STATIC mp_uint_t stringio_write(mp_obj_t o_in, const void *buf, mp_uint_t size, int *errcode) {
(void)errcode;
mp_obj_stringio_t *o = MP_OBJ_TO_PTR(o_in);
check_stringio_is_open(o);
if (o->vstr->fixed_buf) {
stringio_copy_on_write(o);
}
mp_uint_t new_pos = o->pos + size;
if (new_pos < size) {
// Writing <size> bytes will overflow o->pos beyond limit of mp_uint_t.
*errcode = MP_EFBIG;
return MP_STREAM_ERROR;
}
mp_uint_t org_len = o->vstr->len;
if (new_pos > o->vstr->alloc) {
// Take all what's already allocated...
o->vstr->len = o->vstr->alloc;
// ... and add more
vstr_add_len(o->vstr, new_pos - o->vstr->alloc);
}
// If there was a seek past EOF, clear the hole
if (o->pos > org_len) {
memset(o->vstr->buf + org_len, 0, o->pos - org_len);
}
memcpy(o->vstr->buf + o->pos, buf, size);
o->pos = new_pos;
if (new_pos > o->vstr->len) {
o->vstr->len = new_pos;
}
return size;
}
// Unbuffered, inefficient implementation of readline() for raw I/O files.
STATIC mp_obj_t stream_unbuffered_readline(uint n_args, const mp_obj_t *args) {
struct _mp_obj_base_t *o = (struct _mp_obj_base_t *)args[0];
if (o->type->stream_p == NULL || o->type->stream_p->read == NULL) {
// CPython: io.UnsupportedOperation, OSError subclass
nlr_raise(mp_obj_new_exception_msg(&mp_type_OSError, "Operation not supported"));
}
mp_int_t max_size = -1;
if (n_args > 1) {
max_size = MP_OBJ_SMALL_INT_VALUE(args[1]);
}
vstr_t *vstr;
if (max_size != -1) {
vstr = vstr_new_size(max_size);
} else {
vstr = vstr_new();
}
int error;
while (max_size == -1 || max_size-- != 0) {
char *p = vstr_add_len(vstr, 1);
if (p == NULL) {
nlr_raise(mp_obj_new_exception_msg_varg(&mp_type_MemoryError, "out of memory"));
}
mp_int_t out_sz = o->type->stream_p->read(o, p, 1, &error);
if (out_sz == -1) {
nlr_raise(mp_obj_new_exception_msg_varg(&mp_type_OSError, "[Errno %d]", error));
}
if (out_sz == 0) {
// Back out previously added byte
// TODO: This is a bit hacky, does it supported by vstr API contract?
// Consider, what's better - read a char and get OutOfMemory (so read
// char is lost), or allocate first as we do.
vstr_add_len(vstr, -1);
break;
}
if (*p == '\n') {
break;
}
}
// TODO need a string creation API that doesn't copy the given data
mp_obj_t ret = mp_obj_new_str_of_type(STREAM_CONTENT_TYPE(o->type->stream_p), (byte*)vstr->buf, vstr->len);
vstr_free(vstr);
return ret;
}
// Unbuffered, inefficient implementation of readline() for raw I/O files.
STATIC mp_obj_t stream_unbuffered_readline(size_t n_args, const mp_obj_t *args) {
const mp_stream_p_t *stream_p = mp_get_stream_raise(args[0], MP_STREAM_OP_READ);
mp_int_t max_size = -1;
if (n_args > 1) {
max_size = MP_OBJ_SMALL_INT_VALUE(args[1]);
}
vstr_t vstr;
if (max_size != -1) {
vstr_init(&vstr, max_size);
} else {
vstr_init(&vstr, 16);
}
while (max_size == -1 || max_size-- != 0) {
char *p = vstr_add_len(&vstr, 1);
if (p == NULL) {
nlr_raise(mp_obj_new_exception_msg_varg(&mp_type_MemoryError, "out of memory"));
}
int error;
mp_uint_t out_sz = stream_p->read(args[0], p, 1, &error);
if (out_sz == MP_STREAM_ERROR) {
if (mp_is_nonblocking_error(error)) {
if (vstr.len == 1) {
// We just incremented it, but otherwise we read nothing
// and immediately got EAGAIN. This case is not well
// specified in
// https://docs.python.org/3/library/io.html#io.IOBase.readline
// unlike similar case for read(). But we follow the latter's
// behavior - return None.
