blob: e4cd7d451a467b0ae76e186d8fb487bca0bf4512 [file] [log] [blame]
// Copyright 2014 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "gn/c_include_iterator.h"
#include <iterator>
#include "base/logging.h"
#include "base/strings/string_util.h"
#include "gn/input_file.h"
#include "gn/location.h"
namespace {
enum IncludeType {
INCLUDE_NONE,
INCLUDE_SYSTEM, // #include <...>
INCLUDE_USER // #include "..."
};
// Returns a new string piece referencing the same buffer as the argument, but
// with leading space trimmed. This only checks for space and tab characters
// since we're dealing with lines in C source files.
std::string_view TrimLeadingWhitespace(std::string_view str) {
size_t new_begin = 0;
while (new_begin < str.size() &&
(str[new_begin] == ' ' || str[new_begin] == '\t'))
new_begin++;
return str.substr(new_begin);
}
// We don't want to count comment lines and preprocessor lines toward our
// "max lines to look at before giving up" since the beginnings of some files
// may have a lot of comments.
//
// We only handle C-style "//" comments since this is the normal commenting
// style used in Chrome, and do so pretty stupidly. We don't want to write a
// full C++ parser here, we're just trying to get a good heuristic for checking
// the file.
//
// We assume the line has leading whitespace trimmed. We also assume that empty
// lines have already been filtered out.
bool ShouldCountTowardNonIncludeLines(std::string_view line) {
if (base::StartsWith(line, "//", base::CompareCase::SENSITIVE))
return false; // Don't count comments.
if (base::StartsWith(line, "/*", base::CompareCase::SENSITIVE) ||
base::StartsWith(line, " *", base::CompareCase::SENSITIVE))
return false; // C-style comment blocks with stars along the left side.
if (base::StartsWith(line, "#", base::CompareCase::SENSITIVE))
return false; // Don't count preprocessor.
if (base::ContainsOnlyChars(line, base::kWhitespaceASCII))
return false; // Don't count whitespace lines.
return true; // Count everything else.
}
// Given a line, checks to see if it looks like an include or import and
// extract the path. The type of include is returned. Returns INCLUDE_NONE on
// error or if this is not an include line.
//
// The 1-based character number on the line that the include was found at
// will be filled into *begin_char.
IncludeType ExtractInclude(std::string_view line,
std::string_view* path,
int* begin_char) {
static const char kInclude[] = "include";
static const size_t kIncludeLen = std::size(kInclude) - 1; // No null.
static const char kImport[] = "import";
static const size_t kImportLen = std::size(kImport) - 1; // No null.
std::string_view trimmed = TrimLeadingWhitespace(line);
if (trimmed.empty())
return INCLUDE_NONE;
if (trimmed[0] != '#')
return INCLUDE_NONE;
trimmed = TrimLeadingWhitespace(trimmed.substr(1));
std::string_view contents;
if (base::StartsWith(trimmed, std::string_view(kInclude, kIncludeLen),
base::CompareCase::SENSITIVE))
contents = TrimLeadingWhitespace(trimmed.substr(kIncludeLen));
else if (base::StartsWith(trimmed, std::string_view(kImport, kImportLen),
base::CompareCase::SENSITIVE))
contents = TrimLeadingWhitespace(trimmed.substr(kImportLen));
if (contents.empty())
return INCLUDE_NONE;
IncludeType type = INCLUDE_NONE;
char terminating_char = 0;
if (contents[0] == '"') {
type = INCLUDE_USER;
terminating_char = '"';
} else if (contents[0] == '<') {
type = INCLUDE_SYSTEM;
terminating_char = '>';
} else {
return INCLUDE_NONE;
}
// Count everything to next "/> as the contents.
size_t terminator_index = contents.find(terminating_char, 1);
if (terminator_index == std::string_view::npos)
return INCLUDE_NONE;
*path = contents.substr(1, terminator_index - 1);
// Note: one based so we do "+ 1".
*begin_char = static_cast<int>(path->data() - line.data()) + 1;
return type;
}
// Returns true if this line has a "nogncheck" comment associated with it.
bool HasNoCheckAnnotation(std::string_view line) {
return line.find("nogncheck") != std::string_view::npos;
}
} // namespace
const int CIncludeIterator::kMaxNonIncludeLines = 10;
CIncludeIterator::CIncludeIterator(const InputFile* input)
: input_file_(input), file_(input->contents()) {}
CIncludeIterator::~CIncludeIterator() = default;
bool CIncludeIterator::GetNextIncludeString(
IncludeStringWithLocation* include) {
std::string_view line;
int cur_line_number = 0;
while (lines_since_last_include_ <= kMaxNonIncludeLines &&
GetNextLine(&line, &cur_line_number)) {
std::string_view include_contents;
int begin_char;
IncludeType type = ExtractInclude(line, &include_contents, &begin_char);
if (HasNoCheckAnnotation(line))
continue;
if (type != INCLUDE_NONE) {
include->contents = include_contents;
include->location = LocationRange(
Location(input_file_, cur_line_number, begin_char),
Location(input_file_, cur_line_number,
begin_char + static_cast<int>(include_contents.size())));
include->system_style_include = (type == INCLUDE_SYSTEM);
lines_since_last_include_ = 0;
return true;
}
if (ShouldCountTowardNonIncludeLines(line))
lines_since_last_include_++;
}
return false;
}
bool CIncludeIterator::GetNextLine(std::string_view* line, int* line_number) {
if (offset_ == file_.size())
return false;
size_t begin = offset_;
while (offset_ < file_.size() && file_[offset_] != '\n')
offset_++;
line_number_++;
*line = file_.substr(begin, offset_ - begin);
*line_number = line_number_;
// If we didn't hit EOF, skip past the newline for the next one.
if (offset_ < file_.size())
offset_++;
return true;
}