forked from openkylin/platform_build
474 lines
12 KiB
C++
474 lines
12 KiB
C++
#include "files.h"
|
|
#include <stdio.h>
|
|
#include <string.h>
|
|
#include <stdlib.h>
|
|
#include <errno.h>
|
|
#include <sys/stat.h>
|
|
#include <unistd.h>
|
|
#include <dirent.h>
|
|
#include <fnmatch.h>
|
|
#include <string.h>
|
|
#include <stdlib.h>
|
|
|
|
static bool
|
|
is_comment_line(const char* p)
|
|
{
|
|
while (*p && isspace(*p)) {
|
|
p++;
|
|
}
|
|
return *p == '#';
|
|
}
|
|
|
|
static string
|
|
path_append(const string& base, const string& leaf)
|
|
{
|
|
string full = base;
|
|
if (base.length() > 0 && leaf.length() > 0) {
|
|
full += '/';
|
|
}
|
|
full += leaf;
|
|
return full;
|
|
}
|
|
|
|
static bool
|
|
is_whitespace_line(const char* p)
|
|
{
|
|
while (*p) {
|
|
if (!isspace(*p)) {
|
|
return false;
|
|
}
|
|
p++;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
static bool
|
|
is_exclude_line(const char* p) {
|
|
while (*p) {
|
|
if (*p == '-') {
|
|
return true;
|
|
}
|
|
else if (isspace(*p)) {
|
|
p++;
|
|
}
|
|
else {
|
|
return false;
|
|
}
|
|
}
|
|
return false;
|
|
}
|
|
|
|
void
|
|
split_line(const char* p, vector<string>* out)
|
|
{
|
|
const char* q = p;
|
|
enum { WHITE, TEXT, IN_QUOTE } state = WHITE;
|
|
while (*p) {
|
|
if (*p == '#') {
|
|
break;
|
|
}
|
|
|
|
switch (state)
|
|
{
|
|
case WHITE:
|
|
if (!isspace(*p)) {
|
|
q = p;
|
|
state = (*p == '"') ? IN_QUOTE : TEXT;
|
|
}
|
|
break;
|
|
case IN_QUOTE:
|
|
if (*p == '"') {
|
|
state = TEXT;
|
|
break;
|
|
}
|
|
[[fallthrough]];
|
|
case TEXT:
|
|
if (state != IN_QUOTE && isspace(*p)) {
|
|
if (q != p) {
|
|
const char* start = q;
|
|
size_t len = p-q;
|
|
if (len > 2 && *start == '"' && start[len - 1] == '"') {
|
|
start++;
|
|
len -= 2;
|
|
}
|
|
out->push_back(string(start, len));
|
|
}
|
|
state = WHITE;
|
|
}
|
|
break;
|
|
}
|
|
p++;
|
|
}
|
|
if (state == TEXT) {
|
|
const char* start = q;
|
|
size_t len = p-q;
|
|
if (len > 2 && *start == '"' && start[len - 1] == '"') {
|
|
start++;
|
|
len -= 2;
|
|
}
|
|
out->push_back(string(start, len));
|
|
}
|
|
}
|
|
|
|
static void
|
|
add_file(vector<FileRecord>* files, const FileOpType fileOp,
|
|
const string& listFile, int listLine,
|
|
const string& sourceName, const string& outName)
|
|
{
|
|
FileRecord rec;
|
|
rec.listFile = listFile;
|
|
rec.listLine = listLine;
|
|
rec.fileOp = fileOp;
|
|
rec.sourceName = sourceName;
|
|
rec.outName = outName;
|
|
files->push_back(rec);
|
|
}
|
|
|
|
static string
|
|
replace_variables(const string& input,
|
|
const map<string, string>& variables,
|
|
bool* error) {
|
|
if (variables.empty()) {
|
|
return input;
|
|
}
|
|
|
|
// Abort if the variable prefix is not found
|
|
if (input.find("${") == string::npos) {
|
|
return input;
|
|
}
|
|
|
|
string result = input;
|
|
|
|
// Note: rather than be fancy to detect recursive replacements,
|
|
// we simply iterate till a given threshold is met.
