Source code

Revision control

Other Tools

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*-
* vim: set ts=8 sts=2 et sw=2 tw=80:
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#include "vm/Printer.h"
#include "mozilla/PodOperations.h"
#include "mozilla/Printf.h"
#include "mozilla/RangedPtr.h"
#include <stdarg.h>
#include <stdio.h>
#include "ds/LifoAlloc.h"
#include "js/CharacterEncoding.h"
#include "util/Memory.h"
#include "util/Text.h"
#include "util/Windows.h"
#include "vm/JSContext.h"
using mozilla::PodCopy;
namespace {
class GenericPrinterPrintfTarget : public mozilla::PrintfTarget {
public:
explicit GenericPrinterPrintfTarget(js::GenericPrinter& p) : printer(p) {}
bool append(const char* sp, size_t len) override {
return printer.put(sp, len);
}
private:
js::GenericPrinter& printer;
};
} // namespace
namespace js {
void GenericPrinter::reportOutOfMemory() {
if (hadOOM_) {
return;
}
hadOOM_ = true;
}
bool GenericPrinter::hadOutOfMemory() const { return hadOOM_; }
bool GenericPrinter::printf(const char* fmt, ...) {
va_list va;
va_start(va, fmt);
bool r = vprintf(fmt, va);
va_end(va);
return r;
}
bool GenericPrinter::vprintf(const char* fmt, va_list ap) {
// Simple shortcut to avoid allocating strings.
if (strchr(fmt, '%') == nullptr) {
return put(fmt);
}
GenericPrinterPrintfTarget printer(*this);
if (!printer.vprint(fmt, ap)) {
reportOutOfMemory();
return false;
}
return true;
}
const size_t Sprinter::DefaultSize = 64;
bool Sprinter::realloc_(size_t newSize) {
MOZ_ASSERT(newSize > (size_t)offset);
char* newBuf = (char*)js_realloc(base, newSize);
if (!newBuf) {
reportOutOfMemory();
return false;
}
base = newBuf;
size = newSize;
base[size - 1] = '\0';
return true;
}
Sprinter::Sprinter(JSContext* cx, bool shouldReportOOM)
: context(cx),
#ifdef DEBUG
initialized(false),
#endif
shouldReportOOM(shouldReportOOM),
base(nullptr),
size(0),
offset(0) {
}
Sprinter::~Sprinter() {
#ifdef DEBUG
if (initialized) {
checkInvariants();
}
#endif
js_free(base);
}
bool Sprinter::init() {
MOZ_ASSERT(!initialized);
base = js_pod_malloc<char>(DefaultSize);
if (!base) {
reportOutOfMemory();
return false;
}
#ifdef DEBUG
initialized = true;
#endif
*base = '\0';
size = DefaultSize;
base[size - 1] = '\0';
return true;
}
void Sprinter::checkInvariants() const {
MOZ_ASSERT(initialized);
MOZ_ASSERT((size_t)offset < size);
MOZ_ASSERT(base[size - 1] == '\0');
}
UniqueChars Sprinter::release() {
checkInvariants();
if (hadOOM_) {
return nullptr;
}
char* str = base;
base = nullptr;
offset = size = 0;
#ifdef DEBUG
initialized = false;
#endif
return UniqueChars(str);
}
char* Sprinter::stringAt(ptrdiff_t off) const {
MOZ_ASSERT(off >= 0 && (size_t)off < size);
return base + off;
}
char& Sprinter::operator[](size_t off) {
MOZ_ASSERT(off < size);
return *(base + off);
}
char* Sprinter::reserve(size_t len) {
InvariantChecker ic(this);
while (len + 1 > size - offset) { /* Include trailing \0 */
if (!realloc_(size * 2)) {
return nullptr;
}
}
char* sb = base + offset;
offset += len;
return sb;
}
bool Sprinter::put(const char* s, size_t len) {
InvariantChecker ic(this);
const char* oldBase = base;
const char* oldEnd = base + size;
char* bp = reserve(len);
if (!bp) {
return false;
}
/* s is within the buffer already */
if (s >= oldBase && s < oldEnd) {
/* buffer was realloc'ed */
if (base != oldBase) {
s = stringAt(s - oldBase); /* this is where it lives now */
}
memmove(bp, s, len);
} else {
js_memcpy(bp, s, len);
}
bp[len] = '\0';
return true;
}
bool Sprinter::putString(JSString* s) {
InvariantChecker ic(this);
JSLinearString* linear = s->ensureLinear(context);
if (!linear) {
return false;
}
