/**
Test if this argument contains any errors. Detected errors include
syntax errors in command substitutions, improperly escaped
characters and improper use of the variable expansion operator.
*/
parser_test_error_bits_t parse_util_detect_errors_in_argument(const parse_node_t &node, const wcstring &arg_src, parse_error_list_t *out_errors)
{
assert(node.type == symbol_argument);
int err=0;
wchar_t *paran_begin, *paran_end;
int do_loop = 1;
wcstring working_copy = arg_src;
while (do_loop)
{
const wchar_t *working_copy_cstr = working_copy.c_str();
switch (parse_util_locate_cmdsubst(working_copy_cstr,
¶n_begin,
¶n_end,
false))
{
case -1:
{
err=1;
if (out_errors)
{
append_syntax_error(out_errors, node, L"Mismatched parenthesis");
}
return err;
}
case 0:
{
do_loop = 0;
break;
}
case 1:
{
const wcstring subst(paran_begin + 1, paran_end);
// Replace the command substitution with just INTERNAL_SEPARATOR
size_t cmd_sub_start = paran_begin - working_copy_cstr;
size_t cmd_sub_len = paran_end + 1 - paran_begin;
working_copy.replace(cmd_sub_start, cmd_sub_len, wcstring(1, INTERNAL_SEPARATOR));
parse_error_list_t subst_errors;
err |= parse_util_detect_errors(subst, &subst_errors, false /* do not accept incomplete */);
/* Our command substitution produced error offsets relative to its source. Tweak the offsets of the errors in the command substitution to account for both its offset within the string, and the offset of the node */
size_t error_offset = cmd_sub_start + 1 + node.source_start;
parse_error_offset_source_start(&subst_errors, error_offset);
if (out_errors != NULL)
{
out_errors->insert(out_errors->end(), subst_errors.begin(), subst_errors.end());
}
break;
}
}
}
wcstring unesc;
if (! unescape_string(working_copy, &unesc, UNESCAPE_SPECIAL))
{
if (out_errors)
{
append_syntax_error(out_errors, node, L"Invalid token '%ls'", working_copy.c_str());
}
return 1;
}
else
{
/* Check for invalid variable expansions */
const size_t unesc_size = unesc.size();
for (size_t idx = 0; idx < unesc_size; idx++)
{
switch (unesc.at(idx))
{
case VARIABLE_EXPAND:
case VARIABLE_EXPAND_SINGLE:
{
wchar_t next_char = (idx + 1 < unesc_size ? unesc.at(idx + 1) : L'\0');
if (next_char != VARIABLE_EXPAND &&
next_char != VARIABLE_EXPAND_SINGLE &&
! wcsvarchr(next_char))
{
err=1;
if (out_errors)
{
parse_util_expand_variable_error(node, unesc, idx, node.source_start, out_errors);
}
}
break;
}
}
}
}
return err;
}
/// Expand all environment variables in the string *ptr.
///
/// This function is slow, fragile and complicated. There are lots of little corner cases, like
/// $$foo should do a double expansion, $foo$bar should not double expand bar, etc. Also, it's easy
/// to accidentally leak memory on array out of bounds errors an various other situations. All in
/// all, this function should be rewritten, split out into multiple logical units and carefully
/// tested. After that, it can probably be optimized to do fewer memory allocations, fewer string
/// scans and overall just less work. But until that happens, don't edit it unless you know exactly
/// what you are doing, and do proper testing afterwards.
///
/// This function operates on strings backwards, starting at last_idx.
///
/// Note: last_idx is considered to be where it previously finished procesisng. This means it
/// actually starts operating on last_idx-1. As such, to process a string fully, pass string.size()
/// as last_idx instead of string.size()-1.
static bool expand_variables(const wcstring &instr, std::vector<completion_t> *out, size_t last_idx,
parse_error_list_t *errors) {
const size_t insize = instr.size();
// last_idx may be 1 past the end of the string, but no further.
assert(last_idx <= insize && "Invalid last_idx");
if (last_idx == 0) {
append_completion(out, instr);
return true;
}
// Locate the last VARIABLE_EXPAND or VARIABLE_EXPAND_SINGLE
bool is_single = false;
size_t varexp_char_idx = last_idx;
while (varexp_char_idx--) {
const wchar_t c = instr.at(varexp_char_idx);
if (c == VARIABLE_EXPAND || c == VARIABLE_EXPAND_SINGLE) {
is_single = (c == VARIABLE_EXPAND_SINGLE);
break;
}
}
if (varexp_char_idx >= instr.size()) {
// No variable expand char, we're done.
append_completion(out, instr);
return true;
}
// Get the variable name.
const size_t var_name_start = varexp_char_idx + 1;
size_t var_name_stop = var_name_start;
while (var_name_stop < insize) {
const wchar_t nc = instr.at(var_name_stop);
if (nc == VARIABLE_EXPAND_EMPTY) {
var_name_stop++;
break;
}
if (!valid_var_name_char(nc)) break;
var_name_stop++;
}
assert(var_name_stop >= var_name_start && "Bogus variable name indexes");
const size_t var_name_len = var_name_stop - var_name_start;
// It's an error if the name is empty.