vstr_clear(&vstr);
return mp_const_none;
} else {
goto done;
}
}
mp_raise_OSError(error);
}
if (out_sz == 0) {
done:
// Back out previously added byte
// Consider, what's better - read a char and get OutOfMemory (so read
// char is lost), or allocate first as we do.
vstr_cut_tail_bytes(&vstr, 1);
break;
}
if (*p == '\n') {
break;
}
}
return mp_obj_new_str_from_vstr(STREAM_CONTENT_TYPE(stream_p), &vstr);
}
STATIC mp_uint_t stringio_write(mp_obj_t o_in, const void *buf, mp_uint_t size, int *errcode) {
mp_obj_stringio_t *o = o_in;
mp_uint_t remaining = o->vstr->alloc - o->pos;
if (size > remaining) {
// Take all what's already allocated...
o->vstr->len = o->vstr->alloc;
// ... and add more
vstr_add_len(o->vstr, size - remaining);
}
memcpy(o->vstr->buf + o->pos, buf, size);
o->pos += size;
if (o->pos > o->vstr->len) {
o->vstr->len = o->pos;
}
return size;
}
STATIC mp_obj_t stream_read_generic(size_t n_args, const mp_obj_t *args, byte flags) {
// What to do if sz < -1? Python docs don't specify this case.
// CPython does a readall, but here we silently let negatives through,
// and they will cause a MemoryError.
mp_int_t sz;
if (n_args == 1 || ((sz = mp_obj_get_int(args[1])) == -1)) {
return stream_readall(args[0]);
}
const mp_stream_p_t *stream_p = mp_get_stream(args[0]);
#if MICROPY_PY_BUILTINS_STR_UNICODE
if (stream_p->is_text) {
// We need to read sz number of unicode characters. Because we don't have any
// buffering, and because the stream API can only read bytes, we must read here
// in units of bytes and must never over read. If we want sz chars, then reading
// sz bytes will never over-read, so we follow this approach, in a loop to keep
// reading until we have exactly enough chars. This will be 1 read for text
// with ASCII-only chars, and about 2 reads for text with a couple of non-ASCII
// chars. For text with lots of non-ASCII chars, it'll be pretty inefficient
// in time and memory.
vstr_t vstr;
vstr_init(&vstr, sz);
mp_uint_t more_bytes = sz;
mp_uint_t last_buf_offset = 0;
while (more_bytes > 0) {
char *p = vstr_add_len(&vstr, more_bytes);
int error;
mp_uint_t out_sz = mp_stream_read_exactly(args[0], p, more_bytes, &error);
if (error != 0) {
vstr_cut_tail_bytes(&vstr, more_bytes);
if (mp_is_nonblocking_error(error)) {
// With non-blocking streams, we read as much as we can.
// If we read nothing, return None, just like read().
// Otherwise, return data read so far.
// TODO what if we have read only half a non-ASCII char?
if (vstr.len == 0) {
vstr_clear(&vstr);
return mp_const_none;
}
break;
}
mp_raise_OSError(error);
}
if (out_sz < more_bytes) {
// Finish reading.