size_t length = JS::GetDeflatedUTF8StringLength(linear);
char* buffer = reserve(length);
if (!buffer) {
return false;
}
mozilla::DebugOnly<size_t> written =
JS::DeflateStringToUTF8Buffer(linear, mozilla::Span(buffer, length));
MOZ_ASSERT(written == length);
buffer[length] = '\0';
return true;
}
ptrdiff_t Sprinter::getOffset() const { return offset; }
void Sprinter::reportOutOfMemory() {
if (hadOOM_) {
return;
}
if (context && shouldReportOOM) {
ReportOutOfMemory(context);
}
hadOOM_ = true;
}
bool Sprinter::jsprintf(const char* format, ...) {
va_list ap;
va_start(ap, format);
bool r = vprintf(format, ap);
va_end(ap);
return r;
}
const char js_EscapeMap[] = {
// clang-format off
'\b', 'b',
'\f', 'f',
'\n', 'n',
'\r', 'r',
'\t', 't',
'\v', 'v',
'"', '"',
'\'', '\'',
'\\', '\\',
'\0'
// clang-format on
};
static const char JSONEscapeMap[] = {
// clang-format off
'\b', 'b',
'\f', 'f',
'\n', 'n',
'\r', 'r',
'\t', 't',
'"', '"',
'\\', '\\',
'\0'
// clang-format on
};
template <QuoteTarget target, typename CharT>
bool QuoteString(Sprinter* sp, const mozilla::Range<const CharT> chars,
char quote) {
MOZ_ASSERT_IF(target == QuoteTarget::JSON, quote == '\0');
using CharPtr = mozilla::RangedPtr<const CharT>;
const char* escapeMap =
(target == QuoteTarget::String) ? js_EscapeMap : JSONEscapeMap;
if (quote) {
if (!sp->putChar(quote)) {
return false;
}
}
const CharPtr end = chars.end();
/* Loop control variables: end points at end of string sentinel. */
for (CharPtr t = chars.begin(); t < end; ++t) {
/* Move t forward from s past un-quote-worthy characters. */
const CharPtr s = t;
char16_t c = *t;
while (c < 127 && c != '\\') {
if (target == QuoteTarget::String) {
if (!IsAsciiPrintable(c) || c == quote || c == '\t') {
break;
}
} else {
if (c < ' ' || c == '"') {
break;
}
}
++t;
if (t == end) {
break;
}
c = *t;
}
{
ptrdiff_t len = t - s;
ptrdiff_t base = sp->getOffset();
if (!sp->reserve(len)) {
return false;
}
for (ptrdiff_t i = 0; i < len; ++i) {
(*sp)[base + i] = char(s[i]);
}
(*sp)[base + len] = '\0';
}
if (t == end) {
break;
}
/* Use escapeMap, \u, or \x only if necessary. */
const char* escape;
if (!(c >> 8) && c != 0 &&
(escape = strchr(escapeMap, int(c))) != nullptr) {
if (!sp->jsprintf("\\%c", escape[1])) {
return false;
}
} else {
/*
* Use \x only if the high byte is 0 and we're in a quoted string,
* because ECMA-262 allows only \u, not \x, in Unicode identifiers
* (see bug 621814).
*/
if (!sp->jsprintf((quote && !(c >> 8)) ? "\\x%02X" : "\\u%04X", c)) {
return false;
}
}
}
/* Sprint the closing quote and return the quoted string. */
if (quote) {
if (!sp->putChar(quote)) {
return false;
}
}
return true;
}
template bool QuoteString<QuoteTarget::String, Latin1Char>(
Sprinter* sp, const mozilla::Range<const Latin1Char> chars, char quote);
template bool QuoteString<QuoteTarget::String, char16_t>(
Sprinter* sp, const mozilla::Range<const char16_t> chars, char quote);
template bool QuoteString<QuoteTarget::JSON, Latin1Char>(
Sprinter* sp, const mozilla::Range<const Latin1Char> chars, char quote);
template bool QuoteString<QuoteTarget::JSON, char16_t>(
Sprinter* sp, const mozilla::Range<const char16_t> chars, char quote);
bool QuoteString(Sprinter* sp, JSString* str, char quote /*= '\0' */) {
JSLinearString* linear = str->ensureLinear(sp->context);
if (!linear) {
return false;
}
JS::AutoCheckCannotGC nogc;
return linear->hasLatin1Chars() ? QuoteString<QuoteTarget::String>(
sp, linear->latin1Range(nogc), quote)
: QuoteString<QuoteTarget::String>(
sp, linear->twoByteRange(nogc), quote);
}
UniqueChars QuoteString(JSContext* cx, JSString* str, char quote /* = '\0' */) {
Sprinter sprinter(cx);
if (!sprinter.init()) {