|
|
|
|
int retries = 1000;
|
|
bool did_replace;
|
|
|
|
do {
|
|
did_replace = false;
|
|
for (map<string, string>::const_iterator it = variables.begin();
|
|
it != variables.end(); ++it) {
|
|
string::size_type pos = 0;
|
|
while((pos = result.find(it->first, pos)) != string::npos) {
|
|
result = result.replace(pos, it->first.length(), it->second);
|
|
pos += it->second.length();
|
|
did_replace = true;
|
|
}
|
|
}
|
|
if (did_replace && --retries == 0) {
|
|
*error = true;
|
|
fprintf(stderr, "Recursive replacement detected during variables "
|
|
"substitution. Full list of variables is: ");
|
|
|
|
for (map<string, string>::const_iterator it = variables.begin();
|
|
it != variables.end(); ++it) {
|
|
fprintf(stderr, " %s=%s\n",
|
|
it->first.c_str(), it->second.c_str());
|
|
}
|
|
|
|
return result;
|
|
}
|
|
} while (did_replace);
|
|
|
|
return result;
|
|
}
|
|
|
|
int
|
|
read_list_file(const string& filename,
|
|
const map<string, string>& variables,
|
|
vector<FileRecord>* files,
|
|
vector<string>* excludes)
|
|
{
|
|
int err = 0;
|
|
FILE* f = NULL;
|
|
long size;
|
|
char* buf = NULL;
|
|
char *p, *q;
|
|
int i, lineCount;
|
|
|
|
f = fopen(filename.c_str(), "r");
|
|
if (f == NULL) {
|
|
fprintf(stderr, "Could not open list file (%s): %s\n",
|
|
filename.c_str(), strerror(errno));
|
|
err = errno;
|
|
goto cleanup;
|
|
}
|
|
|
|
err = fseek(f, 0, SEEK_END);
|
|
if (err != 0) {
|
|
fprintf(stderr, "Could not seek to the end of file %s. (%s)\n",
|
|
filename.c_str(), strerror(errno));
|
|
err = errno;
|
|
goto cleanup;
|
|
}
|
|
|
|
size = ftell(f);
|
|
|
|
err = fseek(f, 0, SEEK_SET);
|
|
if (err != 0) {
|
|
fprintf(stderr, "Could not seek to the beginning of file %s. (%s)\n",
|
|
filename.c_str(), strerror(errno));
|
|
err = errno;
|
|
goto cleanup;
|
|
}
|
|
|
|
buf = (char*)malloc(size+1);
|
|
if (buf == NULL) {
|
|
// (potentially large)
|
|
fprintf(stderr, "out of memory (%ld)\n", size);
|
|
err = ENOMEM;
|
|
goto cleanup;
|
|
}
|
|
|
|
if (1 != fread(buf, size, 1, f)) {
|
|
fprintf(stderr, "error reading file %s. (%s)\n",
|
|
filename.c_str(), strerror(errno));
|
|
err = errno;
|
|
goto cleanup;
|
|
}
|
|
|
|
// split on lines
|
|
p = buf;
|
|
q = buf+size;
|
|
lineCount = 0;
|
|
while (p<q) {
|
|
if (*p == '\r' || *p == '\n') {
|
|
*p = '\0';
|
|
lineCount++;
|
|
}
|
|
p++;
|
|
}
|
|
|
|
// read lines
|
|
p = buf;
|
|
for (i=0; i<lineCount; i++) {
|
|
int len = strlen(p);
|
|
q = p + len + 1;
|
|
if (is_whitespace_line(p) || is_comment_line(p)) {
|
|
;
|
|
}
|
|
else if (is_exclude_line(p)) {
|
|
while (*p != '-') p++;
|
|
p++;
|
|
excludes->push_back(string(p));
|
|
}
|
|
else {
|
|
vector<string> words;
|
|
|
|
split_line(p, &words);
|
|
|
|
#if 0
|
|
printf("[ ");
|
|
for (size_t k=0; k<words.size(); k++) {
|
|
printf("'%s' ", words[k].c_str());
|
|
}
|
|
printf("]\n");
|
|
#endif
|
|
FileOpType op = FILE_OP_COPY;
|
|
string paths[2];
|
|
int pcount = 0;
|
|
string errstr;
|
|
for (vector<string>::iterator it = words.begin(); it != words.end(); ++it) {
|
|
const string& word = *it;
|
|
if (word == "rm") {
|
|
if (op != FILE_OP_COPY) {
|
|
errstr = "Error: you can only specifiy 'rm' or 'strip' once per line.";
|
|
break;
|
|
}
|
|
op = FILE_OP_REMOVE;
|
|
} else if (word == "strip") {
|
|
if (op != FILE_OP_COPY) {
|
|
errstr = "Error: you can only specifiy 'rm' or 'strip' once per line.";
|
|
break;
|
|
}
|
|
op = FILE_OP_STRIP;
|
|
} else if (pcount < 2) {
|
|
bool error = false;
|
|
paths[pcount++] = replace_variables(word, variables, &error);
|
|
if (error) {
|
|
err = 1;
|
|
goto cleanup;
|
|
}
|
|
} else {
|
|
errstr = "Error: More than 2 paths per line.";
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (pcount == 0 && !errstr.empty()) {
|
|
errstr = "Error: No path found on line.";
|
|
}
|
|
|
|
if (!errstr.empty()) {
|
|
fprintf(stderr, "%s:%d: bad format: %s\n%s\nExpected: [SRC] [rm|strip] DEST\n",
|
|
filename.c_str(), i+1, p, errstr.c_str());
|
|
err = 1;
|
|
} else {
|
|
if (pcount == 1) {
|
|
// pattern: [rm|strip] DEST
|
|
paths[1] = paths[0];
|
|
}
|
|
|
|
add_file(files, op, filename, i+1, paths[0], paths[1]);
|
|
}
|
|
}
|
|
p = q;
|
|
}
|
|
|
|
cleanup:
|
|
if (buf != NULL) {
|
|
free(buf);
|
|
}
|
|
if (f != NULL) {
|
|
fclose(f);
|
|
}
|
|
return err;
|
|
}
|
|
|
|
|
|
int
|
|
locate(FileRecord* rec, const vector<string>& search)
|
|
{
|
|
if (rec->fileOp == FILE_OP_REMOVE) {
|
|
// Don't touch source files when removing a destination.
