
*/
static void failed(void)
{
	char *e;

	/* delete intermediate pre-processor file if needed */
	if (i_tmpfile) {
		if (!direct_i_file) {
			unlink(i_tmpfile);
		}
		free(i_tmpfile);
		i_tmpfile = NULL;
	}

	/* delete the cpp stderr file if necessary */
	if (cpp_stderr) {
		unlink(cpp_stderr);
		free(cpp_stderr);
		cpp_stderr = NULL;
	}

	/* strip any local args */
	args_strip(orig_args, "--ccache-");

	if ((e=getenv("CCACHE_PREFIX"))) {
		char *p = find_executable(e, MYNAME);
		if (!p) {
			perror(e);
			exit(1);
		}
		args_add_prefix(orig_args, p);
	}

	execv(orig_args->argv[0], orig_args->argv);
	cc_log("execv returned (%s)!\n", strerror(errno));
	perror(orig_args->argv[0]);
	exit(1);
}


/* return a string to be used to distinguish temporary files 
   this also tries to cope with NFS by adding the local hostname 
*/
static const char *tmp_string(void)
{
	static char *ret;

	if (!ret) {
		char hostname[200];
		strcpy(hostname, "unknown");
#if HAVE_GETHOSTNAME
		gethostname(hostname, sizeof(hostname)-1);
#endif
		hostname[sizeof(hostname)-1] = 0;
		asprintf(&ret, "%s.%u", hostname, (unsigned)getpid());
	}

	return ret;
}


/* run the real compiler and put the result in cache */
static void to_cache(ARGS *args)
{
	char *path_stderr;
	char *tmp_stdout, *tmp_stderr, *tmp_hashname;
	struct stat st1, st2;
	int status;

	x_asprintf(&tmp_stdout, "%s/tmp.stdout.%s", cache_dir, tmp_string());
	x_asprintf(&tmp_stderr, "%s/tmp.stderr.%s", cache_dir, tmp_string());
	x_asprintf(&tmp_hashname, "%s/tmp.hash.%s.o", cache_dir, tmp_string());

	args_add(args, "-o");
	args_add(args, tmp_hashname);

	/* Turn off DEPENDENCIES_OUTPUT when running cc1, because
	 * otherwise it will emit a line like
	 *
	 *  tmp.stdout.vexed.732.o: /home/mbp/.ccache/tmp.stdout.vexed.732.i
	 *
	 * unsetenv() is on BSD and Linux but not portable. */
	putenv("DEPENDENCIES_OUTPUT");
	
	if (getenv("CCACHE_CPP2")) {
		args_add(args, input_file);
	} else {
		args_add(args, i_tmpfile);
	}
	status = execute(args->argv, tmp_stdout, tmp_stderr);
	args_pop(args, 3);

	if (stat(tmp_stdout, &st1) != 0 || st1.st_size != 0) {
		cc_log("compiler produced stdout for %s\n", output_file);
		stats_update(STATS_STDOUT);
		unlink(tmp_stdout);
		unlink(tmp_stderr);
		unlink(tmp_hashname);
		failed();
	}
	unlink(tmp_stdout);

	if (status != 0) {
		int fd;
		cc_log("compile of %s gave status = %d\n", output_file, status);
		stats_update(STATS_STATUS);

		fd = open(tmp_stderr, O_RDONLY);
		if (fd != -1) {
			if (strcmp(output_file, "/dev/null") == 0 ||
			    rename(tmp_hashname, output_file) == 0 || errno == ENOENT) {
				if (cpp_stderr) {
					/* we might have some stderr from cpp */
					int fd2 = open(cpp_stderr, O_RDONLY);
					if (fd2 != -1) {
						copy_fd(fd2, 2);
						close(fd2);
						unlink(cpp_stderr);
						cpp_stderr = NULL;
					}
				}