if (var_name_len == 0) {
if (errors) {
parse_util_expand_variable_error(instr, 0 /* global_token_pos */, varexp_char_idx,
errors);
}
return false;
}
// Get the variable name as a string, then try to get the variable from env.
const wcstring var_name(instr, var_name_start, var_name_len);
// Do a dirty hack to make sliced history fast (#4650). We expand from either a variable, or a
// history_t. Note that "history" is read only in env.cpp so it's safe to special-case it in
// this way (it cannot be shadowed, etc).
history_t *history = nullptr;
maybe_t<env_var_t> var{};
if (var_name == L"history") {
// We do this only on the main thread, matching env.cpp.
if (is_main_thread()) {
history = reader_get_history();
}
} else if (var_name != wcstring{VARIABLE_EXPAND_EMPTY}) {
var = env_get(var_name);
}
// Parse out any following slice.
// Record the end of the variable name and any following slice.
size_t var_name_and_slice_stop = var_name_stop;
bool all_values = true;
const size_t slice_start = var_name_stop;
// List of indexes, and parallel array of source positions of each index in the variable list.
std::vector<long> var_idx_list;
std::vector<size_t> var_pos_list;
if (slice_start < insize && instr.at(slice_start) == L'[') {
all_values = false;
const wchar_t *in = instr.c_str();
wchar_t *slice_end;
// If a variable is missing, behave as though we have one value, so that $var[1] always
// works.
size_t effective_val_count = 1;
if (var) {
effective_val_count = var->as_list().size();
} else if (history) {
effective_val_count = history->size();
}
size_t bad_pos = parse_slice(in + slice_start, &slice_end, var_idx_list, var_pos_list,
effective_val_count);
if (bad_pos != 0) {
append_syntax_error(errors, slice_start + bad_pos, L"Invalid index value");
return false;
}
var_name_and_slice_stop = (slice_end - in);
}
if (!var && !history) {
// Expanding a non-existent variable.
if (!is_single) {
// Normal expansions of missing variables successfully expand to nothing.
return true;
} else {
// Expansion to single argument.
// Replace the variable name and slice with VARIABLE_EXPAND_EMPTY.
wcstring res(instr, 0, varexp_char_idx);
if (!res.empty() && res.back() == VARIABLE_EXPAND_SINGLE) {
res.push_back(VARIABLE_EXPAND_EMPTY);
}
res.append(instr, var_name_and_slice_stop, wcstring::npos);
return expand_variables(res, out, varexp_char_idx, errors);
}
}
// Ok, we have a variable or a history. Let's expand it.
// Start by respecting the sliced elements.
assert((var || history) && "Should have variable or history here");
wcstring_list_t var_item_list;
if (all_values) {
if (history) {
history->get_history(var_item_list);
} else {
var->to_list(var_item_list);
}
} else {
// We have to respect the slice.
if (history) {
// Ask history to map indexes to item strings.
// Note this may have missing entries for out-of-bounds.
auto item_map = history->items_at_indexes(var_idx_list);
for (long item_index : var_idx_list) {
auto iter = item_map.find(item_index);
if (iter != item_map.end()) {
var_item_list.push_back(iter->second);
}
}
} else {
const wcstring_list_t &all_var_items = var->as_list();
for (long item_index : var_idx_list) {
// Check that we are within array bounds. If not, skip the element. Note:
// Negative indices (`echo $foo[-1]`) are already converted to positive ones
// here, So tmp < 1 means it's definitely not in.
// Note we are 1-based.
if (item_index >= 1 && size_t(item_index) <= all_var_items.size()) {
var_item_list.push_back(all_var_items.at(item_index - 1));
}
}
}
}
if (is_single) {
wcstring res(instr, 0, varexp_char_idx);
if (!res.empty()) {
if (res.back() != VARIABLE_EXPAND_SINGLE) {
res.push_back(INTERNAL_SEPARATOR);
} else if (var_item_list.empty() || var_item_list.front().empty()) {
// First expansion is empty, but we need to recursively expand.
res.push_back(VARIABLE_EXPAND_EMPTY);
}
}
// Append all entries in var_item_list, separated by spaces.
// Remove the last space.
if (!var_item_list.empty()) {
for (const wcstring &item : var_item_list) {
res.append(item);
res.push_back(L' ');
}
res.pop_back();
}
res.append(instr, var_name_and_slice_stop, wcstring::npos);
return expand_variables(res, out, varexp_char_idx, errors);
} else {
// Normal cartesian-product expansion.
for (const wcstring &item : var_item_list) {
if (varexp_char_idx == 0 && var_name_and_slice_stop == insize) {
append_completion(out, item);
} else {
wcstring new_in(instr, 0, varexp_char_idx);
if (!new_in.empty()) {
if (new_in.back() != VARIABLE_EXPAND) {
new_in.push_back(INTERNAL_SEPARATOR);
} else if (item.empty()) {
new_in.push_back(VARIABLE_EXPAND_EMPTY);
}
}
new_in.append(item);
new_in.append(instr, var_name_and_slice_stop, wcstring::npos);
if (!expand_variables(new_in, out, varexp_char_idx, errors)) {
return false;
}
}
}
}
return true;
}