// TODO what if we have read only half a non-ASCII char?
vstr_cut_tail_bytes(&vstr, more_bytes - out_sz);
if (out_sz == 0) {
break;
}
}
// count chars from bytes just read
for (mp_uint_t off = last_buf_offset;;) {
byte b = vstr.buf[off];
int n;
if (!UTF8_IS_NONASCII(b)) {
// 1-byte ASCII char
n = 1;
} else if ((b & 0xe0) == 0xc0) {
// 2-byte char
n = 2;
} else if ((b & 0xf0) == 0xe0) {
// 3-byte char
n = 3;
} else if ((b & 0xf8) == 0xf0) {
// 4-byte char
n = 4;
} else {
// TODO
n = 5;
}
if (off + n <= vstr.len) {
// got a whole char in n bytes
off += n;
sz -= 1;
last_buf_offset = off;
if (off >= vstr.len) {
more_bytes = sz;
break;
}
} else {
// didn't get a whole char, so work out how many extra bytes are needed for
// this partial char, plus bytes for additional chars that we want
more_bytes = (off + n - vstr.len) + (sz - 1);
break;
}
}
}
return mp_obj_new_str_from_vstr(&mp_type_str, &vstr);
}
#endif
vstr_t vstr;
vstr_init_len(&vstr, sz);
int error;
mp_uint_t out_sz = mp_stream_rw(args[0], vstr.buf, sz, &error, flags);
if (error != 0) {
vstr_clear(&vstr);
if (mp_is_nonblocking_error(error)) {
// https://docs.python.org/3.4/library/io.html#io.RawIOBase.read
// "If the object is in non-blocking mode and no bytes are available,
// None is returned."
// This is actually very weird, as naive truth check will treat
// this as EOF.
return mp_const_none;
}
mp_raise_OSError(error);
} else {
vstr.len = out_sz;
return mp_obj_new_str_from_vstr(STREAM_CONTENT_TYPE(stream_p), &vstr);
}
}
// Unbuffered, inefficient implementation of readline() for raw I/O files.
STATIC mp_obj_t stream_unbuffered_readline(uint n_args, const mp_obj_t *args) {
struct _mp_obj_base_t *o = (struct _mp_obj_base_t *)args[0];
if (o->type->stream_p == NULL || o->type->stream_p->read == NULL) {
// CPython: io.UnsupportedOperation, OSError subclass
nlr_raise(mp_obj_new_exception_msg(&mp_type_OSError, "Operation not supported"));
}
mp_int_t max_size = -1;
if (n_args > 1) {
max_size = MP_OBJ_SMALL_INT_VALUE(args[1]);
}
vstr_t *vstr;
if (max_size != -1) {
vstr = vstr_new_size(max_size);
} else {
vstr = vstr_new();
}
int error;
while (max_size == -1 || max_size-- != 0) {
char *p = vstr_add_len(vstr, 1);
if (p == NULL) {
nlr_raise(mp_obj_new_exception_msg_varg(&mp_type_MemoryError, "out of memory"));
}
mp_uint_t out_sz = o->type->stream_p->read(o, p, 1, &error);
if (out_sz == MP_STREAM_ERROR) {
if (is_nonblocking_error(error)) {
if (vstr->len == 1) {
// We just incremented it, but otherwise we read nothing
// and immediately got EAGAIN. This is case is not well
// specified in
// https://docs.python.org/3/library/io.html#io.IOBase.readline
// unlike similar case for read(). But we follow the latter's
// behavior - return None.
vstr_free(vstr);
return mp_const_none;
} else {
goto done;
}
}
nlr_raise(mp_obj_new_exception_arg1(&mp_type_OSError, MP_OBJ_NEW_SMALL_INT(error)));
}
if (out_sz == 0) {
done:
// Back out previously added byte
// Consider, what's better - read a char and get OutOfMemory (so read
// char is lost), or allocate first as we do.
vstr_cut_tail_bytes(vstr, 1);
break;
}
if (*p == '\n') {
break;
}
}
// TODO need a string creation API that doesn't copy the given data
mp_obj_t ret = mp_obj_new_str_of_type(STREAM_CONTENT_TYPE(o->type->stream_p), (byte*)vstr->buf, vstr->len);
vstr_free(vstr);
return ret;
}