				/* we can use a quick method of
                                   getting the failed output */
				copy_fd(fd, 2);
				close(fd);
				unlink(tmp_stderr);
				if (i_tmpfile && !direct_i_file) {
					unlink(i_tmpfile);
				}
				exit(status);
			}
		}
		
		unlink(tmp_stderr);
		unlink(tmp_hashname);
		failed();
	}

	x_asprintf(&path_stderr, "%s.stderr", hashname);

	if (stat(tmp_stderr, &st1) != 0 ||
	    stat(tmp_hashname, &st2) != 0 ||
	    rename(tmp_hashname, hashname) != 0 ||
	    rename(tmp_stderr, path_stderr) != 0) {
		cc_log("failed to rename tmp files - %s\n", strerror(errno));
		stats_update(STATS_ERROR);
		failed();
	}

	cc_log("Placed %s into cache\n", output_file);
	stats_tocache(file_size(&st1) + file_size(&st2));

	free(tmp_hashname);
	free(tmp_stderr);
	free(tmp_stdout);
	free(path_stderr);
}

/* find the hash for a command. The hash includes all argument lists,
   plus the output from running the compiler with -E */
static void find_hash(ARGS *args)
{
	int i;
	char *path_stdout, *path_stderr;
	char *hash_dir;
	char *s;
	struct stat st;
	int status;
	int nlevels = 2;
	char *input_base;
	char *tmp;
	
	if ((s = getenv("CCACHE_NLEVELS"))) {
		nlevels = atoi(s);
		if (nlevels < 1) nlevels = 1;
		if (nlevels > 8) nlevels = 8;
	}

	hash_start();

	/* when we are doing the unifying tricks we need to include
           the input file name in the hash to get the warnings right */
	if (enable_unify) {
		hash_string(input_file);
	}

	/* we have to hash the extension, as a .i file isn't treated the same
	   by the compiler as a .ii file */
	hash_string(i_extension);

	/* first the arguments */
	for (i=1;i<args->argc;i++) {
		/* some arguments don't contribute to the hash. The
		   theory is that these arguments will change the
		   output of -E if they are going to have any effect
		   at all, or they only affect linking */
		if (i < args->argc-1) {
			if (strcmp(args->argv[i], "-I") == 0 ||
			    strcmp(args->argv[i], "-include") == 0 ||
			    strcmp(args->argv[i], "-L") == 0 ||
			    strcmp(args->argv[i], "-D") == 0 ||
			    strcmp(args->argv[i], "-idirafter") == 0 ||
			    strcmp(args->argv[i], "-isystem") == 0) {
				i++;
				continue;
			}
		}
		if (strncmp(args->argv[i], "-I", 2) == 0 ||
		    strncmp(args->argv[i], "-L", 2) == 0 ||
		    strncmp(args->argv[i], "-D", 2) == 0 ||
		    strncmp(args->argv[i], "-idirafter", 10) == 0 ||
		    strncmp(args->argv[i], "-isystem", 8) == 0) {
			continue;
		}

		if (strncmp(args->argv[i], "--specs=", 8) == 0 &&
		    stat(args->argv[i]+8, &st) == 0) {
			/* if given a explicit specs file, then hash that file, but
			   don't include the path to it in the hash */
			hash_file(args->argv[i]+8);
			continue;
		}

		/* all other arguments are included in the hash */
		hash_string(args->argv[i]);
	}

	/* the compiler driver size and date. This is a simple minded way
	   to try and detect compiler upgrades. It is not 100% reliable */
	if (stat(args->argv[0], &st) != 0) {
		cc_log("Couldn't stat the compiler!? (argv[0]='%s')\n", args->argv[0]);
		stats_update(STATS_COMPILER);
		failed();
	}
	hash_int(st.st_size);
	hash_int(st.st_mtime);

	/* possibly hash the current working directory */
	if (getenv("CCACHE_HASHDIR")) {
		char *cwd = gnu_getcwd();
		if (cwd) {
			hash_string(cwd);
			free(cwd);
		}
	}

	/* ~/hello.c -> tmp.hello.123.i 
	   limit the basename to 10
	   characters in order to cope with filesystem with small
	   maximum filename length limits */
	input_base = str_basename(input_file);
	tmp = strchr(input_base, '.');
	if (tmp != NULL) {
		*tmp = 0;
	}
	if (strlen(input_base) > 10) {
		input_base[10] = 0;
	}