return nullptr;
}
if (!QuoteString(&sprinter, str, quote)) {
return nullptr;
}
return sprinter.release();
}
bool JSONQuoteString(Sprinter* sp, JSString* str) {
JSLinearString* linear = str->ensureLinear(sp->context);
if (!linear) {
return false;
}
JS::AutoCheckCannotGC nogc;
return linear->hasLatin1Chars() ? QuoteString<QuoteTarget::JSON>(
sp, linear->latin1Range(nogc), '\0')
: QuoteString<QuoteTarget::JSON>(
sp, linear->twoByteRange(nogc), '\0');
}
Fprinter::Fprinter(FILE* fp) : file_(nullptr), init_(false) { init(fp); }
#ifdef DEBUG
Fprinter::~Fprinter() { MOZ_ASSERT_IF(init_, !file_); }
#endif
bool Fprinter::init(const char* path) {
MOZ_ASSERT(!file_);
file_ = fopen(path, "w");
if (!file_) {
return false;
}
init_ = true;
return true;
}
void Fprinter::init(FILE* fp) {
MOZ_ASSERT(!file_);
file_ = fp;
init_ = false;
}
void Fprinter::flush() {
MOZ_ASSERT(file_);
fflush(file_);
}
void Fprinter::finish() {
MOZ_ASSERT(file_);
if (init_) {
fclose(file_);
}
file_ = nullptr;
}
bool Fprinter::put(const char* s, size_t len) {
MOZ_ASSERT(file_);
int i = fwrite(s, /*size=*/1, /*nitems=*/len, file_);
if (size_t(i) != len) {
reportOutOfMemory();
return false;
}
#ifdef XP_WIN
if ((file_ == stderr) && (IsDebuggerPresent())) {
UniqueChars buf = DuplicateString(s, len);
if (!buf) {
reportOutOfMemory();
return false;
}
OutputDebugStringA(buf.get());
}
#endif
return true;
}
LSprinter::LSprinter(LifoAlloc* lifoAlloc)
: alloc_(lifoAlloc), head_(nullptr), tail_(nullptr), unused_(0) {}
LSprinter::~LSprinter() {
// This LSprinter might be allocated as part of the same LifoAlloc, so we
// should not expect the destructor to be called.
}
void LSprinter::exportInto(GenericPrinter& out) const {
if (!head_) {
return;
}
for (Chunk* it = head_; it != tail_; it = it->next) {
out.put(it->chars(), it->length);
}
out.put(tail_->chars(), tail_->length - unused_);
}
void LSprinter::clear() {
head_ = nullptr;
tail_ = nullptr;
unused_ = 0;
hadOOM_ = false;
}
bool LSprinter::put(const char* s, size_t len) {
// Compute how much data will fit in the current chunk.
size_t existingSpaceWrite = 0;
size_t overflow = len;
if (unused_ > 0 && tail_) {
existingSpaceWrite = std::min(unused_, len);
overflow = len - existingSpaceWrite;
}
// If necessary, allocate a new chunk for overflow data.
size_t allocLength = 0;
Chunk* last = nullptr;
if (overflow > 0) {
allocLength =
AlignBytes(sizeof(Chunk) + overflow, js::detail::LIFO_ALLOC_ALIGN);
LifoAlloc::AutoFallibleScope fallibleAllocator(alloc_);
last = reinterpret_cast<Chunk*>(alloc_->alloc(allocLength));
if (!last) {
reportOutOfMemory();
return false;
}
}
// All fallible operations complete: now fill up existing space, then
// overflow space in any new chunk.
MOZ_ASSERT(existingSpaceWrite + overflow == len);
if (existingSpaceWrite > 0) {
PodCopy(tail_->end() - unused_, s, existingSpaceWrite);
unused_ -= existingSpaceWrite;
s += existingSpaceWrite;
}
if (overflow > 0) {
if (tail_ && reinterpret_cast<char*>(last) == tail_->end()) {
// tail_ and last are consecutive in memory. LifoAlloc has no
// metadata and is just a bump allocator, so we can cheat by
// appending the newly-allocated space to tail_.
unused_ = allocLength;
tail_->length += allocLength;
} else {
// Remove the size of the header from the allocated length.
size_t availableSpace = allocLength - sizeof(Chunk);
last->next = nullptr;
last->length = availableSpace;
unused_ = availableSpace;
if (!head_) {
head_ = last;
} else {
tail_->next = last;
}
tail_ = last;
}
PodCopy(tail_->end() - unused_, s, overflow);
MOZ_ASSERT(unused_ >= overflow);
unused_ -= overflow;
}
MOZ_ASSERT(len <= INT_MAX);
return true;
}
} // namespace js