|
|
rec->sourceMod = 0;
|
|
rec->sourceSize = 0;
|
|
rec->sourceIsDir = false;
|
|
return 0;
|
|
}
|
|
|
|
int err;
|
|
|
|
for (vector<string>::const_iterator it=search.begin();
|
|
it!=search.end(); it++) {
|
|
string full = path_append(*it, rec->sourceName);
|
|
struct stat st;
|
|
err = stat(full.c_str(), &st);
|
|
if (err == 0) {
|
|
rec->sourceBase = *it;
|
|
rec->sourcePath = full;
|
|
rec->sourceMod = st.st_mtime;
|
|
rec->sourceSize = st.st_size;
|
|
rec->sourceIsDir = S_ISDIR(st.st_mode);
|
|
return 0;
|
|
}
|
|
}
|
|
|
|
fprintf(stderr, "%s:%d: couldn't locate source file: %s\n",
|
|
rec->listFile.c_str(), rec->listLine, rec->sourceName.c_str());
|
|
return 1;
|
|
}
|
|
|
|
void
|
|
stat_out(const string& base, FileRecord* rec)
|
|
{
|
|
rec->outPath = path_append(base, rec->outName);
|
|
|
|
int err;
|
|
struct stat st;
|
|
err = stat(rec->outPath.c_str(), &st);
|
|
if (err == 0) {
|
|
rec->outMod = st.st_mtime;
|
|
rec->outSize = st.st_size;
|
|
rec->outIsDir = S_ISDIR(st.st_mode);
|
|
} else {
|
|
rec->outMod = 0;
|
|
rec->outSize = 0;
|
|
rec->outIsDir = false;
|
|
}
|
|
}
|
|
|
|
string
|
|
dir_part(const string& filename)
|
|
{
|
|
int pos = filename.rfind('/');
|
|
if (pos <= 0) {
|
|
return ".";
|
|
}
|
|
return filename.substr(0, pos);
|
|
}
|
|
|
|
static void
|
|
add_more(const string& entry, bool isDir,
|
|
const FileRecord& rec, vector<FileRecord>*more)
|
|
{
|
|
FileRecord r;
|
|
r.listFile = rec.listFile;
|
|
r.listLine = rec.listLine;
|
|
r.sourceName = path_append(rec.sourceName, entry);
|
|
r.sourcePath = path_append(rec.sourceBase, r.sourceName);
|
|
struct stat st;
|
|
int err = stat(r.sourcePath.c_str(), &st);
|
|
if (err == 0) {
|
|
r.sourceMod = st.st_mtime;
|
|
}
|
|
r.sourceIsDir = isDir;
|
|
r.outName = path_append(rec.outName, entry);
|
|
more->push_back(r);
|
|
}
|
|
|
|
static bool
|
|
matches_excludes(const char* file, const vector<string>& excludes)
|
|
{
|
|
for (vector<string>::const_iterator it=excludes.begin();
|
|
it!=excludes.end(); it++) {
|
|
if (0 == fnmatch(it->c_str(), file, FNM_PERIOD)) {
|
|
return true;
|
|
}
|
|
}
|
|
return false;
|
|
}
|
|
|
|
static int
|
|
list_dir(const string& path, const FileRecord& rec,
|
|
const vector<string>& excludes,
|
|
vector<FileRecord>* more)
|
|
{
|
|
string full = path_append(rec.sourceBase, rec.sourceName);
|
|
full = path_append(full, path);
|
|
|
|
DIR *d = opendir(full.c_str());
|
|
if (d == NULL) {
|
|
return errno;
|
|
}
|
|
|
|
vector<string> dirs;
|
|
|
|
struct dirent *ent;
|
|
while (NULL != (ent = readdir(d))) {
|
|
if (0 == strcmp(".", ent->d_name)
|
|
|| 0 == strcmp("..", ent->d_name)) {
|
|
continue;
|
|
}
|
|
if (matches_excludes(ent->d_name, excludes)) {
|
|
continue;
|
|
}
|
|
string entry = path_append(path, ent->d_name);
|
|
bool is_directory = (ent->d_type == DT_DIR);
|
|
add_more(entry, is_directory, rec, more);
|
|
if (is_directory) {
|
|
dirs.push_back(entry);
|
|
}
|
|
}
|
|
closedir(d);
|
|
|
|
for (vector<string>::iterator it=dirs.begin(); it!=dirs.end(); it++) {
|
|
list_dir(*it, rec, excludes, more);
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
int
|
|
list_dir(const FileRecord& rec, const vector<string>& excludes,
|
|
vector<FileRecord>* files)
|
|
{
|
|
return list_dir("", rec, excludes, files);
|
|
}
|
|
|
|
FileRecord::FileRecord() {
|
|
fileOp = FILE_OP_COPY;
|
|
}
|
|
|