	/* now the run */
	x_asprintf(&path_stdout, "%s/%s.tmp.%s.%s", cache_dir,
		   input_base, tmp_string(), 
		   i_extension);
	x_asprintf(&path_stderr, "%s/tmp.cpp_stderr.%s", cache_dir, tmp_string());

	if (!direct_i_file) {
		/* run cpp on the input file to obtain the .i */
		args_add(args, "-E");
		args_add(args, input_file);
		status = execute(args->argv, path_stdout, path_stderr);
		args_pop(args, 2);
	} else {
		/* we are compiling a .i or .ii file - that means we
		   can skip the cpp stage and directly form the
		   correct i_tmpfile */
		path_stdout = input_file;
		if (create_empty_file(path_stderr) != 0) {
			stats_update(STATS_ERROR);
			cc_log("failed to create empty stderr file\n");
			failed();
		}
		status = 0;
	}

	if (status != 0) {
		if (!direct_i_file) {
			unlink(path_stdout);
		}
		unlink(path_stderr);
		cc_log("the preprocessor gave %d\n", status);
		stats_update(STATS_PREPROCESSOR);
		failed();
	}

	/* if the compilation is with -g then we have to include the whole of the
	   preprocessor output, which means we are sensitive to line number
	   information. Otherwise we can discard line number info, which makes
	   us less sensitive to reformatting changes 

	   Note! I have now disabled the unification code by default
	   as it gives the wrong line numbers for warnings. Pity.
	*/
	if (!enable_unify) {
		hash_file(path_stdout);
	} else {
		if (unify_hash(path_stdout) != 0) {
			stats_update(STATS_ERROR);
			failed();
		}
	}
	hash_file(path_stderr);

	i_tmpfile = path_stdout;

	if (!getenv("CCACHE_CPP2")) {
		/* if we are using the CPP trick then we need to remember this stderr
		   data and output it just before the main stderr from the compiler
		   pass */
		cpp_stderr = path_stderr;
	} else {	
		unlink(path_stderr);
		free(path_stderr);
	}

	/* we use a N level subdir for the cache path to reduce the impact
	   on filesystems which are slow for large directories
	*/
	s = hash_result();
	x_asprintf(&hash_dir, "%s/%c", cache_dir, s[0]);
	x_asprintf(&stats_file, "%s/stats", hash_dir);
	for (i=1; i<nlevels; i++) {
		char *p;
		if (create_dir(hash_dir) != 0) {
			cc_log("failed to create %s\n", hash_dir);
			failed();
		}
		x_asprintf(&p, "%s/%c", hash_dir, s[i]);
		free(hash_dir);
		hash_dir = p;
	}
	if (create_dir(hash_dir) != 0) {
		cc_log("failed to create %s\n", hash_dir);
		failed();
	}
	x_asprintf(&hashname, "%s/%s", hash_dir, s+nlevels);
	free(hash_dir);
}


/* 
   try to return the compile result from cache. If we can return from
   cache then this function exits with the correct status code,
   otherwise it returns */
static void from_cache(int first)
{
	int fd_stderr, fd_cpp_stderr;
	char *stderr_file;
	int ret;
	struct stat st;

	x_asprintf(&stderr_file, "%s.stderr", hashname);
	fd_stderr = open(stderr_file, O_RDONLY);
	if (fd_stderr == -1) {
		/* it isn't in cache ... */
		free(stderr_file);
		return;
	}

	/* make sure the output is there too */
	if (stat(hashname, &st) != 0) {
		close(fd_stderr);
		unlink(stderr_file);
		free(stderr_file);
		return;
	}

	/* the user might be disabling cache hits */
	if (first && getenv("CCACHE_RECACHE")) {
		close(fd_stderr);
		unlink(stderr_file);
		free(stderr_file);
		return;
	}

	utime(stderr_file, NULL);

	if (strcmp(output_file, "/dev/null") == 0) {
		ret = 0;
	} else {
		unlink(output_file);
		if (getenv("CCACHE_HARDLINK")) {
			ret = link(hashname, output_file);
		} else {
			ret = copy_file(hashname, output_file);
		}
	}

	/* the hash file might have been deleted by some external process */
	if (ret == -1 && errno == ENOENT) {
		cc_log("hashfile missing for %s\n", output_file);
		stats_update(STATS_MISSING);
		close(fd_stderr);
		unlink(stderr_file);
		return;
	}
	free(stderr_file);

	if (ret == -1) {
		ret = copy_file(hashname, output_file);
		if (ret == -1) {
			cc_log("failed to copy %s -> %s (%s)\n", 
			       hashname, output_file, strerror(errno));
			stats_update(STATS_ERROR);
			failed();
		}
	}
	if (ret == 0) {
		/* update the mtime on the file so that make doesn't get confused */
		utime(output_file, NULL);